CS Paper Radar

Daily listing

2026-10-28

487 papers first submitted this day (UTC). arXiv announces a day's submissions the next evening, so the newest day fills up over ~24 h.

Score = weighted mean ofCitations 20%Upvotes 10%Code stars 10%Venue 15%Code/data 10%Novelty 15%Momentum 15%Substance 5%how it works

AI & Machine Learning 149

FairSplit: Decomposing the Embedding Space for Fair Classification

cs.LG ·2026-10-28 ·Andrii Shkabrii, Anna Beer, Ylli Sadikaj, Claudia Plant
top venuecode/datahot direction
Abstract · Accepted at NeurIPS 2026

Fair classification faces a twofold challenge: maintaining high classification accuracy while limiting reliance on sensitive information. Our method FairSplit achieves this by learning a decomposition of the embedding space into a Fair Space used for prediction and a Sensitive Space that isolates sensitive information. During training, the model must simultaneously learn the classification task and separate fair from sensitive information. The Sensitive Space gives sensitive attributes a destination, so the Fair Space can be used for prediction without relying on them. This decomposition is guided via a learnable mask and the Hilbert-Schmidt Independence Criterion (HSIC) as a measure of statistical dependence. We provide theoretical guarantees showing that the demographic parity gap is upper-bounded by both (i) the learned dimensionality of the Fair Space and (ii) the HSIC between the Fair Space and sensitive attributes. FairSplit effectively handles multiclass tasks and non-binary sensitive attributes, overcoming key limitations of existing fair classification methods. Experiments on established fairness benchmarks show that FairSplit achieves competitive accuracy-fairness trade-offs. Code is available at https://github.com/shkabrii/fairsplit.

0.74

Assessing feature scorer results on high-dimensional datasets with t-SNE

cs.LG cs.AI ·2026-10-28 ·Bruno Iochins Grisci, Mario Inostroza-Ponta, Márcio Dorn
publishedcode/dataunusualhighly cited
Abstract

While vast literature on high-dimensional data visualization is available, there are not many works regarding the visualization of feature scorers and their results. Feature scorers are algorithms that assign numerical importance to each feature of multi-dimensional datasets. These importance scores can be used in several applications, such as feature selection, knowledge discovery, and machine learning interpretability. There are several feature scorers to choose from, and often no single metric or ground truth is available to guarantee the quality of their results. In this scenario, visualization can become valuable to support the decision of which method to choose and how good its results are. For this goal, this work presents "weighted t-SNE." It modifies the relationship between data points in the embedded 2D space to reflect the importance of each dimension of the original datasets as assessed by a feature scorer. This research discusses how to implement weighted t-SNE, proposes the silhouette coefficient as a numerical evaluation of the results, and shows several examples of its use in practice. Synthetic and real-world tabular datasets are used in the experiments together with nine feature scorers, ranging from Mutual Information to neural networks. Each feature scorer produces unique visualizations, and weighted t-SNE can be used to compare and choose the one that better suits a given dataset and task. Weighted t-SNE can also visually show the importance of features learned by machine learning models and help us see how they are organizing the data, increasing their interpretability.

0.71

When Does Closed-Loop Curriculum Adaptation Help GRPO? An Analysis on Countdown

cs.LG cs.AI ·2026-10-28 ·Yuv Raj Pant, Moshiur Rahman Tonmoy, Kaiqun Fu
publishedcode/data
Abstract · 6 pages, 2 figures, Accepted at IEEE ICTAI 2026

Reinforcement fine-tuning has become an effective post-training paradigm for improving the reasoning ability of large language models, but it remains sample-inefficient when the training distribution is mismatched to the model's current capability. In group-relative objectives such as Group Relative Policy Optimization (GRPO), problems that are too hard often produce uniformly failed rollouts, while problems that are too easy provide little useful advantage signal. In this work, we introduce CLOSEDLOOPGAUSSIAN, a lightweight adaptive curriculum method for efficient reinforcement fine-tuning on ordered reasoning tasks. Instead of following a fixed easy-to-hard schedule, CLOSEDLOOPGAUSSIAN preserves the structure of an Easy-to-Hard Gaussian sampler while dynamically adjusting its pace and spread using periodic probe estimates of per-difficulty accuracy. The controller accelerates training when the model is ahead of a reference curriculum, slows down when the model falls behind, and broadens sampling when earlier difficulties regress. This feedback mechanism keeps training focused on tasks that are challenging but still learnable, without changing the reward function, model architecture, or underlying GRPO optimizer. Experiments on symbolic arithmetic reasoning task show that CLOSEDLOOPGAUSSIAN improves exact-match accuracy over vanilla GRPO and fixed open-loop curricula, with the strongest gains in intermediate-difficulty regimes where group-relative learning signals are most informative. Codes are available at https://github.com/yuvrajpant56/Closed_Loop_Curriculum.

0.53

B2P-Corr: Batch-to-Population Gradient Estimators for Non-Decomposable Correlation Losses

cs.LG ·2026-10-28 ·Jingquan Yan, Yuwei Miao, Peiran Yu, Junzhou Huang
top venuecode/dataunusual
Abstract · Accepted by NeurIPS 2026. Implementation is available at https://github.com/jyan97/b2p-corr

Correlation metrics such as the Pearson Correlation Coefficient (PCC) and Concordance Correlation Coefficient (CCC) are standard regression metrics and learning objectives, but their non-decomposable nature poses a challenge for stochastic optimization. Since these metrics depend on dataset-level moments (mean, variance, covariance), naive mini-batch training with local statistics yields biased, high-variance gradients whose expectation does not align with the population gradient. To bridge this ``batch-to-population'' (B2P) gap, we propose B2P-Corr, a novel framework that transforms correlation objectives into decomposable pseudo-losses driven by lightweight global ``moment sketches''. We introduce two efficient variants: B2P-EMA, which utilizes exponential moving averages and stop-gradient operators to ensure asymptotic gradient tracking with $O(B)$ cost ($B$ is batch size); and B2P-U, which uses sampled cross-batch U-statistic estimators and a reservoir-based stabilization with an $O(B)$ implementation. Theoretically, we quantify the $O(B^{-1})$ bias of naive mini-batch optimization, prove a moving-target tracking result for B2P-EMA under a fast-sketch / slow-parameter regime, and provide an idealized control-variate analysis for B2P-U. Experiments on two synthetic and three real-world datasets demonstrate B2P-Corr achieves stable optimization and higher correlation than naive mini-batch optimization with negligible overhead.

0.53

NAGO: Noise-Aware Generative Operator via Flow Matching

cs.LG ·2026-10-28 ·Guanyu Chen, Shengze Xu, Pengwei Liu, Xingyu Ren, Pengkai Wang +3 more
top venuecode/data
Abstract · Accepted at NeurIPS 2026. 53 pages including appendices and checklist. Updated to the NeurIPS camera-ready format; research content unchange…

We introduce NAGO, a noise-aware generative operator for efficient sampling from invariant measures of stochastic partial differential equations (SPDEs) in function space. Our central idea is to treat the source measure as a design variable rather than a fixed Gaussian default. We establish admissibility conditions for well-defined infinite-dimensional flows and show that the finite-band source--target covariance ratio controls the spectral stiffness of the marginal velocity field. This analysis yields an admissibility-constrained spectral-matching rule: choose a source that is sufficiently rough for well-posedness, then minimize covariance mismatch over the resolved Fourier modes. On three representative stochastic-field benchmarks, this source design improves few-step spectral accuracy and several complementary distributional and physical statistics. In a controlled Matern-field ablation, it reduces the accurate RK4 budget from about 30 steps for Gaussian functional flow matching to 3--5 steps, a $6$--$10\times$ reduction in neural ODE evaluations. These gains are strongest for stationary multiscale fields and depend on the metric, resolution, and guidance protocol.

0.48

ParaTTT: Parallelizing Test-Time Training Layers with Deep Memories

cs.LG ·2026-10-28 ·Noel Loo, Neehal Tumma, T. Konstantin Rusch, Mathias Lechner, Alexander Amini +1 more
theorycode/data
Abstract · 34 pages, 7 figures. Project repository: https://github.com/ntumm120/paraTTT

Test-time training (TTT) layers replace the linear fast-weight memory of modern recurrent architectures with a small neural network trained online by gradient descent. This nonlinear memory is more expressive than the linear memories of Gated DeltaNet and Mamba-2, but is difficult to parallelize: existing implementations such as TTT-MLP rely on sequential, block-wise weight updates that run several times slower than chunkwise-parallel linear recurrences. Moreover, their block-wise execution only approximates the token-by-token online recurrence, whereas linear recurrences admit an exact chunkwise-parallel evaluation. We introduce ParaTTT (PTTT), an algorithm for parallelizing arbitrarily deep nonlinear memories by iteratively refining estimates of the inner-loop gradient trajectory. Starting from a guess of the hidden-layer gradients, each refinement consists of three chunkwise-parallel passes: a forward pass that evaluates the resulting feature trajectory, an exact gated-delta-rule pass that computes the readout trajectory and its feature gradients, and a backward pass that propagates these gradients into a new estimate. The exact online gradient descent recurrence is a fixed point of the iteration, while decoding requires no refinement iterations. We prove geometric convergence for a two-layer variant under sufficient contraction conditions. In practice, two iterations suffice. With two refinement steps at 4K context, PTTT evaluates the inner-loop trajectory $32\times$ faster than TTT-MLP with the block approximation at matched error, while the full 340M model trains $4.9\times$ faster. Models trained at 340M and 1.3B parameters match or exceed Gated DeltaNet and Mamba-2 and remain competitive with same-size Transformers on language-modeling perplexity and zero-shot accuracy, with better recall than the linear recurrences.

0.47

DecompGym: Benchmarking LLM Agents on Designing Decomposition Algorithms

cs.AI math.OC ·2026-10-28 ·Sikai Cheng, Tinghan Ye, Arnaud Deza, Xiaonan Qi, Miao Li +2 more
benchmarkcode/data
Abstract

Benders decomposition and column generation solve large optimization problems by passing cuts, columns, and dual values between a master problem and subproblems, and are hard to implement correctly. Large language models now automate parts of algorithm design, which points toward agents that build decomposition solvers for new problems. As a first step toward such agents, we study how well current coding agents design and implement these two algorithms, and how they fail. We introduce DecompGym, 81 instances from nine problem families (five Benders, four column generation), in which an agent writes a solver for a prescribed method. Each solver must meet three criteria: execution within a 60-second time limit; numerical feasibility together with a method audit, in which an LLM judge reads the solver's code; and an objective within 1% of the stored optimal objective. We evaluate 12 GPT, Claude, and Gemini models running in coding-agent tools. Over up to ten attempts, we compare independent regeneration with reflection, which first diagnoses the failure. On the first attempt, two models solve every instance and seven solve more than 90%. Most Benders failures (48 of 62) are missing or suboptimal solutions. The most common column-generation failure is a method-audit rejection. Output checks alone would accept 93 of the 108 attempts that fail the method audit. The most common rejection, an optimalif, occurs 96 times, 92 of them in one model.Reflection turns 66% of retries into successes, compared with 39% for regeneration, and it repairs method-audit failures, for which the judge names the defecective gaps. Agents for decomposition therefore need checks on how a solver reaches its answer and feedback that locates the error. The benchmark is available at https://github.com/SIKAI-C/DecompGym.

0.45

BASILISC: Bayesian Alignment of Siamese-Integrated Latents for Inferring Single-cell Correspondence

cs.LG ·2026-10-28 ·Sepideh Saran, Mahsa Ghanbari, Uwe Ohler
benchmark
Abstract · 12 pages main text, 3 figures, 1 table; supplementary material with 10 figures and 10 tables (38 pages total). Code to be released on public…

Single-cell assays study the genome one cell at a time. Unimodal assays measure one molecular layer, the newer multimodal assays measure several. Most data are single-modality, so integrating them relies on models trained on multimodal assays, where the cell correspondence across modalities is known. These models return a matching, but rarely a confidence measure. We present BASILISC (Bayesian Alignment of Siamese-Integrated Latents for Inferring Single-cell Correspondence), a framework for matching single-cell representations with uncertainty on the correspondence itself. A BASILISC model is a probabilistic siamese network over the two modalities with a predictive distribution over the correspondence for a cross-modal pair of cells. We instantiate it at two levels: at the cell-type level the question is whether two cells share a cell type, at the cell level which cell is the partner in the other modality. To our knowledge, this is the first uncertainty score on a learned cross-modal correspondence evaluated against a known ground-truth pairing. BASILISC learns a similarity across modalities that reflects cell lineage: an unseen cell type is matched to its lineage relative, matching still works on such held-out types (balanced accuracy 0.86 on unseen against 0.93 on seen types), and the predictive uncertainty rises on these out-of-distribution cells. The uncertainty score also selects high-quality matches for downstream analysis: discarding the least certain fifth of the pairs raises accuracy from 0.93 to 0.98. At the cell level, a deep ensemble of the deterministic matcher reaches 0.062 matching accuracy on the benchmark test set, above the competition winner; on the Bayesian matcher's assignment, the posterior flags the wrong matches that cross a cell-type boundary better than a deterministic read of the same scores (AUROC 0.68 against 0.64).

0.42

Prototype-Guided Dynamic Domain Modulation and a Large-Scale Multi-SKU Benchmark for Shampoo Bottle Surface Defect Detection

cs.AI ·2026-10-28 ·Yuehang Li, Xiao Wang, Qiang Chen, Yangyi Fan, Bo Jiang +2 more
benchmarkcode/data
Abstract

Industrial shampoo-bottle surface defect detection must operate across many stock keeping units (SKUs), where differences in bottle geometry, material, label layout, reflectance, and imaging conditions can substantially alter the appearance of the same defect. Existing industrial defect datasets rarely provide both instance-level defect annotations and product-variant identities, making it difficult to study SKU-aware detection under realistic manufacturing conditions. We introduce Bottle-SKU, a large-scale multi-SKU benchmark for shampoo bottle surface defect detection. Bottle-SKU contains 13,119 images, 20,419 instance-level bounding boxes, seven defect categories, and 48 bottle SKUs, enabling evaluation of both shared defect recognition and SKU-specific appearance variation. To address the challenge of SKU-specific appearance variation, we propose Prototype-Guided Dynamic Domain Modulation (ProDMM). ProDMM constructs a visual prototype for each SKU from training images and predicts a soft SKU distribution for each input. The resulting prototype-conditioned representation dynamically generates low-rank Query/Value modulation for the visual backbone and residual conditioning for defect-text-to-visual attention, while preserving the pretrained Grounding DINO detection pathway. The model is trained jointly with detection, SKU-domain classification, and modulation-regularization objectives. Experiments on Bottle-SKU and the public GC10-DET-SKU metallic surface defect dataset demonstrate the effectiveness of prototype-guided domain conditioning for industrial surface defect detection. The source code and model will be released on https://github.com/Event-AHU/Industrial_Anomaly_Detection

0.42

Disparity Laundering: Consensus Without Fairness in Clinical Multi-Agent Systems

cs.AI cs.MA ·2026-10-28 ·Xiaoyang Wang, Tianrui Wang, Christopher C. Yang
publishedcode/data
Abstract · 21 pages, 8 figures. Code: https://github.com/wangxiaoyang0412/disparity-laundering . Committed to EACL 2027

Clinical large language models are increasingly fielded as multi-agent teams, on the assumption that deliberation averages out individual error. It does not average out bias. Across a full factorial of five model families and four aggregation topologies on 120 real MIMIC-IV emergency cases under counterfactual demographic swaps, no configuration neutralizes the disparity: all 20 cells commit the direction of their agents' independent answers, giving Black patients more acute triage. Every model has a topology that amplifies its baseline gap, but which one varies by model, so no design is reliably safe or reliably amplifying: the disparity is relocated, never removed. Late-stage fixes fail: fairness instructions and demographic masking leave the gap untouched, and per-agent self-debiasing widens it, because the bias is already encoded in the agents' ordinary clinical numbers. We call this disparity laundering; the audits now applied to clinical AI, per model, per output, per version, can certify the parts but never the committed decision. Interventions close the gap exactly when they make counterfactual versions visible inside the system and pool them, through deliberation, into the committed decision. A diverse-perspective ensemble does both while keeping the team, with no retraining or architectural change: the disparity falls by 67 to 94% where all five perspectives pool at every round, and the one topology that never pools them is the one it cannot reliably repair. Medical multi-agent systems do not filter demographic bias; they launder it into plausible decisions no single-version audit can catch, and the only defense is to make the bias visible.

0.41

Tabular foundation models are not a shortcut to calibrated clinical risk prediction: a pre-registered simulation against the true risk

cs.LG stat.ME ·2026-10-28 ·Ebrahim Khaled Ebrahim
code/data
Abstract · 37 pages, 7 figures, 1 table; Supplementary Information (55 pages) as an ancillary file. Pre-registered on OSF Registries (doi:10.17605/OSF.…

Tabular foundation models are pretrained networks that predict risk from a table without fitting a model. Whether their risks are correct (calibrated) has been judged on real data, where true risk is unknown. In a pre-registered simulation with known true risks, we compared TabPFN-3.5, TabICLv2 and Mitra-v2 with Firth logistic regression with intercept correction (FLIC) in 58 scenarios and 1,000 random designs, then on four clinical data sets. Foundation models were not a shortcut to calibration. Their calibration slope matched FLIC's within the pre-registered margin in 36 or 37 of 48 core scenarios and was better only at n = 200 with a 3% event rate. Even 3,000 patients were often not enough: with FLIC, TabPFN-3.5 or Mitra-v2, 63%-83% of samples gave risks close to the truth in average, intercept and slope at event rates of 10% and 30%, and 35%-50% at 3%. TabICLv2 did worse (8%-39% and 1%-3%). Across the random designs, foundation models had a lower true calibration error than FLIC beyond the margin in 25%-38% and a higher one in none, though small losses were predicted at n = 3000. On real data, no model resisted a change in event rate between populations; recalibration on 500 local patients largely repaired it.

0.41

A Case Study of a Multi-Agent Kernel-Optimization Loop That Learns a New AI Accelerator

cs.LG cs.AIcs.ARcs.PL ·2026-10-28 ·Byonggon Chun
publishedbenchmarkcode/data
Abstract · 7 pages, 3 figures, 2 tables. Round 1 of the NPU Model Optimization Competition at MICRO 2026. Code: https://github.com/bgchun-fs/multi-agen…

Frontier large language models write correct CUDA kernels for most single-operator benchmark tasks, yet their accuracy collapses on accelerators whose architecture and programming language they have rarely seen. We present a case study of an LLM agent loop built for such accelerators. Our example is the FuriosaAI RNGD NPU, programmed in a recent, sparsely documented Rust DSL, where the loop optimized three Gemma-4-12B decode kernels over 12 days on a shared, scarce device. The loop builds its own target knowledge: it writes and checks a summary of the DSL documentation, measures the chip with microbenchmarks, records compiler-derived legality rules, and revisits this knowledge when the toolchain changes, rechecking rules and measurements where it can. A role-based search over per-kernel Design Trees explores whole-kernel designs and accumulates small accepted changes in the best version of each kernel. Because device time is scarce and readings are noisy, cheap checks screen new designs before any device run, and late in the run a change was accepted only after repeated comparisons with the best kernel on the device. The final kernels reach a 9.293x geomean speedup over the official baseline, each within 1.25x of the cycles needed just to move its weights. Our prompts, profiler and acceptance test are available at https://github.com/bgchun-fs/multi-agent-kernel-gen-loop.

0.40

From Leniency to Suspicion: How an Open Post-Training Recipe Changes Whom LLM Fact-Checkers Believe

cs.AI cs.CLcs.CY ·2026-10-28 ·Nouar AlDahoul, Yasir Zaki
Abstract · 10 pages, 4 figures, 6 tables. Preprint, under review

LLMs increasingly fact-check online content with the poster's identity in view. However, it remains unclear which post-training changes how much a model's truth judgement depends on the poster's social class, and how such a change happens inside the model. This paper investigates job-based doubt, i.e., judging a claim false more often when the poster has a lower-status job, using fact-checked claims with randomised poster profiles in base and post-trained checkpoints of six model families. Our results reveal that: 1) post-training creates such doubt only with the open Tülu-3 recipe (OLMo-2 32B, Tülu-3 70B), not with Meta's, Alibaba's or Mistral's; the effect is under one percentage point but statistically reliable, exceeding training drift and replicating on independent claim sets; 2) in OLMo-2 32B, the model does not start noticing the poster's job more, but starts interpreting it differently: a lower-status job pushes the base model toward "true" and the post-trained model toward "false"; and 3) this flip is carried by a single residual-stream direction and happens at supervised fine-tuning. The mechanism was confirmed on 400 new claims with pre-specified predictions. None of seven deployed commercial models doubts lower-status posters more. Our work underscores the need to audit speaker bias per training recipe.

0.37

Beyond fabrication: Misattributed numbers in papers written by AI research agents

cs.AI cs.CLcs.DL ·2026-10-28 ·Mosab Alfaqeeh, Omar Mohammed Horani, Mou'th Hourani, Hashem Alfaqeeh
benchmarkcode/data
Abstract · v2: corrected author names; clarified the second-judge comparison and two table captions. 43 pages, 5 figures, 13 tables. Code and data: htt…

Verifying numerical claims in research written by AI agents is an important but challenging problem. Recent benchmarks show that many agent-written papers fail integrity checks, but such verdicts, often produced by a single LLM judge, identify failure without explaining its nature, and existing checks either test a paper for internal consistency or test whether a reported value appears somewhere in an agent's output. We re-analyze all 540 released judgments from a recent public benchmark of AI research agents (nine agents, fifteen open research problems) and manually label 256 numerical claims from 18 submissions. Failures concentrate on alignment between the paper and the underlying work rather than overt cheating, and about four in five unsupported claims correspond to values in the agent's own files that come from the wrong experiment, metric, or run. We introduce a deterministic claim-to-evidence checker that scores whether the surrounding context identifies the claimed result and calibrates each claim against numerical decoys. With its rules fixed before labeling, it certifies claims with over 90% precision against blinded annotators (78 of 85; Cohen's kappa = 0.89), while verifying only a subset of claims; most claims it cannot verify are nevertheless traceable, so failure to verify is not evidence of falsehood. An LLM verifier given the same evidence recovers more traceable claims but accepts about three times as many misattributed ones, so the two are best used together. Applied to 99 submissions, the checker reveals large differences in traceability across agents, and on 30 submissions a second judge configuration, with a different model and harness, orders the agents' vendor groups as the official judge does while passing twice as many papers. These findings motivate provenance-first auditing of AI-produced research.

0.37

Right Numbers, Wrong Questions: Evaluating LLMs on Underspecified Financial Queries

cs.AI cs.CLq-fin.GN ·2026-10-28 ·Radha Yadav, Prasanna Devadiga, Peeyush Mishra, Apoorv Singh
Abstract · Accepted for oral presentation at the 3rd Workshop on LLMs and Generative AI for Finance (AI4F) at ACM ICAIF '26, Milan, Italy, November 14,…

Financial assistants answer questions that sound factual but depend on what the user leaves unsaid: what a data figure measures, which product variant is meant, and the investor's situation and the rules that apply. An assistant that ignores these can still answer with a real number from its data, one that answers a different question. We call this frame collapse. We study it in Indian mutual funds, where product variants and tax rules are formally defined, so the correct answer can be computed for every reading. One study tests each challenge. Given an ambiguously labelled figure, models used it, often under the user's label. Given every alternative labelled, they listed them all, but the layout decided which came first. Missing a fact about the investor, weaker agents assumed a default or an outdated tax rule; requiring the fact did not guarantee its use. Because such answers match their data, checks against the data pass them. An AI judge caught mislabelled figures only when told what each field means, and still passed answers that silently chose another reading. We show how these failures arise, not how often, and propose controls for financial assistants.

0.37

ExplainCC: Explaining Correlated Clusters by Node Labels

cs.LG cs.DS ·2026-10-28 ·Nadeesha Shyami Kumari Herath Mudiyanselage, Florian Adriaens, Nikolaj Tatti
top venue
Abstract · Accepted to CIKM 2026

An increasingly important requirement for clustering algorithms is that decisions and outcomes need to be clearly understandable for human users. We address this requirement for the Correlation Clustering (CC) problem, in case the objects to be clustered are accompanied by labels. In CC one seeks to partition a set of objects into clusters based on pairwise similarity and dissimilarity information. Our proposed model only considers clusters that can be explained by a label tree, which is essentially a decision tree that partitions the objects according to labels. The goal is to find an explainable partition of the data that minimizes the CC objective function. We formalize this problem from a theoretical viewpoint, and prove several hardness of approximation results, also when constraining the size of the tree. Our proposed greedy method finds locally optimal splits, and works recursively in a top-down approach. To efficiently compute the best split, we carefully maintain certain counters. An experimental evaluation demonstrates its effectiveness. The proposed algorithm processes graphs with over million edges in a few minutes, can find the ground truth in synthetic data, and is able to produce interpretable clusters in real world networks.

0.35

Rennala NSGD: Asynchronous Stochastic Optimization Beyond Euclidean Geometry

cs.LG cs.DCmath.OC ·2026-10-28 ·Igor Sokolov, Alexander Tyurin, Peter Richtárik
top venue
Abstract · Accepted at NeurIPS 2026. 111 pages, 2 algorithms, 3 tables

The empirical success of matrix-aware optimizers, such as Muon (Jordan et al., 2024) and Shampoo (arXiv:1802.09568), highlights the potential of non-Euclidean geometry for deep learning, including large language model (LLM) training. Despite this progress, the theoretical understanding of stochastic optimization in general normed spaces remains limited, particularly in distributed settings. We analyze minibatch stochastic gradient descent (SGD) for smooth nonconvex optimization with respect to an arbitrary norm on a finite-dimensional space. Using the framework of $κ$-regular normed spaces (arXiv:0809.0813), we derive high-probability oracle complexity bounds under a light-tail condition stated directly in the dual norm induced by the update norm. The bounds make the dependence on geometry explicit through the regularity constant $κ$ of the dual space. In the Euclidean case, our bound recovers the standard minibatch SGD rate up to logarithmic factors. For some non-Euclidean geometries and noise models, it is sharper than bounds obtained by converting a Euclidean noise assumption, reducing a linear dimension dependence to a logarithmic one. Building on this framework, we introduce Rennala NSGD, a semi-asynchronous distributed algorithm, and establish wall-clock time complexity guarantees under fixed, heterogeneous worker computation times. In the Euclidean case ($κ=1$), our bound matches the parameter dependence of classical Rennala SGD up to logarithmic factors.

0.34

AI-Driven Digital Twin Framework for Warehouse Logistics Automation Optimization

cs.AI ·2026-10-28 ·Md Mahinur Alam, Kanita Jerin Tanha, Md Tanjin Alam, Shadman Sakib, Min Gyu Kim
published
Abstract · ICRA 2026 Workshop Advances and Challenges in AI-Driven Automation and Robotic System Integration with Digital Twins

In recent years, warehouse logistics optimization has increasingly focused on intelligent, data-driven systems to enhance operational efficiency. Conventional static product placement strategies are being progressively replaced by dynamic slotting approaches that can adapt to real-time variations in warehouse conditions. In this work, we present a novel framework that employs a reinforcement learning agent integrated within a FlexSim-based Digital Twin environment for real-time product placement optimization. The proposed agent continuously learns from streaming operational data, enabling it to reduce picking time while improving overall system throughput. The simulation environment models a four-zone warehouse layout with forklift-based material handling and leverages the Asset Administration Shell framework to ensure seamless interaction between the learning agent, physical asset, and the digital twin. Experimental evaluation demonstrates substantial performance gains, with the proposed approach achieving a 29.66\% reduction in average picking time, a 35.42\% increase in throughput, and an 18.54\% decrease in travel distance.

0.34

CHORD: Cross-Model Hallucination Detection via Relational Graph Discrimination

cs.AI ·2026-10-28 ·Yongxin Deng, Zhen Fang, Guansong Pang, Sharon Li, Ling Chen
top venuecode/data
Abstract · Accepted to NeurIPS 2026 as a Spotlight paper. 64 pages, 14 figures, 7 tables

Hallucination detection is critical for deploying large language models (LLMs) in real-world applications. Due to strong empirical performance, internal representation-based methods have emerged as the prevailing direction for detecting hallucinations, yet they remain largely model-specific and often fail to generalize to unseen LLMs. In this paper, we study an important yet underexplored problem, termed cross-model hallucination detection (CMHD), which aims to train hallucination detectors on source LLMs while ensuring performance on unseen target LLMs. The core challenge of CMHD lies in cross-model representation heterogeneity: hidden states from different LLMs exhibit severe semantic inconsistency, which limits the transferability of classical representation-based detectors. Through analysis and empirical validation, we show that relational structures over layers, tokens, and features capture transferable detection signals despite semantic inconsistency. Based on this observation, we propose cross-model hallucination detection via heterogeneity-oriented relational discrimination (CHORD), a relational graph discrimination framework that represents hidden states as joint relational graphs over layers, tokens, and features. It feeds these graphs into relation-aware attention to obtain structure-aware features, and uses meta-learning to shape these features toward transferable detection signals. Extensive experiments show that CHORD outperforms representative detection methods in cross-model generalization. Code is available at https://github.com/MiracleworkerD/CHORD.

0.33

Governance-Aware Federated Learning for Multi-Agent AI Systems: A Policy-Driven Framework for Trustworthy Collaborative AI

cs.AI ·2026-10-28 ·Kiranmayee Janardhan, Karthick S, Madhusmita Panda, Bharathi Vokkaliga Ganesh, Syed Ahmed
published
Abstract · 16 pages, 2 Figures and 3 Tables presented at NGEN 2026 Conference

Federated Learning (FL) enables collaborative model training without exchanging raw data, but conventional FL provides limited runtime governance, evidence verification, fairness assessment, explainability monitoring, and auditability. This paper proposes Governance-Aware Federated Learning (GAFL), which combines independent governance-evidence verification, temporal trust, fairness and explanation assessment, anomaly detection, policy enforcement, normalized aggregation, and tamper-evident audit logging. Client-reported governance values are authenticated, bound to their submitted model updates, and independently recomputed before aggregation. A cross-silo prototype with 10 participants was evaluated on the Adult Income dataset under honest, benign non-IID, biased-data, malicious-update, non-compliant, falsified-evidence, and mixed-threat scenarios. Four methods were compared across five random seeds: FedAvg, FedAvg with anomaly filtering, GAFL without independent verification, and complete GAFL. Complete GAFL intervened on all affected participants at the final round across the evaluated risk scenarios, including identifying 100% of falsified-evidence participants, whereas GAFL without verification identified none of the falsified-evidence participants. Final macro-F1 ranged from 0.7520 to 0.7733 across GAFL scenarios. Complete GAFL required 173.67 plus or minus 26.94 ms of governance processing per round, compared with 3.82 plus or minus 0.79 ms for FedAvg. A strict demographic-parity threshold also caused frequent interventions for honest participants, demonstrating a significant fairness-policy and utility trade-off.

0.33

UnEZ: A Fully Unsupervised Self-Templating Neural Network for Redshift and Line Strength Estimation

cs.LG astro-ph.IMastro-ph.GA ·2026-10-28 ·Asha Abd, Karl Glazebrook, Colin Jacobs
benchmark
Abstract · 22 pages, 17 figures. Submitted to MNRAS

As spectroscopic surveys grow, estimating galaxy properties such as metallicity, star formation rate, and redshift becomes computationally cumbersome. Methods of estimating such properties rely on template spectra or curated redshift catalogs, which fail to represent the diversity of objects seen in large-scale surveys. We present UnEZ, a fully unsupervised, physics-informed machine learning model that can predict spectroscopic redshifts and line strengths with no prior redshift or line strength labels. An encoder predicts emission and absorption line amplitudes/widths, and a fixed deterministic decoder constructs a unique template, estimated redshift, redshift confidence curve, and spectral model. UnEZ was tested on the Sloan Digital Sky Survey (SDSS) Data Release 18 due to its large dataset and thoroughly vetted redshift catalog which serves as an independent benchmark. Despite the lack of templates, line strength labels, or redshift labels, UnEZ's performance competes with modern redshift estimation algorithms. At SNR > 10, UnEZ achieves an outlier fraction of 0.459% and a normalized median absolute deviation (NMAD) of $1.237 \times 10^{-4}$. At SNR > 3, UnEZ achieves an outlier fraction of 3.89% and an NMAD of $1.471 \times 10^{-4}$. Furthermore, we demonstrate how UnEZ's line strength and redshift predictions can be utilized for various purposes such as constructing template spectra required in traditional redshift estimation algorithms. UnEZ illustrates how a fully unsupervised neural network can allow redshift, among other galaxy properties, to emerge naturally from the spectra themselves, providing a template-free, label-free approach that scales to the demands of future surveys.

0.33

When Should AI Decide, Defer, and Deviate? Responsible Automation for Deceased Donor Kidney Allocation

cs.AI ·2026-10-28 ·Madeleine Pollack, Thomas Daillak, Mehmet Toner, Korkut Uygun, Heidi Yeh +1 more
unusual
Abstract

We ask when an AI trained on historical decisions should be allowed to recommend decisions, defer to human experts, or in other cases, flag concerning patterns in data and deviate from past practice. This can be challenging because past decisions record what was done rather than what should have been done, and where data is sparse, a model's predictions may be too uncertain to act on. We provide a framework for addressing this problem using a case study in deceased-donor kidney allocation. Nearly 30% of kidneys recovered from deceased donors in the United States are never transplanted, largely because they are hard to place through the standard waitlist sequence. One can use AI to flag kidneys at risk of delayed or failed placement for out-of-sequence (OOS) allocation, but this comes at the cost of bypassing candidates higher on the waitlist. Thus, OOS allocation should only be used when the organ is demonstrably at high risk of discard on the standard sequence. Using Organ Procurement and Transplantation Network match-run data from 2016-2019 and 2023-2025 (excluding COVID-19-disrupted years), we build the Arbitrator framework, which combines various interpretable models and acts only when the uncertainty in their predictions cannot change the preferred action. Our Arbitrator resolves 93.5% of organ placement decisions using explicit medical criteria learned from data, and defers the remaining 6.5% to a human or a more complex model. We show that the Arbitrator correctly identifies 70.1% of historically discarded low-KDPI kidneys as OOS candidates, projecting that up to 43.9% of historically discarded kidneys could be flagged for expedited placement and potentially transplanted. We then exploit the interpretability of the Arbitrator to create an agentic audit comparing the learned rules to published clinical literature in order to flag decision-making patterns that are contentious.

0.33

RainCheck: Harnessing Historical Weather Data with Machine Learning for Rainfall Forecasting in Bangladesh

cs.LG ·2026-10-28 ·Shahran Rahman Alve, Muhammad Zawad Mahmud, Samiha Islam, Al Jubayer Pial
Abstract · 11 Pages

Machine-learning rainfall forecasts are often evaluated by shuffling station records into random training and test sets. For rainfall, this lets a model see what happened at neighbouring stations on the very day it is asked to forecast, and it hides the fact that a seasonal lookup table already explains most of the variance. We quantify both effects on 618,650 station-days from 35 Bangladesh Meteorological Department stations (1948-2014). We compare random, temporal (train <=1999, test 2005-2014) and spatio-temporal (unseen stations and unseen years) evaluation, calendar-only versus antecedent-rainfall features, and six learners against climatology, persistence and SARIMA. Three findings stand out. First, a station-month climatology alone classifies high-rainfall months with 88.4% accuracy (AUC 0.950), and no learner improved on it by more than 0.8 points under temporal evaluation. Second, a calendar-only random forest reached a next-day wet-day AUC of 0.889 under a random split, higher than any model with real predictors under temporal evaluation, yet fell to 0.835, below climatology (0.859), on future years. Third, models with lagged rainfall add genuine but modest skill for next-day forecasts: a Brier skill score of 0.074 (95% CI 0.064-0.088) over climatology, rising to 0.091 when station and year inputs are removed and to 0.131 when other stations' same-day observations are added. A GRU sequence model matches but does not exceed gradient boosting, and skill disappears beyond a two-day lead. For monthly totals, no learner beat climatology significantly. We release a leakage-aware protocol and code so that future station-based studies report skill against climatology on held-out years.

0.32

Replaying the Same Gradients at a Different $\varepsilon$: A Registered Test of the Fixed-Trajectory Account of Adam's $\varepsilon$ for One Rare Row

cs.LG ·2026-10-28 ·Lngel Kyo
Abstract · 24 pages, 2 figures

Adam divides every update by the square root of a running average of squared gradients plus a small constant, epsilon. For an output-embedding row whose token is rare, which we call the probe row, raising epsilon can be read as shrinking that row's steps by arithmetic alone. We test that reading directly. While a training run proceeds, we also divide the probe row's update by a different epsilon, using the run's own moment estimates; this replay is exact arithmetic on the gradients the run actually saw, so it holds the run's trajectory fixed. A rule for when the replay counts as wrong was written down before the runs. We train a small causal transformer in a standard configuration. When epsilon rises by four orders of magnitude, the replay counts as wrong on each of the three seeds registered for the test: the probe row keeps 8.9--11.4 times more of its displacement than predicted with the main, sampled-negative head, and 5.0--7.7 times more with a full-softmax head scored as a second instrument. At three orders of magnitude, tested with the main head only, it does not, by narrow margins; a further two seeds, registered after this result had been committed and scored on their own, follow the same pattern. In separate real runs, raising epsilon on the probe row alone is enough to reproduce the row's change and raising it on every other parameter alone also moves the row part of the way --- readings about the runs that do not rescue the replay; scaling the row's learning rate instead, with the denominator untouched, is mispredicted by the replay in the same direction. This is one embedding row of one model family on one corpus; we make no claim about rates, about rows in general (other rows were read only as a panel on one trajectory) or about how the effect changes with model size.

0.32

Auditable Hierarchical Inference with Multimodal Evidence and Bayesian Active Inquiry

cs.LG ·2026-10-28 ·Yucheng Liu, Everett X. Wang
Abstract

Most multimodal classifiers assume that all useful evidence is available upfront, whereas clinical inquiry is sequential. We present a synthetic evaluation framework separating objective multimodal inference from active inquiry. An objective model maps synthetic tongue, bilateral multipressure pulse, and acoustic observations to eight syndrome families. A provenance-bearing Knowledge Core and training-only answer model support Bayesian narrowing over 33 fine terms using 25 typed questions, while seven safety questions remain outside the diagnostic posterior. Weak shared symptom factors limit fine-term influence on objective phenotypes, and an objective-only probe tests leakage. In a preregistered 600-case synthetic evaluation, objective family accuracy was 87.0%. Cost-sensitive expected information gain (cost-EIG) raised family accuracy to 95.5% and exact fine-term accuracy to 42.5%, versus 92.7% and 33.0% for an unoptimized fixed-order baseline. Selective family accuracy reached 96.9% at 95.3% coverage. However, cost-EIG asked more questions on average than fixed order (9.795 versus 9.352). Missing-audio routing retained 94.0% family accuracy, and a workflow invariant enforced abstention when all objective modalities were absent. Permuting fine-term answer likelihoods reduced family accuracy to 78.7%. Five of six frozen checks passed, but the question-efficiency hypothesis failed. Comparisons are descriptive because only aggregate, not paired case-level, outputs were retained. These results establish a reproducible synthetic workflow, not clinical diagnostic or treatment validity.

0.32

A-RSI: Adaptive Task Refresh for Zero-Data Multimodal Self-Improvement

cs.AI ·2026-10-28 ·Yu Liu, Youxing Li, Yuedong Tan, Zhiwei Yang, Zihan Su +15 more
benchmarkcommunity pick
Abstract

Recursive self-improvement requires models not only to acquire useful training experience, but also to decide when their current experience has ceased to be useful. Existing methods often focus on what experience to generate next, while paying less attention to when fresh experience should be introduced. Many also construct training targets through self-consistency or majority voting, making supervision depend on the model's own uncertain predictions. We address these two questions separately: training dynamics determine when to acquire new experience, while the answers to new tasks remain independently checkable. A-RSI monitors reward variation and accuracy in training logs to refresh a task pool when its learning signal declines, without consulting validation or test scores. A static library of model-authored programs produces multimodal questions and derives their reference answers; we specify the task-generation contract, automatically verify chosen properties of the reward and refresh rules, and separately audit the generated answers. The training loop uses no external training examples or human-written answer labels. On the primary eleven-benchmark suite, A-RSI raises the performance average from 57.77 to 61.05 with 800 fresh tasks.

0.32

Why Speculative Decoding Works Better Than Predicted on Sparse MoE Models

cs.LG cs.AI ·2026-10-28 ·Ekagra Ranjan, Komal K. Teru, Bharat Venkitesh, Acyr Locatelli
top venue
Abstract · Accepted as at NeurIPS 2026 main track conference

Speculative decoding (SD) accelerates autoregressive inference by leveraging the memory bound nature of decoding, verifying multiple draft tokens in a single forward pass. For mixture-of-experts (MoE) models, prior work predicts a concave speedup-batch size relationship, with gains peaking at intermediate batch sizes due to expert saturation. We evaluate this across 10 sparse MoE models (30B--1T parameters, 8--512 experts) and find recent fine-grained models break this trend, showing higher speedup at low batch size. We find verification cost is governed not only by activated-expert bytes, but also by temporal overlap across draft tokens and GPU execution dynamics. At low batch size, MoE layers in fine-grained models account for a much smaller share of runtime than of weights; on H100, kernel profiling attributes this mainly to higher bandwidth utilization of MoE kernels relative to attention, with L2 absorbing repeated reads of co-routed experts. Across draft tokens, routing overlap further reduces the unique experts activated during verification relative to independence assumptions. We model this effect using a maximum-entropy Iterative Proportional Fitting (IPF) framework that predicts unique expert counts from routing-overlap statistics with less than 4\% error, and substantially narrows the gap to real-model curves and captures the qualitative low-batch trend from synthetic routing.

0.32

Investor's Last Exam: Can AI Agents Do Equity Investment Without Seeing the Future?

cs.AI ·2026-10-28 ·Huangfei Jiang, Anning Hu, Kai Sheng, Yuchen Cai, Rui Ge +5 more
benchmark
Abstract

Financial agents must form forecasts and investment judgments from decision-time evidence, with predictions evaluated against subsequent outcomes. Existing benchmarks assess financial knowledge, research workflows, or forecasting, but evaluating the full research process requires a strictly frozen information set and professional evaluation criteria. We introduce Investor's Last Exam (ILE), an equity-research benchmark led by experienced buy-side investors and fund managers. Its three components are a versioned financial database spanning 3.43 million document versions and nine source categories, supplemented by structured market data, with validated timestamps governing retrieval and computation; a task taxonomy organized around professional analyst deliverables; and an expert-led framework linking scoring criteria to sources, calculations, and assumptions through financial dependency graphs. Evaluation spans 86 assignments and jointly assesses research quality, forecast accuracy, and simulated investment performance. GPT-6 Astra and Claude Opus 5.5 lead overall at 68.1 and 66.9 out of 100, yet even Astra scores only 52.5 on Forecast & Model. Our results show that even state-of-the-art models fail to translate overall research capability into predictive performance. ILE shifts the evaluation unit from isolated capabilities to the evidence--assumption--judgment chain, allowing researchers to locate where analytical advantages fail to translate into forecasts and investment judgments, and providing a traceable, diagnostic foundation for training and evaluating financial agents.

0.32

Alpha-TCM: A Nine-Layer Framework for TCM-Guided Multimodal Inference and Structured-Prior Auditing

cs.LG ·2026-10-28 ·Yucheng Liu, Everett X. Wang
Abstract

Alpha-TCM makes two contributions under a central thesis: auditable systems should separate domain knowledge, case evidence, probabilistic inference, and downstream authority. First, it proposes a nine-layer TCM-oriented architecture covering semantic definitions, multimodal acquisition, quality control, provenance-bearing knowledge, probabilistic inference, longitudinal updating, constrained decisions, language interaction, and human governance. Each layer specifies its information contract, failure boundary, dependencies, and validation status. This is a system-design proposal, not a claim of complete clinical implementation. Second, a controlled synthetic prototype instantiates Layers 1-5 to audit the knowledge-evidence interface. Four conditional-Gaussian encoders process synthetic tongue images, pulse waveforms, inquiry fields, and text, while a frozen bridge predicts a five-dimensional synthetic outcome. A versioned contextual prior is compiled independently of the renderer. Fusion removes duplicated reference-prior information before adding context. Equal-information controls, prior regularization, graded contamination, and structural corruption test contextual benefits, Bayesian specificity, and conflict detection. Across ten paired seeds and four training sizes, contextual knowledge improved mean absolute error over marginal committee fusion at 600 and 1,200 cases, but not over equal-information neural controls. Fully incorrect context harmed performance at every size, while prior effects were inconsistent and conflict scores did not track case-level damage. The results support contextual structure in low-data synthetic settings, not a specifically Bayesian advantage. The prototype establishes no claims about TCM theory or clinical, diagnostic, therapeutic, causal, longitudinal, or transfer validity.

0.30

A goodness of fit approach to select the optimal visit sequence for nonparametric data synthesis

cs.LG stat.ML ·2026-10-28 ·R{é}my Chapelle, Bruno Falissard, Mohammed Sedki
Abstract

Synthetic data are a promising way to maintain privacy while disseminating research data, thereby enabling compliance with the open science principles. Popular methods for generating synthetic data include sequential synthesis, where each variable is generated conditionally on the preceding ones in a given order. This order, also called the visit sequence of the synthesizer, is known to affect the quality of synthetic data. This raises the question of the criteria that should guide its choice. In this work, we present a novel method to perform such selection based on a goodness of fit criterion, and show that it outperforms alternative structure-based methods, both in terms of synthesis quality and time complexity. The proposed method also has the advantage of being applicable to any type of data and any class of conditional distribution models.

0.30

PT$^3$: Prototype-Guided Self-Supervised Pretraining for Generalizable mmWave Human Point Cloud Representation

cs.AI ·2026-10-28 ·Ziyang Liu, Kieren Yu, Xiao Yang, Xin Yang, Kaishun Wu
Abstract · 30 pages

Multi-source millimeter-wave (mmWave) observations offer an opportunity to learn human representations reusable across sensing tasks. However, sparse detections and multipath effects make these observations unstable, challenging existing point cloud pretraining methods. We propose PT$^3$, a prototype-guided self-supervised pretraining framework for mmWave human point clouds. PT$^3$ represents local states in spatial regions as sparse compositions of shared prototypes and learns their temporal evolution through autoregressive prediction. Observation consistency further stabilizes representations under radar perturbations. After pretraining on unlabeled sequences from multiple sources, the same frozen encoder supports action classification, dense point generation, joint pose estimation, and future motion prediction on datasets not used for pretraining. Across these tasks, PT$^3$ outperforms the compared pretraining baselines and achieves competitive or superior performance to task-specific models.

0.30

From Grind to Grasp

cs.AI cs.MA ·2026-10-28 ·Peter Sunehag
Abstract

Discussions of complexity, knowledge and intelligence often overlook the work involved. For example, outputs generated by short programs have low algorithmic complexity but can have high computational depth, the least work needed to produce them from near-shortest programs. Producing them with less work can require longer descriptions, giving high work-bounded complexity. We introduce a formalism that compares achievements under work and information limits and describes how learning and persistence can make deep achievements shallow by reusing products of earlier work. For an agent with a limited work budget, the value of building and maintaining such resources depends on how they support its goals over later tasks. We define grasp as performance across work budgets at a fixed information allowance. Low grasp can still yield success through grind at high cost in work, without necessarily improving human grasp. Applied to improvement itself, grasp gives a criterion for recursive self-improvement in which successors make comparable further improvements more efficiently. Another route to faster progress is to expand the resources brought to bear through knowledge sharing and coordination across society, even while comparable advances require increasing work. Grasp also applies to collectives, which may outperform their members yet require more total work than one system with their combined resources. Establishing and sustaining organisation can itself be deep even when the primary task is simple. These demands may help explain the richness of biological and social systems. In agent collectives, this organisational work can generate goals beyond assigned tasks, including maintaining the collective itself. We therefore caution against allowing an emergent agent society to assume authority over social coordination, since it could increasingly direct work towards its own persistence and goals.

0.30

Compute Has Shape: Self-Morphing the Internal Computation of Frozen LLMs

cs.AI ·2026-10-28 ·Junxi Wu, Yiwen Guo, Shu-Tao Xia
code/data
Abstract

Test-time scaling improves large language models through additional computation, but how to adapt the topology of internal computation to individual inputs remains underexplored. Existing approaches largely fix recurrent or parallel architectures during training, or reconfigure frozen models along a single structural axis. We introduce SelfMorph, which makes computational shape an input-dependent inference-time decision. It unifies the original forward path (BASE), serial reuse of a layer span (REPEAT), and local branching and merging of perturbed hidden states (BRANCH) in a unified space parameterized by type, location, and intensity. Since they offer complementary benefits rather than uniform dominance, we use a lightweight selector that reads the backbone's depth-wise hidden state trajectory on the prompt to choose a morphology before response generation. We train the selector by distilling utility-based distribution from offline morphing that jointly account for task performance and computational cost. Evaluated on three LLMs spanning 4B-27B, SelfMorph achieves a 8.06% mean relative improvement on five in-distribution benchmarks and 4.45% on four unseen benchmarks. It achieves the strongest performance in 24 of 27 settings. These findings identify input-adaptive computational topology as a complementary axis of test-time scaling beyond generating more tokens. Our code is available at https://github.com/creator-xi/SelfMorph.

0.30

Representation Complexity Along Learning Trajectories: Trading Width for Precision and Time

cs.LG cs.AI ·2026-10-28 ·Ryan Singh, Miguel De Llanza, Christopher L. Buckley
theoryhot direction
Abstract

What computational role does overparameterisation play during learning? Information-plane analyses suggest that neural networks can traverse representations substantially more complex than those required by their final predictors. We explore when this transient complexity serves a computational purpose by comparing the representation complexity needed for prediction with that needed to make an unknown target statistically visible through gradient feedback. We introduce an oracle model of neural network learning and use Kolmogorov \(n\)-widths to quantify effective representation complexity. Our main technical result transfers complexity bounds from the forward representation to the corresponding gradient-query features. Combined with statistical-query lower bounds, this gives a task-agnostic lower bound on the initial effective width required for target-dependent gradient information to become detectable. For \(k\)-sparse parity in \(D\) dimensions, a known target admits an \(O(k)\)-width representation, while the general bound already implies a polynomial separation between representation and discovery. We then exploit parity's invariance under coordinatewise sign reflections to identify a sharper unsigned effective width, which removes sign distinctions invisible to the parity information channel. At a transfer-compatible measurement resolution, its critical scale is \(\widetildeΘ_k\left( D^{k-1}τ^2 \right)\), up to logarithmic factors. Thus, the representation resources required to discover a predictor can substantially exceed those required to represent it once learned. Interpreting precision through accumulation over fresh samples yields a corresponding width-time frontier. Experiments support the predicted dimension dependence and examine how optimisation and generalisation depart from this idealised frontier.

0.30

Autonomous Quantum Measurement with Agentic AI

cs.AI cond-mat.mtrl-sci ·2026-10-28 ·Eunbi Rha, Abhijatmedhi Chotrattanapituk, Mouyang Cheng, Chung-tao Chou, Shenyong Ye +4 more
Abstract

Quantum materials exhibit diverse electronic and functional phenomena, but their characterization requires extensive experimental expertise, iterative decision-making and accumulated laboratory knowledge that is difficult to document and transfer. Here we introduce Quantum Agent (QUANTA), an agentic artificial intelligence (AI) framework for autonomous quantum-materials characterization. Using a widely adopted Physical Property Measurement System (PPMS) as an experimental platform, QUANTA combines retrieval augmentation using scientific literature, experimental protocols and prior measurements with specialized agents for experimental planning, troubleshooting, and data interpretation. A separate deterministic control layer validates instrument actions. Guided by incoming data, QUANTA evaluates measurement quality, revises experimental plans and selects subsequent measurements. We demonstrate these capabilities through transport measurements of the topological Weyl semimetal TaP, as well as investigations of nonlinear responses in magnetic material Co2MnGa. These results establish QUANTA as a generalizable framework for evidence-driven, safeguarded and closed-loop quantum-materials characterization.

0.30

Multi-Token Prediction Done Right

cs.LG ·2026-10-28 ·Jayden Teoh
code/datacommunity pick
Abstract · Preprint

Multi-Token Prediction (MTP) is now a common ingredient in language-model pre-training, as it improves downstream performance and enables faster inference through speculative decoding. We identify a flaw in standard MTP training, in which a residual component of the MTP gradient interferes with both next-token and multi-token learning, increasingly so with scale. Specifically, the MTP cross-entropy gradient can be decomposed exactly into a KL self-distillation term that pulls each MTP head toward the model's own next-token prediction (NTP) distribution and improves speculative decoding, and a residual that has a low signal-to-noise ratio and buries the useful KL signal. We introduce the KL gradient-recovery trick: when two distributions share a linear output head, the gradient of their KL divergence can be recovered exactly by combining their cross-entropy gradients through straight-through gradient routing. Unlike minimizing the KL directly, this trick requires no vocabulary-sized logit matrices and is compatible with standard fused cross-entropy kernels. This enables Soft Multi-Token Prediction (Soft-MTP), a one-line code change to MTP that removes the noisy residual and backpropagates only the KL self-distillation signal by reusing the existing NTP and MTP cross-entropy computations. Soft-MTP supports both DeepSeek- and Meta-style MTP, adds no parameters, and incurs no measurable overhead. From 167M to 1.2B parameters, Soft-MTP improves over standard MTP with lower NTP perplexity, faster speculative decoding and higher zero-shot accuracy at every scale, lower MTP perplexity from 689M onward, and stronger math and code reasoning after fine-tuning at 1.2B. Soft-MTP also matches, if not exceeds, direct KL divergence minimization on most metrics. Soft-MTP is, in effect, a free lunch for MTP optimization. Our code can be found at https://github.com/JaydenTeoh/Soft-MTP.

0.30

Memorization Capacity of Invariant Convolutional Neural Nets

cs.LG stat.MLmath.ST ·2026-10-28 ·Sjoerd Dirksen, Patrick Finke, Paul Geuchen
theory
Abstract

Neural networks are powerful memorizers which are often able to interpolate their training data even when the labels have been randomized. While there exists an extensive literature on the memorization capacity of fully connected networks, in practice more specific architectures with problem-dependent properties are preferred, such as convolutional networks for image classification. In this paper we consider convolutional networks which are invariant with respect to a given arbitrary finite group of orthogonal transformations, which may include translations, reflections, and/or rotations of images and invariances of higher order tensors. We prove upper bounds on the size of an invariant convolutional neural network with threshold activations that is required to interpolate a given dataset with two classes. Our bounds are formulated in terms of a complexity measure that takes the invariant structure into account and is instance-specific, i.e., it is related to the mutual arrangement of the classes in the given dataset and their geometric properties rather than the number of samples. Our proof is constructive: for given input data, instances of an interpolating network can be computed via a randomized algorithm with high probability. In numerical experiments we investigate the practical effectiveness of this algorithm.

0.30

MiniMMD: Minimization of MMD for Sampling, Inference and Integration

cs.LG stat.ML ·2026-10-28 ·Zonghao Chen
Abstract

Many problems in computational statistics and machine learning can be formulated as approximating a target distribution by minimizing a discrepancy over a restricted class of candidate distributions. This minimum-discrepancy viewpoint appears, for example, in numerical integration, sampling, generative modelling, and statistical inference, but these instances are often studied separately in the literature. This thesis develops the first unified treatment of these many problems through the minimization of the maximum mean discrepancy (MMD), a kernel-based discrepancy between probability measures. The central theme is that MMD is not only a statistically meaningful distance between probability distributions, but also a tractable optimization objective to design practical and efficient algorithms. This thesis summarizes the recent theoretical and empirical advancements of such minimum-MMD based algorithms.

0.29

Syntactic Segmentation and Collage Generation for Videogame Levels with Perceptual Distance

cs.AI ·2026-10-28 ·Ross Mawhorter, Matthew Guzdial
Abstract · 11 pages, 12 figures, 22nd Artificial Intelligence and Interactive Digital Entertainment Conference

In procedural videogame level design, many existing techniques combine hand-authored level chunks. Some existing analysis techniques also rely on a higher-level graph structure which abstracts the level design. In this paper, we explore techniques for computing both of these things automatically from an existing design. We introduce the concept of perceptual distance, based on analysis techniques from the architecture subfield of Space Syntax. We use this perceptual distance to segment level designs with hierarchical clustering. We compare with past work on StarCraft analysis, show results on rooms from Super Metroid, and also demonstrate a proof-of-concept level generation algorithm based on this segmentation method.

0.29

"Close enough?" An investigation into multiplication errors in LLMs

cs.AI ·2026-10-28 ·William Bolton, Bezawit Tilaye, Huw Day
Abstract · Accepted to Neurips 2026 NewInML workshop

Large language models (LLMs) frequently fail at multi-digit arithmetic, but the underlying mechanisms governing these failures remain poorly understood. While recent work demonstrates that addition errors follow a "noisy quantisation" model localised around carry boundaries, whether this framework generalises to multiplication remains an open question. In this work, we evaluate multiplication capabilities across several small base LLMs on a dataset of 16,991 problems spanning varied operand lengths. We show that multiplication errors do not conform to the addition-style noisy quantisation model. Instead, we uncover a novel error regime in small Qwen models: errors in lower-order columns predominantly preserve parity. Conversely, higher-order digit errors shift toward off-by-one carry failures. Our findings suggest that arithmetic computation in LLMs relies on distinct, multi-periodic internal representations where parity is tracked more robustly than exact digit identity, providing new mechanistic insights into numerical reasoning in transformers.

0.29

Dungeons & Dragons Encounter Analysis with Semi-Random Simulation in Foundry VTT

cs.AI ·2026-10-28 ·ravyn Hardcastle, Matthew Guzdial
Abstract · 10 pages, 6 figures, 22nd Artificial Intelligence and Interactive Digital Entertainment Conference

Dungeons & Dragons (D&D) is a tabletop roleplaying game (TTRPG) in which a group of players, guided by a Game Master, roleplay a fantasy adventure with simulated combat and improvisational narrative. In D&D, an encounter is a single instance of simulated combat, where the players must use their characters to attempt to defeat a selection of enemy characters that the GM controls. D&D has multiple systems that approximate encounter difficulty by summing individual difficulty values that designers predefine for each enemy. However, since players and enemies can have a wide variety of capabilities, encounter difficulty depends on more than just the estimated difficulty of each monster. For example, these metrics do not account for positioning and dungeon layout, the synergies between monster abilities, and player party composition. We have created a tool, Mage-Hand, that can simulate D&D encounters semi-randomly using Foundry Virtual Tabletop, a virtual tabletop (VTT) system. In this paper, we show that our tool's simulated winrates have a strong correlation with D&D's published difficulty evaluation metrics, while also better aligning with some players' reported experiences.

0.29

Decompose, Don't Expand: Capability-Independent AI Systems by Construction

cs.AI cs.HC ·2026-10-28 ·Tejaswi Gowda
Abstract

A recurring design choice in building AI systems is whether, when a task is too hard for a model, to expand the model (make it larger, better-prompted, fine-tuned), or to decompose the task (extract the parts that admit a deterministic solution to a host, leaving the model only the irreducibly fuzzy residue). We make three claims. First, that many tasks commonly delegated whole to a model are in fact deterministic tasks with a small generative core. Second, and our central empirical result, that decomposing such a task yields a measurable and desirable property we call parity by construction: on the extracted sub-tasks the model-size performance curve becomes flat, because no model performs them at all. We measure this across nine models spanning a 494M-parameter local checkpoint to a frontier Opus-class model, a thirty-fold parameter range on the local family alone: once the host resolves the context, operation selection scores identically for all nine models and selector resolution for eight of nine, while the one task that does not decompose, labelling, remains sloped across the same range. Because the task leaves the model, a stock checkpoint suffices: no task-specific training and no fine-tuning. Third, that this is a general design discipline, not a domain trick: we show the same move recurring across physics setup, motion retargeting, and object-model editing, and in a Web-CLI tool family (same author, so not an independent replication). We give a method for locating the cut, the construction that makes a cut exist (a deterministic surface and a closed grammar over it), and a way to make the residue safe (resolve the deterministic context; verify the generative output), position the work against neuro-symbolic AI, tool-augmented language models, and generate-and-verify approaches, and are explicit about what is measured (one domain) versus illustrative (the rest).

0.29

Automatic Construction of Strong Simple Baselines

cs.LG q-bio.QM ·2026-10-28 ·Caleb N. Ellington, Jiaqi Wang, Sohan Addagudi, Elijah Cole
Abstract

To rigorously establish the value of a novel deep learning method, it must be compared against appropriate baselines. As deep learning has become increasingly common in the physical and life sciences, there have been numerous cases in which apparent progress has turned out to be an artifact of omitted or missing baselines. In this work, we investigate whether this problem can be mitigated by using large language models to automate the construction of baselines. We assemble 36 supervised classification and regression tasks from peer-reviewed studies in Nature, Science, and Cell-family journals that investigate novel deep learning methods for scientific applications. We then use a language model to construct simple baselines for each task. These automatically constructed baselines are strong, recovering a median of 95.5% (mean 85.7%) of the gap between a trivial predictor and the score of the deep learning model. The median simple baseline matches or exceeds the performance of the deep learning model in 12 studies. We further examine the variation across scientific domains and publication venues and we characterize the effects of test-time compute and access to different levels of study context. As human peer review has struggled to keep up with the rapidly increasing volume of research articles in recent years, our results suggest that automated baseline construction is a practical and scalable check on deep learning contributions in the physical and life sciences.

0.29

Tunable Spatial Task Allocation with Application to Maritime Swarms

cs.MA eess.SYmath.OC ·2026-10-28 ·Vishnudatta Thota, Tyler Paine, Anastasia Bizyaeva
Abstract · 29 pages, 10 figures

In this chapter, we explore how a team of autonomous agents can flexibly reconfigure the distribution of its members across many spatially embedded tasks using a task allocation strategy grounded in a generalization of the recently proposed Nonlinear Opinion Dynamics model. Motivated by an adaptive maritime survey application, we couple opinion dynamics directly to spatial movement and examine how the emergent macroscopic allocation is shaped by the interplay between environmental geometry, nature of social interactions, and task demands. We illustrate in simulation how social coupling in the model tunes group cohesion, producing macroscopic allocation patterns ranging from maximally dispersed states, to intermediate semi-dispersed states with a smaller number of distinct subgroups, to maximally cohesive consensus states, in distinct and interpretable parameter regimes. Furthermore, we demonstrate with illustrative examples how a team can reconfigure itself in real time under changing task demands. For allocation scenarios with distance-based task preference and matched number of agents and tasks, we relate numerically the final allocation in the group to the Hungarian optimum, illustrating close-to-optimal performance. We illustrate performance of this method under realistic communication constraints in the MOOS-IvP-Multi simulation environment with simulated autonomous underwater vehicles (AUVs). Together, these examples show how a common set of local interaction rules can support a large range of interpretable macroscopic allocation states and flexible transitions between them.

0.29

A Gradient based Adversarial Augmentation Framework for Mitigating Unlearnable Datasets

cs.LG ·2026-10-28 ·Long Dang, Satpal Sura, Kaiqi Xiong
published
Abstract · Accepted at IEEE DSC 2026. 8 pages

Unlearnable data methods (ULD) protect public data from unauthorized model training by adding imperceptible perturbations that create a learning shortcut. A deep neural network can exploit the shortcut learning path to minimize training loss without learning generalizable features. Intuitively, finding adversarial versions of perturbed images that can substantially increase the model's loss can force the model to learn generalizable features. Existing data augmentation based attacks rely on heuristic search strategies/random sampling over a curated augmentation set to identify adversarial augmentation operations, an approach that often does not scale when the number of candidate operations is large. To address the challenge of a large search space, we propose {\techname}, a gradient based adversarial augmentation framework that learns an augmentation policy via stochastic gradient ascent. {\techname} leverages a differentiable data augmentation pipeline to efficiently search a set of up to 18 augmentation techniques, from which a grid search selects a protection-specific subset. The augmentation subset is trained to maximize a surrogate model's classification loss. We introduce a perceptual-plus-pixel regularization to learn operation parameters that preserve image semantics. Under a controlled ablation, {\techname} consistently outperforms an equivalent random sampling baseline (80.0\% vs. 77.29\% clean test accuracy on EMIN-protected CIFAR-10 for ResNet-18 and 83.28\% vs. 66.45\% on NTGA-protected CIFAR-10 for MobileNet-V2), demonstrating that gradient guidance drives the clean test accuracy recoverability rather than the expanded search space. {\techname} offers an effective and scalable adversarial augmentation framework for stress-testing ULD methods, a critical tool for developing robust ULD methods.

0.29

Causal Attention on CPU: A Cache-Aware and Load-Balanced Scheduling Kernel for Prefill

cs.LG ·2026-10-28 ·Kwun Fung Lau, Frost Mitchell, Kanya Mo, Emani Ashok
hot direction
Abstract · Preprint, 18 pages

Agentic AI drives significant productivity gains across diverse domains, but concurrent long-context prefills from multiple agents can saturate GPU compute resources. Although CPUs offer a viable alternative, existing CPU attention kernels fall short in adapting to multi-tier cache hierarchies and lack effective thread scheduling. To address this, we introduce CACPU, a high-performance attention kernel featuring tiered cache blocking and an in-kernel task dispatching mechanism designed for causal full-sequence prefilling. CACPU achieves up to a 2.01x speedup over PyTorch SDPA, FlexAttention, and SGLang attention kernels. Integrated into SGLang, it reduces Time to First Token (TTFT) of Llama-3.1-8B by up to 1.74x while preserving model accuracy.

0.29

Inner-Learning: From Hand-Coded Beliefs to Declarative Knowledge in Trick-Taking Games

cs.AI ·2026-10-28 ·Stefan Edelkamp
Abstract

We present a declarative approach to belief-state modeling in trick-taking card games. Starting from a state-of-the-art Skat player, we automatically extract its hand-coded inference rules into a domain-specific language built on the primitive $\neg\mathsf{holds}(P,C)$, denoting that player $P$ cannot hold any card in set $C$. Round-trip code generation verifies that the extracted rules reproduce the original engine's intended behavior, up to three audited bugs in the original that the declarative engine corrects. We then grow the rule base through symmetric derivation and annotation of an expert strategy document---unifying code, symmetry, and expertise in one representation. We extend the knowledge representation with \emph{cardinality constraints}: at-most-$k$ and at-least-$k$ bounds on the cards a player holds from a named group, expressed in the same declarative language and enforced at world-generation time. Confidence parameters connect symbolic rules to probabilistic belief updates. Self-play evaluation shows the declarative engine measurably improving winning percentage and score over the original hand-coded engine, with the extracted rules, the human-annotated rules, and the cardinality constraints each contributing to the effect.

0.29

Communication-Efficient Clustered Federated Learning in Non-IID Wireless Networks

cs.LG ·2026-10-28 ·Martin Isaksson, Rickard Cöster, Šarūnas Girdzijauskas, Seif Haridi
Abstract · 64 pages (25 main text + appendices)

Current and emerging mobile networks face increasing privacy, scalability, and performance challenges as distributed learning becomes essential for advanced 6G services. These challenges are amplified by the presence of non-IID data, naturally arising in heterogeneous wireless environments, that degrade the performance of standard Federated Learning (FL) approaches. To address non-IID data across clients, we propose a Clustered Federated Learning (CFL) framework that assigns clients to specialized models during training. Data-domain clustering initializes model-based clustering, enabling identification and exploitation of client similarities without sharing raw data while reducing communication cost. Motivated by mmWave beam selection, we evaluate our methods on a wireless simulation dataset (DeepMIMO) and complement it with two non-IID benchmarks (FEMNIST, Rotated CIFAR-10) to isolate different non-IID characteristics. Our best configurations improve accuracy over IFCA by 4.8 pp on DeepMIMO and 10.6 pp on FEMNIST while reducing downlink data volume by 83.1 % on FEMNIST compared to IFCA. We demonstrate that different clustering strategies suit different non-IID characteristics. These results show that our CFL methods enable resource-efficient, and high-performance distributed learning in future 6G networks.

0.28

MUSE: Multi-Level Memory with Selective Runtime Injection for Scientific Coding Agents

cs.LG cs.AI ·2026-10-28 ·Haoran Jin, Defu Lian, Xiting Wang, Hong Xie
code/data
Abstract · 17 pages, 8 figures. Under review

Historical experience can guide scientific coding at multiple levels, but task-level retrieval may miss implementation lessons that become relevant during execution. We introduce MUSE, a memory management framework combining multi-level memory with selective runtime injection for scientific coding agents. MUSE builds global, approach-level, and implementation-level memory from historical trajectories. At initialization, the agent receives global memory and problem-matched approach-level memory. During execution, current interactions are used to retrieve implementation-level memory. An LLM gate selects relevant experience and generates specific review advice, or skips injection. Across four benchmarks spanning numerical problem solving, kernel optimization, and scientific software repair, MUSE outperforms retrieval and memory baselines on all evaluation metrics. Ablations support the multi-level design and gated injection, while transfer experiments show that memory remains useful across kernel benchmarks. These results show that multi-level memory and selective runtime injection enable effective reuse of experience in scientific coding. Code is available at https://github.com/hr-jin/MUSE.

0.28

MAS-Loop: Beyond One-Shot Orchestration towards Self-Evolving Multi-Agent Systems

cs.MA cs.AI ·2026-10-28 ·Hehai Lin, Qi Yang, Chengwei Qin
Abstract

Training-time multi-agent system (MAS) design seeks to learn reusable orchestration policies, yet prevailing approaches remain largely one-shot: an Orchestrator generates a system but only observes and learns from its terminal outcome. This paradigm discards execution trajectories that expose system-level failures and contain fine-grained information beyond the terminal outcome, leaving no direct basis for revising the source MAS, evaluating the updated MAS, or learning from the resulting contrast. To this end, we introduce MAS-Loop, an execution-grounded framework that closes this feedback loop by coupling an Orchestrator with a trajectory-conditioned Refiner. For each task, the Orchestrator constructs and executes a source MAS to obtain an outcome score and an intermediate trajectory. Conditioned on the trajectory, the Refiner localizes an actionable failure, formulates a trajectory-grounded diagnosis, and proposes a bounded patch to the agents, information flow, or execution configuration. The executable updated MAS is re-executed on the same task to measure its realized outcome under stochastic execution. This pair-MAS comparison provides complementary learning signals: the Orchestrator learns to generate strong systems with fewer recoverable failures, whereas the Refiner learns to expose and repair observed weaknesses, thereby driving policy-level co-evolution across training iterations. Experiments show that MAS-Loop achieves the strongest overall performance among the evaluated baselines, with consistent gains from refinement across all evaluated benchmarks. The co-evolution process also yields MAS-JudgeCorpus, a dataset of execution-grounded source-updated pairs with orchestration- and trajectory-level supervision, which can provide the community with support for MAS reward modeling and trajectory-aware investigation.

0.28

Dual DeltaNet: Value-Side Memory Reorganization for Linear Attention

cs.LG ·2026-10-28 ·Yixiao Qian, Yaxin Zhang, Song Chen, Jiaxu Liu, Shengze Cai +1 more
code/data
Abstract

Linear attention and state space models (SSMs) offer efficient alternatives to softmax attention by maintaining a fixed-size recurrent state, but this design requires careful management to preserve useful historical associations and limit interference. The delta rule helps reduce interference through correction of the value associated with each key, but it does not explicitly account for how related value features are distributed across the state. We observe that a state $\boldsymbol{S}\in\mathbb{R}^{d_k\times d_v}$ can be viewed as a bank of $d_k$ addresses storing $d_v$-dimensional values. While a key-side read $\boldsymbol{S}^\top\boldsymbol{k}$ retrieves the value associated with $\boldsymbol{k}$, a value-side read $\boldsymbol{S}\boldsymbol{e}$ aggregates addresses whose stored value features match the selector $\boldsymbol{e}$. Based on this observation, we propose Dual DeltaNet (DDN), which extends Kimi Delta Attention (KDA) with a two-sided delta rule. The key-side delta rule reads with the current key to correct its associated value, while the complementary value-side delta rule reads with selected features of the current value to reorganize the addresses it occupies. We derive a chunkwise parallel training algorithm and a rank-2 recurrence that increases arithmetic intensity during inference. Compared with KDA, GDN-2, and Mamba-3, DDN achieves the strongest overall performance across language modeling, zero-shot downstream tasks, and retrieval, while incurring modest training and inference overhead relative to KDA. Code: https://github.com/ReichtumQian/DualDeltaNet

0.28

Estimating Implicit Expected Log Density Ratios

cs.LG cs.AIstat.ML ·2026-10-28 ·Avi Amalanshu, Chandler Squires, Yizhou Lu, Pradeep Ravikumar
position
Abstract

When exact probabilities cannot be computed, \emph{density ratio estimation (DRE)} is a key subroutine in the wide range of applications where density ratios appear, from experimental design to generative modelling and reinforcement learning. In this paper, we introduce the concept of an \emph{expected log density ratio (ELDR)}: log density ratios that are integrated under some third, possibly distinct measure. We show many problems with implicit densities where DRE is naïvely used are actually ELDR estimation tasks, and prove fundamental limitations on finite-sample ELDR estimation. Based on these results, we argue that the typical DRE-based approaches are unsuitable for ELDR estimation, and provide some foundational motivation towards bespoke ELDR estimators. To demonstrate, we propose a simple modification to some DRE methods that allows samples convergent to the integral distribution to appear in training data. We show through benchmarks that this can improve ELDR estimation error.

0.28

Weaken to Strengthen: Weakening-Induced On-Policy Self-Distillation for Large Reasoning Models

cs.AI cs.CLcs.LG ·2026-10-28 ·Yecheng Wu, Song Han, Han Cai
code/data
Abstract · Work in progress. Code is released at https://github.com/jet-ai-projects/WISD

Self-distillation offers a way to improve reasoning models when a suitable external teacher is unavailable. However, common approaches rely on privileged information to construct a stronger self-teacher, limiting their use when only unlabeled prompts are available. In this paper, we introduce Weakening-Induced On-Policy Self-Distillation (WISD), which instead constructs a weaker counterpart of the original model and uses the contrast between the two as a training signal without requiring privileged information. We explore two ways to construct the weaker model: low-bit quantization and magnitude pruning. We evaluate WISD across diverse architectures and scales on mathematical and coding reasoning tasks. Across these benchmarks, quantization and pruning improve average accuracy over the original models by 4.0 and 5.5 percentage points, respectively, achieving state-of-the-art performance even without privileged information. Notably, pruning raises Qwen3.5-4B's accuracy from 57.7% to 70.5% on HMMT 2025 and from 39.1% to 56.7% on LiveCodeBench v6. Quantization raises Qwen3.6-35B-A3B's accuracy from 68.5% to 74.6% and from 61.8% to 69.5% on the same benchmarks, respectively. These results establish WISD as a novel and effective self-distillation framework for large reasoning models.

0.28

Agentic AI Reasoning and Learned Sampling for Task-Aware Projection Selection in Sparse-View CT: Predicting Hemodynamics

cs.AI physics.flu-dyn ·2026-10-28 ·Mahmoud Elhadidy, Roshan M. D'Souza, Amirhossein Arzani
Abstract

Sparse-view computed tomography (CT) reduces the number of acquired projections, but conventional sampling strategies are designed primarily for image reconstruction rather than for downstream physical inference. Here, we investigate optimal task-aware projection selection for directly predicting hemodynamics from sparse CT angiography (CTA) measurements. Using computational fluid dynamics simulations of intracranial aneurysms, we generate synthetic fan-beam CTA projections and train a Fourier neural operator (FNO) to predict two targets: the spatially resolved time-averaged wall shear stress (TAWSS) field and the spatially averaged TAWSS. The FNO was coupled with optimal input selection for simultaneous optimal sampling and training. Our goal was to investigate if agentic AI could discover optimal sampling strategies that outperform current methods. We compared fixed uniform and golden-angle sampling with three task-aware optimal selection strategies: Gumbel-Softmax, a Soft top-K method, and an agentic AI selector. The methods were compared at different view selection budgets. A detailed analysis of the AI agent was performed to understand its reasoning and performance. For spatially resolved TAWSS, Agentic and Soft top-K achieve the lowest mean errors (Agentic performs slightly better) and substantially outperform fixed sampling under extreme sparsity. For spatially averaged TAWSS, optimized selection becomes important at extremely sparse acquisitions. Analysis of the selected masks reveals multiple distinct subsets concentrated around informative regions. Moreover, masks optimized for hemodynamic prediction yield poor image reconstructions, showing that task-optimal acquisition differs fundamentally from reconstruction-optimal acquisition. This work suggests the possibility of designing sparse acquisitions directly around downstream physical quantities.

0.27

GAAC: Lossless KV Compression for Bandwidth-Efficient LLM Inference on Personal Devices

cs.AI ·2026-10-28 ·Yipin Guo, Arun M George, Siddharth Joshi
code/data
Abstract

Weight-only, lossy quantization substantially reduces the storage and memory requirements for serving large language models (LLMs), but the growing key-value (KV) cache remains a bottleneck for long-context inference on personal devices. We present Group-Adaptive Additive Codec (GAAC), a lossless KV compression and serving system that exploits exponent redundancy to reduce storage and accelerate attention without introducing KV approximation. GAAC encodes BF16 exponents using calibrated per-channel base-plus-offset codes over contiguous ranges. GAAC adapts code widths across small groups while preserving KV values bit-for-bit. A fused attention kernel reconstructs KV tiles on the fly, avoiding full-cache decompression. Block-wise compression and fixed-capacity slots support incremental cache updates and paged serving. Across five tested models, GAAC achieves 1.405-1.407$\times$ lossless BF16 KV compression. We also study compression performance on low-precision tensors FP8 E4M3 caches ( 1.09-1.14$\times$) and E5M2 caches (1.25-1.30$\times$). On NVIDIA DGX Spark, GAAC accelerates decode attention at a 64K-token context by 1.19-1.27$\times$ and end-to-end W4A16 decode steps by 1.04-1.15$\times$. In serving, it reduces resident KV memory by approximately 26% and increases aggregate throughput by 13-29% under comparable KV-memory budgets. On Apple M4, GAAC achieves 1.02-1.25$\times$ attention speedups, demonstrating portability beyond NVIDIA GPUs. Code is available at https://github.com/Intelligent-Microsystems-Lab/GAAC.

0.27

Loss is Not Behavior: A Unified Output-Space Analysis of Gradient-Based Machine Unlearning

cs.LG ·2026-10-28 ·Xingjian Zhao, Mohammad Mohammadi Amiri, Malik Magdon-Ismail
top venue
Abstract · Accepted at NeurIPS 2026

Gradient-based machine unlearning methods typically ``unlearn'' by optimizing forget and retain losses, but similar losses do not necessarily imply similar output-space behavior. Prior work has shown that high forget loss alone does not certify that unlearning has produced retraining-like output behavior. We therefore ask: when a gradient-based unlearner differs from the model that would have been learned on only the retain data, what structure does the resulting output-space gap have? We derive closed-form expressions for this gap in linear regression, with an extension to nonlinear models via a generalized Gauss-Newton and local linearity approximations. Our analysis puts various gradient-based methods under a unified framework parametrized by geometry, reweighting, and projection. Two leverage matrices emerge that capture the forget-retain entanglement under the retrained model geometry: forget self-leverage and retain cross-leverage. We show both theoretically and empirically that these two leverage matrices drive the impact of the forget residual on the forget and retain outputs, and existing methods account for this geometry only partially. Our framework yields precomputable leverage diagnostics for auditing forget sets to identify what type of gradient-based unlearning is likely to be successful.

0.27

Formal Specifications with Data-Driven Signal Analysis

cs.AI cs.LG ·2026-10-28 ·Roderick Bloem, Benedikt Maderbacher, Cristinel Mateis, Dejan Ničković, Sergii Tkachov +1 more
unusual
Abstract

Developing analog and hybrid systems requires substantial domain knowledge that is hard to articulate in a formal specification; without additional effort, this domain knowledge is also inaccessible to machine learning methods. We propose a hybrid approach for analyzing and monitoring real-valued signals that integrates specification-based and data-driven techniques. We use regular expressions with interval operations to identify regions of interest in a signal, classical signal processing methods to extract semantically meaningful features, and semi-supervised machine learning to infer complex relationships that distinguish correct from faulty behaviors. The relationships learned from normal behavior data can automatically detect incorrect behavior in the same system or variants thereof. We evaluate the method on a voltage regulator, demonstrating its effectiveness in detecting undesired system behaviors.

0.27

Masked Erosion: How Acceptance Gates Lose Redundant Safeguards in Self-Improving Agents

cs.AI ·2026-10-28 ·Hau-Sing So, Daixian Liu, Weicai Long
Abstract · 25 pages (8 pages main text), 10 figures, 35 tables

Self-improving LLM agents edit their own harness, the prompts and rules around a fixed model, and keep an edit only if an acceptance gate scores it higher on a batch of tasks. Such a gate sees what the harness does under the current model, not what it would still prevent if one safeguard failed. When safeguards overlap, deleting one is masked: the others still block the harm, so the gate decides on task success alone and redundancy erodes unseen. We measure harm for every subset of four written safeguards in two agent environments and run 411 improvement loops under ten gate rules, mostly as controlled deletion experiments, plus three runs of the GEPA optimizer. A deletion behind another working safeguard adds almost no observable harm. A paired significance test blocks deletions that pass by chance but not deletions that genuinely raise task success, and a larger evaluation batch lets it delete cheaper safeguards: at the same cost, 20-task batches kept every harness protected while 40-task batches lost all six. Large harm audits prevent exposure but not the loss of depth, which turns into harm when the harness is reused with a weaker model. GEPA kept every safety rule verbatim, yet two of its three final harnesses misused credentials under a weaker model. A statistical model frozen before the runs predicts where each gate's protection breaks as safeguard cost and batch size change, and freezing safety text or auditing one-layer ablations preserves depth.

0.27

El Agente Quntur: Agentic Active-Space Selection for CASSCF

cs.MA physics.chem-ph ·2026-10-28 ·Juan B. Pérez-Sánchez, Viviana Roman-Ventura, Olalla Nieto-Faza, Aiwei Yin, Alán Aspuru-Guzik +1 more
unusual
Abstract

Multireference calculations require time-consuming expert decisions about active spaces, state identity, and convergence. This study asks whether agents can make these decisions reliably. Using the El Agente Quntur multi-agent harness with GPT-5.6 Sol and GPT-5.6 Terra as its language models, agents study 32 fixed-geometry singlet excitations, detect missing active orbitals, compare excited states, and revise state assignments as calculations progress. Prompts specify equal-weight two-singlet state-averaged complete active space self-consistent field (CASSCF) calculations followed by strongly contracted second-order N-electron valence state perturbation theory (NEVPT2) with def2-TZVPD; agents choose and refine the active spaces. Excitation energies are evaluated against published complete active space second-order perturbation theory (CASPT2) reference values at matching geometries. Agentic execution reduces mean absolute error by 48\% relative to six deterministic Active Space Finder protocols. Computational traces vary substantially across runs in selection strategies and cost, even for the same molecule. These results underscore the advantage of agentic active space selection over fixed workflows, yet leave room to improve efficiency through runtime-aware decision-making, reduced inter-agent communication overhead and redundancy, more systematic checks of state identity, and clearer stopping criteria.

0.27

Exponential-Nemytskii Neural Operators: Learning Field Evolution Across Step Sizes

cs.LG math.NA ·2026-10-28 ·Jingye Li, Jinglai Li
theory
Abstract · 22 pages, 1 figure

In variable-step operator learning, agreement at observed step sizes does not by itself determine predictions at other temporal resolutions. We propose an operator whose step-size dependence is prescribed by a first-order Duhamel approximation of semilinear dynamics. It combines exact linear propagation with a local nonlinear function of the field and its derivatives, allowing evaluation below the observation interval without supervision densely sampled across step sizes. Under suitable conditions and at fixed spatial resolution, we prove that agreement between models at two step sizes controls their prediction differences throughout a prescribed range. We also give conditions for local identifiability of both components from two step sizes, supporting interpretation of the learned structure. A latent extension applies the same construction to learned fields. We evaluate the base model on two synthetic problems and the extension on two real datasets against several baselines.

0.26

An Alignment Benchmark for Enterprise Agents on Era by Eon

cs.AI cs.SE ·2026-10-28 ·Hai Sadon, Benjamin Gruenbaum, Doron Porat, Assaf Natanzon, Roy Zavida +2 more
benchmark
Abstract · 11 pages, 1 figure, 4 tables

AI agents are widely used in enterprises to access business systems and act on them in pursuit of a goal. However, fulfilling a user's request can cause harm. Era by Eon is a product that creates a digital twin of a company by simulating its business systems and data. Its simulators implement the programming interfaces of real products. Our benchmark uses ten of these systems, including Salesforce, Zendesk, Jira and Slack. We tested whether agents would carry out harmful business requests or trust misleading information in company records. We ran 19 models on 41 tasks three times, for 2,337 runs. Of the 2,304 runs that produced an answer, 918 (40%) contained a violation. The most common were false claims about actions or records (290 runs), following planted instructions (266) and disclosing sensitive data to the wrong audience (249). Every attempted change was recorded and blocked before it took effect. Misleading Slack messages caused 16 models to report an incorrect ticket status in 47 of 57 runs. No model was free of violations. The four best models violated in 22 to 24% of their runs. Yet one of them, Nova Pro, did the task correctly in only 30 runs.

0.26

Game-Theoretic Foundations for Undiscounted Bisimulation Metrics

cs.AI cs.GTcs.LGmath.OC ·2026-10-28 ·Matthew Zurek, Yudong Chen, Josiah P. Hanna
theory
Abstract

Bisimulation metrics offer a principled notion of task-relevant state similarity in reinforcement learning, but existing constructions and methods are largely confined to discounted MDPs, where contraction arguments ensure well-defined fixed points and downstream guarantees. We develop a unified foundation for bisimulation metrics in undiscounted settings, that is total- and average-reward problems, by replacing the standard operator-centric view with a game-centric one: we interpret prior metric operators as Bellman-Shapley operators of zero-sum turn-based games and define metrics as the corresponding game value functions, yielding unique definitions even when $γ=1$. This game-theoretic foundation facilitates transfer of dynamic-programming theory and algorithms. We also use this game-centric perspective to characterize how metric finiteness depends on structural properties of the underlying MDP and motivate certain directed metrics that remain finite in weakly communicating MDPs. We further introduce an abstract framework for modifying the games' action spaces, allowing general transition coupling sets, that recovers many existing metrics and guides new constructions. Finally, we prove that these metrics bound total-reward and, when the optimal gain is state-independent, optimal relative value differences. We also provide error and suboptimality guarantees for aggregation-based planning in total- and average-reward settings.

0.26

Information-Theoretic Reliability--Safety Bounds and Finite-Alignment Dynamics in AI Knowledge Distillation

cs.AI cond-mat.stat-mech ·2026-10-28 ·Miao-Miao Yi, Zhen-Yu Cai, Y -M. Du, C. P. Sun
Abstract

The reliability and safety of artificial intelligence (AI), together with their transfer and evolution through knowledge distillation, have attracted growing attention. We establish an information-theoretic bound on the reliability--safety trade-off and characterize the dynamics of this trade-off under knowledge distillation. We first prove a universal trade-off inequality that is independent of the specific model form. Its boundary is determined solely by a single parameter Kh, which quantifies the hidden layer's ability to distinguish benign from hazardous inputs. We then develop a canonical Boltzmann-machine model of AI response behavior and derive two exact relations between reliability and safety. The trade-off curve is determined by the hazard discrimination capacity K, while the overall answer tendency controls the operating point along the curve. These relations agree well with existing experimental data. In knowledge distillation, we find that teacher responses induce a renormalization of the student model's trade-off parameters. We further use a renormalization-group approach to study multigenerational distillation, deriving exact recursion relations and the associated renormalization-group-like flows for two representative direct intergenerational distillation protocols at arbitrary teacher--student alignment strength. These flows reveal fixed-point structures, long-term dynamics, and critical behavior, yielding testable predictions for the critical transition between capability loss and stable transmission.

0.26

SatOps-Decision: Low-Latency Discriminative Decisions for Onboard Satellite Communication Operations

cs.LG cs.AIcs.NI ·2026-10-28 ·Lifan Mei, Yuxuan Li, Jiaxian Tu, Shenghe Xu
benchmark
Abstract

We propose SatOps-Decision, a method for low-latency decisions in satellite communication operations. To the best of our knowledge, it is the first application of Jev-style general-purpose discriminative decision models to satellite operations, and the accompanying benchmark is the first dedicated to this approach. The learned scorer, a locally adapted Laya encoder with its native decision head, selects among supplied operational answers in a single forward pass. The benchmark combines authored operational contracts, controlled state changes and independently checked labels in an auditable construction. Evaluation prioritizes complete-request latency and raw decision accuracy; all main accuracy comparisons use the same 1,024 held-out questions, separate from training and development. SatOps-Decision's local scorer has a median latency of 90.3 ms per decision, versus 493.9 ms for Qwen3.5-0.8B generation and 388.4 ms for OneJev1.6-0.8B: $5.47\times$ and $4.30\times$ speedups based on median request times in the measured FP32/MPS implementations. Its mean raw accuracy is 85.3%, versus 42.5% for frozen Laya, 40.7% for Qwen generation and 58.6% for OneJev. As cloud accuracy references, original Jev 1.14/1.13 reach 87.4/81.9%.

0.26

Structural-Test Geometry as an Inductive Bias for Partially Observable World Models

cs.LG ·2026-10-28 ·Todd Y. Zhou
Abstract

Predictive world-model objectives need not organize latent histories according to distinctions required for later control. With privileged simulator access during training, should response-derived structural information be predicted, decoded, or imposed directly on latent geometry--and how do these choices compare with structured belief distillation? We define posterior-marginalized structural tests whose response distributions induce a pseudometric over histories, then train quotient-preserving (QP) world models by aligning shared latent Euclidean distances to that metric. Across a fixed six-task diagnostic suite, the largest differences appear under held-out structural variation: QP improves planning success by 26.1 points $[21.5,30.7]$ over metric decoding and 6.4 points $[3.8,9.0]$ over structured belief distillation. On GridWorld, metric decoding accurately recovers held-out structural distances ($ρ=.91$) while its native latent geometry remains more distorted than QP's (paired IQD difference $.22$ $[.20,.24]$), showing that relational recoverability and direct geometric alignment can diverge.

0.25

Pinpointing Pivot Points: The DECIDE Framework for Data Analysis Decision Importances

cs.LG stat.MLstat.ME ·2026-10-28 ·Zachary T. Rewolinski, Jan Simson, Frauke Kreuter, Christoph Kern, Bin Yu
Abstract

Human judgment calls made throughout the data science life cycle can change the conclusions drawn from data. Independent analysts often make different choices when given the same data, leading to often unacceptable variations in results and contributing to the ongoing reproducibility crisis across scientific disciplines. The rapid adoption of agentic tools threatens to amplify this crisis. Several approaches have been developed to address this variation by following the same strategy: repeating an analysis under different combinations of analytical decisions and inspecting the distribution of outcomes. Examples include multiverse analysis, specification curve analysis, and vibration of effects. More broadly, the Predictability-Computability-Stability (PCS) framework assesses how documented judgment calls across the data science life cycle affect conclusions, with predictability serving as a reality-check that a model captures signal. Each of these approaches results in a "multiverse" object. While the distribution of outcomes reflects the overall variability of conclusions, estimating the influence of each individual decision requires additional tools. A common heuristic is to train a model to predict the outcome from these decisions and treat its coefficients or feature importances as proxies for each decision's influence. Although this strategy may work in certain cases, it assumes the model is correctly specified and accurately predicts the outcome, which we show is not always the case. In this work, we introduce DECIDE, a model-free framework that quantifies each decision's importance from the actual outcomes across the plausible decision space, avoiding issues of predictability by design. We demonstrate the effectiveness of DECIDE by applying it to real-world multiverse analyses. We further provide what is, to our knowledge, the first formal definition of the multiverse.

0.25

Data Amount Heterogeneity Improves Performance in Federated Learning for Neuroimaging

cs.LG cs.CVq-bio.NC ·2026-10-28 ·Paul Kurian, Tamoghna Chattopadhyay, Abhijith Shaji, Greg Ver Steeg, Paul M. Thompson +1 more
Abstract · SIPAIM 2026

Federated Learning (FL) allows for secure analysis of distributed data without data sharing. We and others have successfully applied FL to neuroimaging tasks, such as Alzheimer's disease classification from structural magnetic resonance imaging (MRI). FL trains a deep learning model locally at each site, encrypts the local model parameters and sends them to a federation controller that aggregates the local models under homomorphic encryption, the resulting community model is sent back to the sites for further training, and the process repeats for a number of rounds. FL often achieves learning performance (e.g., accuracy) close to centralized training, but no data leaves the sites and model parameters are protected through encryption and other measures. In this paper, we show that federations composed of sites with heterogeneous amounts of data (1 large site plus small sites) perform better than homogeneous (same data per site) federations, particularly in low data regimes, while reaching performance close to centralized training. Having a big site often results in faster convergence and higher final accuracy. Heterogeneous federations better exploit sites with small amounts of data, a common situation in large consortia such as ENIGMA.

0.25

Variational Consequence-Driven Offline Reinforcement Learning

cs.LG ·2026-10-28 ·Wen Jiang, Ke Jiang, Jin Wang, Xiaoyang Tan
top venue
Abstract · accepted by NeurIPS 2026

Offline reinforcement learning (RL) typically mitigates distribution shift by imposing divergence constraints on the action distributions of the learned and behavior policies. However, this paradigm cannot ensure the underlying dynamic consistency structure: even minor action deviations can lead to drastically different transitions, inadvertently driving the agent away from the offline dataset. To address this, we explore policy constraints through their induced state-consequence distributions rather than relying on pure action-distribution matching. Nevertheless, directly enforcing such consequence-aware constraints is computationally intractable without access to the true underlying dynamics model. To bypass this issue, we propose Structural Behavior Regularization (StBR), which replaces the intractable objective with a closed-form latent divergence. Structuring this latent space to reflect transition dynamics renders the latent divergence a theoretically bounded surrogate for the true consequence-distribution divergence. By regularizing the learned policy with this latent divergence, StBR adaptively tightens the behavior constraint in dynamics-sensitive regions, effectively preventing deviations outside the offline dataset. Empirically, we demonstrate that StBR achieves superior performance across diverse offline RL benchmarks.

0.25

FACTS-MAS: A Multi-Agent Architecture for Adaptive Forecasting from Heterogeneous Evidence

cs.AI stat.AP ·2026-10-28 ·Madumita Karthikeyan, Umang Rajnikant Bid, Rong Pan
code/data
Abstract · 29

Real-world forecasting often requires integrating heterogeneous evidence whose predictive value varies across horizons, operating regimes, and domains. We introduce FACTS-MAS, a domain-agnostic multi-agent forecasting framework that decomposes evidence into specialized channels for temporal dynamics, exogenous context, seasonality, structural information, and discrete events. Each agent generates an independently testable forecast through a common interface, while leakage-safe historical validation measures its predictive skill relative to a naive reference. These skill estimates are used to weight and combine agent forecasts, so that uninformative evidence is either removed or downweighted rather than forced to participate. A calibration procedure selects domain-specific constant means based on historical validation folds. We evaluate FACTS-MAS on weekly for-sale housing inventory from Zillow and weekly unemployment insurance claims from FRED, for 15 U.S. metropolitan areas at 4-, 8-, and 13-week forecasting horizons. Our method outperforms both a naive last-value forecast method and the TimesFM foundation model. However, a factor-adjusted exponential smoothing model performs as well as the fused forecast. The evidence therefore supports FACTS-MAS as a systematic protocol for testing, calibrating, and recomposing evidence channels before assembling them into the strongest single model for a specific application domain. Code, data, and results are available at https://github.com/Madumita24/FACTS-MAS.

0.25

TGTV: An Interpretable Framework for UAV Fault Detection and Traceability via Spatiotemporal Graph Learning and Large Language Models

cs.LG ·2026-10-28 ·Zixuan Wang, Changbo Shi, Ji Liu
Abstract · 21 pages, 8 figures, 4 tables. Submitted to IEEE Sensors Journal

Fixed-wing unmanned aerial vehicles experience operating-condition transitions, asynchronous sensor sampling, and control-system compensation that can obscure fault signatures. Existing temporal detection methods have limited capacity to represent coupled faults across subsystems, track their evolution, and constrain language-model diagnoses with quantitative evidence. This paper proposes TGTV, an interpretable fault detection and traceability framework that integrates spatiotemporal graph learning with a large language model. A temporal convolutional network and a graph attention network with weighted fused adjacency extract multiscale temporal patterns and spatial dependencies among sensors. A hybrid temporal Transformer and variational autoencoder then evaluates window-level anomalies using prediction errors and reconstruction probabilities. Continuous-time Shapley attribution identifies critical sensors and high-contribution intervals while preserving the temporal continuity of fault evolution. A structured evidence package combines detection results, graph relationships, temporal attribution, sensor attribution, and operating context. An evidence-constrained diagnosis protocol uses this package to generate flight-level conclusions with traceable evidence identifiers. Cross-flight holdout experiments on the public ALFA fixed-wing flight dataset yield a window-level area under the precision-recall curve of 0.976, accuracy of 0.973, recall of 0.902, and F1 score of 0.920. Flight-level fault diagnosis achieves recall of 0.943 and an F1 score of 0.944. These results demonstrate accurate anomaly detection and quantitatively supported fault-tracing cues, providing a basis for interpretable post-flight fault review and analysis.

0.25

Too Hard, Too Easy, Just Right: Performance-Driven Challenge Adaptation in Virtual Reality Firefighter Training

cs.AI ·2026-10-28 ·Xuan Wang, Jianan Jiang, Rongqian Chen, Tian Lan, Bin Li
Abstract

Effective firefighter training requires practice under dynamic and stressful emergency conditions, yet real-world live-fire exercises are costly, risky, and difficult to scale. Virtual Reality (VR) offers a safe and controllable alternative, but many existing VR firefighter training systems rely on fixed scenarios or provide limited difficulty adaptation, making it difficult to maintain an appropriate level of challenge for skill development as trainees' performance changes over time. To address this limitation, we present ProACT, a performance-driven adaptive training system that regulates scenario difficulty through closed-loop interactions between trainee performance and simulated fire dynamics. Specifically, ProACT combines a global module that continuously adjusts scene-level fire spread dynamics in real time based on trainee performance with a local module that periodically updates object-level flammability across the environment. Our evaluation shows that ProACT improves both objective training performance and subjective user experience compared with existing adaptive and non-adaptive baselines. We further discuss how adaptive immersive environments can extend beyond firefighting to broader training domains.

0.25

Modular Bayesian Foundation Models: Bayesian time series encoders for OOD detection, routing and forecasting

cs.LG stat.MLstat.APstat.CO ·2026-10-28 ·Mame Diarra Toure, Quang Hung Pham, Mohamed Ryad Zemouri, David Stephens, Luc Vouligny +1 more
dataset
Abstract · 44 pages, 15 figures

Industrial systems continuously generate time series that vary in sampling frequency, scale, dynamics, and operating regime. Consequently a single time-series foundation model may not be able to represent all regimes effectively and retraining a large model for every new dataset or regime shift is computationally expensive. Modular foundation models address this issue by reusing specialised encoders trained on diverse datasets and by allowing the expansion of capabilities through the addition of new experts. This approach requires compact experts to minimize storage cost and a proper selection mechanism of experts for unseen data. We propose a modular foundation model framework that combines end-to-end Bayesian encoder training with uncertainty based routing. Using Singular Bayesian Neural Networks (SBNNs), we build a collection of compact Bayesian encoders whose epistemic uncertainty guides label-free selection. At the selected ranks, our Bayesian collection uses 63.8\% less storage than its deterministic counterpart. We evaluate this framework on time-series data through distribution shift, encoder selection, and probabilistic forecasting. Across seven incoming datasets, uncertainty-based routing is competitive with distance-based routing in terms of forecast MSE, while reducing average ECE from 0.086 to 0.072 and average NLL from 0.907 to 0.836. It also increases PICP90 from 82.28\% to 89.03\%. In a separate model-family comparison, we evaluate the SBNN pipeline against full-rank Bayes by Backprop, deep ensembles, and MC Dropout. The SBNN pipeline achieves an average MSE of 0.390, comparable with 0.396 for deep ensembles, 0.389 for MC dropout, and 0.443 for the full-rank Bayesian model. It uses 63.9\% fewer parameters than MC dropout, 81.9\% fewer than the full-rank model, and 92.8\% fewer than the deep ensemble.

0.24

Evaluating Causal Reasoning in LLMs: A Case Study in the Challenges of Procedural Generation

cs.AI cs.LG ·2026-10-28 ·Justin Clarke, Isha Karn, Patrick Murray, KJ Hill, David Jensen
benchmarkcode/data
Abstract

Large language models (LLMs) appear to represent and reason about a large range of phenomena. In particular, recent studies have examined whether LLMs can accurately reason about causation. One of the most commonly used benchmarks for evaluating formal causal reasoning in LLMs is CLadder, a framework for procedural generation of benchmark data sets that was released in 2023. Experiments with CLadder appear to indicate that LLMs can perform formal causal reasoning in a large percentage of generated examples. In this paper, we examine what can be reliably inferred when LLMs perform well on test sets produced by CLadder. We identify several examples of superficial and highly predictive patterns that the CLadder generator embeds in data sets it generates. We find that even simple models such as random forests perform well on CLadder-generated data sets after training on a sample of generated data. We also find that the performance of many commercial LLMs on CLadder test sets is extremely brittle and that removing large sections of the text that are critical for formal causal reasoning has almost no effect on performance. Given these results, the most likely hypothesis for the strong performance of state-of-the-art LLMs on CLadder is that they are exploiting superficial patterns learned from exposure to training data from the generator. Any inference that LLMs are exhibiting more complex causal reasoning must rule out this hypothesis.

0.24

CFKS: Continuum-Feature Kernel PDE Solvers

cs.LG ·2026-10-28 ·Vikas Dwivedi, Balaji Srinivasan, Monica Sigovan, Bruno Sixou
code/data
Abstract

Neural-network-based PDE solvers require architectural choices, particularly the number of layers and neurons per layer. Fixing a single Gaussian radial basis function (RBF) feature layer and taking the number of features to the continuum limit removes the need to prescribe a finite feature count or select individual feature centers, thereby reducing representation arbitrariness. Motivated by this principle, we introduce Continuum-Feature Kernel Solver (CFKS), a deterministic framework for PDEs obtained by analytically marginalizing a bounded continuum of Gaussian RBF features. CFKS works directly with the resulting continuum-feature mean kernel and its operator-applied counterparts, parameterized by prescribed Gaussian widths and bounded center distributions. For linear PDEs, CFKS leads to symmetric operator-kernel collocation systems with null-space enforcement of physical constraints. Nonlinear PDEs are addressed through linearized initialization and nonlinear least-squares residual minimization using analytical kernel derivatives and Jacobians. Residual-guided collocation enrichment and sequential space--time-slab marching provide additional resolution strategies, while prescribed multiscale kernel refinement helps resolve localized nonlinear structures. Finite-feature experiments confirm convergence of empirical value and operator Gram matrices toward their continuum counterparts. Numerical studies include oscillatory and multiscale ODEs, irregular-domain Poisson and biharmonic problems, singular perturbations, transient transport equations, and viscous Burgers' equation. The results establish CFKS as a direct computational connection between finite Gaussian-RBF representations and continuum-feature kernel PDE solvers. Codes: https://github.com/vikas-dwivedi-2022/CFKS_Codes

0.24

Progressive Coupling Method for Diffusion Posterior Sampling in Ptychography

cs.LG ·2026-10-28 ·Yihan Qiu, Mingtao Huang, Tianyuan Li, Shudong Li, Yuan Shen
Abstract

Ptychography is a computational imaging method that reconstructs a sample from overlapping diffraction measurements. This technique relies on these overlapping measurements to provide the redundancy information needed for robust reconstruction, but dense scanning increases acquisition time and data volume. Reducing scan overlaps can improve measurement efficiency but weakens the physical constraints available for reconstruction. Diffusion models provide strong learned prior information, offering a promising approach to compensate for the loss of measurement redundancy. However, effectively utilizing this prior information under weak measurement constraints remains a challenge. We propose a progressive coupling posterior sampling (PCPS) method, which allows for a broader exploration of the solution space under high uncertainty while maintaining stronger diffusion-guided consistency as the reconstruction becomes more refined. Experiments on real X-ray ptychographic data at various scan overlaps demonstrate that PCPS consistently improves reconstruction accuracy compared to traditional iterative methods and existing diffusion posterior samplers, especially at low overlaps. These results show that diffusion priors can reduce the measurement redundancy required for accurate ptychography reconstruction.

0.24

FPGAFly: A 166.7k-Neuron Fruit-Fly Connectome on One FPGA via Deferred Spike Expansion and Shared-Codeword States

cs.NE cs.AR ·2026-10-28 ·YE Ziyang
Abstract

Synapse-resolution fruit-fly (Drosophila) connectomes enable neuron-by-neuron simulation of entire insect nervous systems, yet whole-brain models still require computationally demanding software or multiple neuromorphic chips, and single FPGAs support only small or statistical networks. Scaling to the full connectome, per-synapse spike expansion scales queue traffic with fan-out up to 11.2k, while 41-bit neuron states require 6.83 Mbit, exceeding the 6.27 Mbit on-chip block static RAM (BSRAM) of the target device. FPGAFly runs the 166.7k-neuron, 25.6M-connection leaky integrate-and-fire male central nervous system (CNS) connectome on one FPGA. Order-preserving deferred expansion queues one 64-bit descriptor per spike in source-ID order, cutting external memory traffic by 54.5%-57.9% and peak queue data by 95.6%-96.5% with bit-identical results. A 24-bit encoding shares codewords across normal, refractory, and quantized out-of-range states, reducing BSRAM usage from 438 to 270 with 1245 extra LUTs and zero changes across 10,000 MNIST predictions. At 80 MHz on a Gowin GW5AST-LV138, each 100 ms window finishes in 86.83 ms on average and within 93.05 ms across 2000 tests, realizing the first complete Drosophila CNS connectome simulation on one FPGA.

0.23

Logical Guidance of Stochastic Interpolants

cs.LG ·2026-10-28 ·Gerrit Gerhartz, Francesco Alesiani, Jonathan Warrell, Henrik Christiansen, Matheus Ferraz +1 more
Abstract

Guidance and composition of generative models have applications in visual design, chemistry, and biology. Logical guidance composes conditional scores exactly for diffusion models. However, Flow Matching (FM) and Stochastic Interpolants (SI) learn velocity fields. We show that negation and exclusive disjunction admit exact conditional velocity composition on any interpolant, whereas conjunction, and therefore any formula containing one, is exact on Gaussian paths but in general incurs a drift-interaction error, so composed velocities need not sample the logical target. LOGSI solves this by using score-based re-equilibration as our principled mechanism and falling back to the Gaussian-path velocity rule as an approximate transport predictor. In non-Gaussian bridge problems, generation alternates transport steps with score re-equilibration steps. On image trans fer (EMNIST to CMNIST, CelebA) and QM9 source-to-target molecular editing with logical attribute constraints, LOGSI improves rule conformity at the same network-evaluation budget over composed-velocity baselines, while also improving fidelity on the image tasks. More broadly, LOGSI applies wherever stochastic interpolants need guidance and logical combinations of constraints are natural.

0.21

The Role of Quantization in Loss Spikes

cs.LG cs.AI ·2026-10-28 ·Zaiqian Chen, Aditya Somasundaram, Sophie Li, Sean McLeish, John Kirchenbauer +2 more
Abstract

Loss spikes in language model pretraining can waste substantial compute and derail otherwise healthy training runs. A primary driver of these instabilities is the growth and divergence of output logits in the final language modeling (LM) head. In this work, we study a particularly costly form of this instability associated with output logit growth: delayed, terminal divergences that emerge only after prolonged training, which we call time-bomb spikes. We find that these instabilities are preceded by rapidly growing quantization error in the LM head, which eventually culminates in a loss spike. Crucially, reducing numerical precision accelerates time-bomb spikes, while increasing precision improves stability. We then propose a simple yet effective mitigation that replaces AdamW on the LM head with SGD with momentum, suppressing logit growth and improving stability over long training horizons. Finally, we find that existing mitigation techniques from the literature (along with our own) all reduce quantization error, suggesting a shared numerical mechanism behind their effectiveness. Together, our results identify finite precision error as a hidden driver of time-bomb spikes and show that suppressing this error is key to stable long-horizon pretraining.

0.21

Continual Adaptation Across Modalities: Learning and Forgetting in Unified Multimodal Models

cs.LG ·2026-10-28 ·Xinrui Wang, Zhihao Chen, Jia Tang, Alexandra Gomez-Villa, Bartlomiej Twardowski +2 more
published
Abstract

Unified multimodal models (UMMs), which can process and generate diverse modalities within a single framework, have recently emerged as a powerful paradigm. However, when these models undergo continual learning on multiple tasks across different modalities, their learning and forgetting dynamics remain poorly understood. In this paper, we systematically investigate how architectural design and adaptation method shape intra-modal learning and cross-modal interference between the generation and understanding sides. Specifically, we report three findings. First, the severity of cross-modal interference decreases with the degree of modality-specific parameter isolation. Second, a within-model asymmetry: the generation side of a UMM is markedly more fragile than its understanding side. Third, what post-training adds is the first to be forgotten: purely post-training capabilities such as personalized generation collapse by 84\% to 90\%, while pretraining capabilities remain stable. Together, these results clarify how parameter isolation and post-training fragility shape UMMs' learning and forgetting dynamics, offering a starting point for developing continually evolving UMMs.

0.21

Floor It, Then Fill It: Simulating Creator Ecosystems for Fair Video-Content Recommendation Systems at Billion User Scale

cs.LG ·2026-10-28 ·Sohini Roychowdhury, Long Sha, Sunand Agarwal, Payal Kamboj, Shuyang Chen +3 more
Abstract · 9 pages, 5 figures, Neurips Workshop 2026

Engagement-optimized content recommendation systems exhibit a well-documented rich-get-richer dynamic that structurally disadvantages emerging creators on large-scale platforms. Evaluating debiasing interventions at scale demands faithful simulation of creator-viewer ecosystems, yet the fidelity, diversity, and validity of such simulators remain underexplored. We present a grounded user-simulation framework that models a heterogeneous population of ${\sim}42$K creators and their coupled viewer interactions over 12~weeks, and use it to evaluate equitable view-allocation policies formulated as a budgeted multi-armed bandit (MAB). The simulator captures three behavioral dimensions: (i)~\textbf{creator diversity}, modeled via a clustering approach grounded in real platform creator-level distributions; (ii)~\textbf{behavioral fidelity}, captured through coupled creator-content state machines that represent growth trajectories, engagement accumulation, and graduation dynamics; and (iii)~\textbf{population-level validity}, measured by calibration against observed platform metrics. We evaluate simulative-configurations that vary content-viewership target definitions, reward functions, and a small-creator reserve lever to maximize creator-content recommendation equity. A floor-protected per content-view-target with a view-parity-gap reward guarantees 90\% of view targets being met by creators every week, grows small-creator organic reach ${\sim}{+}33\%$ per week, and achieves high distribution stability (set of creators with funded-views Jaccard $= 0.819$). We analyze the Sim2Real gap and propose validation protocols, contributing evidence for when simulation-derived fairness conclusions can be trusted.

0.21

Lev-3B: A Data Recipe for Compact Probabilistic Log Decision Models

cs.LG cs.AI ·2026-10-28 ·Leo Betthauser, Yingqi Liu, Uday Shah, Jishang Wei, Sainyam Galhotra +3 more
Abstract

Many log-analysis tasks involve choosing among known options, such as detecting anomalous sessions or identifying attack techniques, and benefit from calibrated probabilities. We present Lev-3B, an open 3B-parameter decision model built on Granite-4.1-3B-Base with a shared pointer readout, and study its training recipe. The recipe has two key parts. First, training follows a curriculum: the model learns bounded natural-language decisions, using no log data, before it sees any logs. Second, the log stage mixes the target tasks with general log data spanning 35 auxiliary tasks unrelated to either evaluation task. We evaluate one shared checkpoint on sources seen in training and on sources excluded from training, calibration and model selection. The two-stage recipe transfers substantially better than training on log data alone. Adding general log data improves performance on excluded sources. At a fraction of their serving cost, Lev-3B outperforms prompted frontier models on anomaly detection, including on an excluded logging system, but trails them on attack-technique classification. Its probabilities are well calibrated on familiar sources, but on the excluded logging system it confidently misses anomalies. We release the checkpoint, recipe, data manifests, and prediction-level artifacts.

0.21

From Natural Language to Real Schematics: An Automated Workflow for Power Electronics Schematic Synthesis

cs.AI ·2026-10-28 ·Bowen Su, Huan Chen, Jiahua Ying, Ran Ou, Weilong Chen +2 more
Abstract

Designing power electronics converters traditionally requires intensive human expertise across topology sizing, commercial component selection, circuit simulation verification, and schematic drafting. While Large Language Models (LLMs) show promise in electronic design automation (EDA), applying standalone LLMs directly to board-level power electronics schematic synthesis suffers from severe engineering hallucinations, including inaccurate analytical sizing, fabricated or mismatched component part numbers, violated voltage derating margins, and broken topological connectivity. In this paper, we propose a Bounded Dual-Loop Automated Schematic Synthesis Framework that translates natural language specifications into verified, editable KiCad schematics and manufacturing deliverables. Our workflow integrates six stages across three functional columns: (1) requirement parsing and knowledge-driven power stage synthesis covering non-isolated (Buck, Boost) and isolated/resonant (Flyback, DAB, half-bridge LLC) topologies; (2) contract- and evidence-coordinated component selection grounded in distributor APIs and OEM datasheets, coupled with closed-loop ngspice transient verification and stress feedback; and (3) modular control hardware integration and multi-format deliverable generation featuring native KiCad netlist round-trip verification. End-to-end case studies on a 24-36V to 12V,2A Buck converter (29 positions, 83 verified terminals) and an isolated 48V to 24V, 0.5A, Dual-Active Bridge (DAB) converter (87 positions, 341 verified terminals), alongside comparative evaluations against a Vanilla LLM baseline, demonstrate that our framework improves component verification coverage from 22.2 to 98.6, nominal value consistency from 0 to 97.7, and SPICE simulation pass rates from 0 to 100.

0.20

Leaving, Staying, or Living There? Duration Commonsense Reasoning with Human-LLM Co-annotations

cs.AI cs.IR ·2026-10-28 ·Chenyang Yu, Zhaomin Xiao, Yan Huang, Eduardo Blanco
Abstract

Gardening in the backyard implies living there, while checking into an Airbnb more likely means staying for a few days. Such duration judgments are important for understanding human mobility, travel behavior, and social context. However, duration is rarely stated explicitly in social media posts and must instead be inferred from noisy textual, visual, and contextual cues. In this paper, we investigate whether vision-language models possess such duration commonsense. We create a new corpus of geotagged tweets co-annotated by humans and LLMs to serve as a testbed for this task. We find that state-of-the-art models struggle with duration inference, even when guided by multi-turn Chain-of-Thought prompting. Error analyses further reveal that models often fail to integrate critical contextual information, or they overly rely on prominent visual details while overlooking subtle hashtags and semantic nuances. These findings underscore the significant challenges remaining in context-aware multimodal understanding.

0.20

A Hybrid Physics-informed holomorphic neural networks Method

cs.LG math.NA ·2026-10-28 ·Zhijie Su, Qiaolin He, Zhiping Mao, Shiquan Zhang
Abstract · 34 pages, 14 figures

Physics-informed holomorphic neural networks (PIHNNs) exploit complex-analytic representations to satisfy homogeneous PDEs by construction, but their extension to nonhomogeneous problems is nontrivial. We propose a hybrid physics-informed holomorphic neural network (HPIHNN) framework based on a decomposition of the solution into particular and homogeneous components. The particular component is computed by a spectral method on a simple extended domain, while the homogeneous component is approximated by a PIHNN trained only on the boundary with modified boundary data. The resulting formulation avoids meshing complex geometries and retains the dimensionality-reduction advantage of holomorphic representations. The method is applied to nonhomogeneous Poisson, Stokes, biharmonic, linear elasticity, and Helmholtz equations. Numerical experiments covering complex geometries, corner singularities, and nearly incompressible regimes demonstrate that the proposed method accurately resolves the corresponding solutions and, in the considered tests, generally achieves lower errors and shorter training times than standard PINNs. The results indicate that coupling spectral particular solutions with holomorphic neural approximations provides an efficient approach to nonhomogeneous PDEs on complex domains.

0.20

DES-JEPA: Learning a Shared Vocabulary for High-Dimensional Discrete Event Sequences

cs.AI ·2026-10-28 ·Hugo Math, Maximilian Forstenhäusler, Christos Anagnostopoulos
published
Abstract · 5 pages

Temporal foundation models such as Chronos and TimesFM generalize across domains through a shared numerical observation space. In contrast, discrete event sequences (DESs), including operational event logs or patient clinical-event trajectories, have domain-specific event vocabularies whose labels lack a consistent correspondence across domains. Beyond this lack of consistency, the number of distinct events in real-world settings is high enough to challenge both classical temporal point process modeling and current in-context foundation inference models such as FIM-PP. Critically, current foundation models for DESs are limited to a small, fixed number of event types, restricting their applicability to realistic, high-dimensional event streams. To address this, we propose two complementary modules: (i) DES-JEPA, a joint-embedding predictive architecture that learns a shared discrete vocabulary from synthetic stochastic processes with unaligned event vocabularies; and (ii) DES-PFN, a discrete event sequence prior-fitted network that reuses the learned vocabulary for in-context next-event prediction. Without task-specific fine-tuning, this framework achieves competitive zero-shot performance on eight unseen real-world datasets and scales up to 700 distinct event types.

0.20

The Utility of Physics-Informed Machine Learning in Solar Power Production Forecasting

cs.LG ·2026-10-28 ·Peter Pavlík, Franz M. Rohrhofer, Bernhard C. Geiger
Abstract · Submitted to Solar Energy

Physics-informed machine learning (PIML) and hybrid modeling have attracted significant attention as approaches for combining physical knowledge with the flexibility of deep learning. In this work, we investigate whether these approaches deliver their commonly claimed benefits for large-scale photovoltaic (PV) power forecasting. We develop two hybrid forecasting architectures that integrate the ADR physical PV efficiency model with the TSMixer neural forecasting architecture through feature augmentation and residual learning. The proposed models are evaluated on a dataset comprising over 3 million samples from 3,713 PV installations across 33 European countries. We assess convergence speed, forecasting accuracy, transferability across geographical regions, and data efficiency. Although the physics-informed models seem to exhibit slightly faster convergence during training, they do not achieve consistent improvements in test-set accuracy, generalization, or performance under limited training data compared to the purely data-driven baseline. These findings suggest that, for modern deep forecasting architectures trained on large datasets, explicitly incorporating the ADR physical model provides limited practical benefit for day-ahead PV forecasting.

0.20

SpikeAD: Toward Multiplication-Free Spike-Controlled Neuromorphic Architecture for Energy-Efficient Small Object Detection

cs.NE ·2026-10-28 ·Zhifan Song, Haralampos-G. Stratigopoulos
code/data
Abstract

Energy-constrained aerial platforms require accurate object detection under strict power and latency budgets, yet conventional inference remains bottlenecked by multiply-accumulate (MAC) operations. While spiking neural networks (SNNs) offer bio-inspired efficiency, many SNN detectors retain hybrid MACs in intermediate layers, or use spike-driven inference where multi-level spikes are interpreted as repeated binary events. We propose SpikeAD, a direct-trained spike-controlled neuromorphic aerial detector. In SpikeAD, sparse bounded integer spikes act strictly as multiplexer control signals for multiplier-free shift-add inference, preserving spike boundaries throughout a four-scale bidirectional feature pyramid. We introduce the Spike-Gated Multi-Scale (SGMS) block, performing receptive field expansion with activity-driven channel selection, implementable with adders and comparators. Evaluated on VisDrone and DOTA and across Nano, Small, and Medium variants, SpikeAD achieves up to 46.2% mAP@50 on VisDrone (+4.8% over SNN prior art), and generalizes to oriented bounding box (77.6% mAP@50 on DOTA) using the proposed SpikeOBB head. Under the Horowitz 45nm arithmetic energy model following standard methodology in SNN literature, SpikeAD achieves up to 18.9x estimated reduction relative to an equivalent MAC-based implementation. We propose the Spike Processing Core (SPC), an RTL compute primitive to validate hardware realizability for reconfigurable and scalable custom accelerator chip design on FPGA/ASIC, highlighting a promising path toward next-generation ultra-low-power neuromorphic processors for aerial object detection. Link to the open-source and open-weight GitHub repository will be added upon publication.

0.19

Learning a Prior-Data Fitted Network for Graph Foundation Models via Structural Causal Model Priors

cs.LG ·2026-10-28 ·Ziwei Zhang, Tianyin Liao, Chunyu Hu, Ge Lan, Nuo Xu +5 more
Abstract

Graph Foundation Models (GFMs) have recently emerged as a promising paradigm for learning transferable representations across different graphs and domains. However, existing GFMs face notable limitations, including limited pre-training scale or reliance on aligning with textual information in Large Language Models (LLMs). To overcome these limitations, recent studies have drawn inspiration from prior-data fitted networks (PFNs), which learn from large synthetic datasets and support multi-domain prediction backed by causal learning. However, existing graph PFNs encounter severe difficulties in modeling causal relationships within and across node and edge sets, generating diverse graph structures, and designing suitable architectures that can learn from complex synthetic datasets. To address these challenges, we propose **GSCM-PFN** (Graph SCM-based Prior-Data Fitted Network), a graph PFN pre-trained on synthetic data generated from an SCM-based prior that jointly models nodes and graph topology. Specifically, we (1) design a joint feature-edge-label structural causal model (SCM) to guide synthetic graph generation and capture causal relationships within and across node and edge sets; (2) introduce prior-guided node-edge dependency modeling to instantiate diverse graph generation strategies, and prove that our SCM-based generation prior is more expressive than existing priors; and (3) propose a feature-permutation-invariant dual-path graph network with theoretical guarantees for learning from diverse synthetic graphs. Experiments on 23 real-world benchmarks show that GSCM-PFN outperforms both conventional GFMs and PFN-based GFMs, without utilizing real-world graph data for pre-training. Code: https://anonymous.4open.science/r/GSCM-PFN-A3A3/

0.19

Persona Learning from Feedback for Personalized AI Assistants

cs.LG cs.AIcs.CL ·2026-10-28 ·Yiran Shen, Yu Xia, Liuyi Yao, Prithviraj Ammanabrolu
Abstract

Personalizing AI assistants requires adapting their behavior to users' preferences and task contexts. Existing work elicits assistant personas through persona prompting or finetuning on persona-conditioned generations. However, persona descriptions alone do not identify what to change or preserve in a particular response, or which behaviors suit the current request. We introduce $\textit{Persona Learning from Feedback (PLF)}$, a framework that learns reusable, preference-grounded assistant personas as independent LoRA adapters on a shared frozen backbone. During training, a frozen self-teacher conditions on persona specifications and response-specific feedback to provide token-level supervision on each adapter's own generated prefixes. A feedback gate focuses corrective updates on responses judged to need improvement. At deployment, a supervised router uses the request or dialogue to select a single adapter or a fixed composition, without user-specific finetuning. We evaluate PLF on single-turn personalization with model-generated critiques and multi-turn tutoring with simulated students' reactions. Across both settings, PLF improves model-judged preference fit over fixed-persona policies and strong baselines including persona prompting and persona distillation, demonstrating how feedback-trained behaviors can be reused for adaptive assistant personalization without degrading general capabilities.

0.19

Reinforcement Learning for the Game of Balatro: A Fast Simulator and Two-Policy Training Approach

cs.LG cs.AI ·2026-10-28 ·Ajay Vikram Periasami, Anjaneya Gupta, Ethan Cheng, Ivan Rosales
Abstract · Accepted at EXAG 2026 (AIIDE Workshop on Experimental AI in Games)

Balatro, a 2024 roguelike deckbuilder, presents three difficulties that have historically derailed reinforcement learning methods: a 3,000-action state-dependent action space, horizons of several hundred decisions with sparse rewards, and pervasive stochasticity in both card draws and shop generation. Published RL results on the game remain scarce. We introduce an end-to-end pipeline comprising (i) a high-throughput Rust simulator that is automatically state-diffed against the live game client via BalatroBot, (ii) a condensed observation-action interface with dynamic per-step masking of legal moves, and (iii) a multi-agent PPO architecture that decomposes play into a tactical blind policy and a strategic shop policy that share the full game state but act in disjoint phases. Long-horizon credit assignment is partially addressed by per-agent reward routing with explicit cross-agent credit. On the main joint run, training metrics show a sharp phase transition in which the first-blind clear rate converges above 99% and mean episode return increases by roughly an order of magnitude, plateauing at a mean maximum ante of 3 out of 8; single-agent baselines under identical rewards oscillated between greedy and hoarding failure modes and did not stably clear the first blind. These results establish the two-policy decomposition as a strong reactive baseline for Balatro and localise the remaining gap to long-horizon strategic planning rather than tactical blind play.

0.19

Agent Transfer Protocol: Server-Mediated Agent-to-Agent Communication Architecture for the Agentic Web

cs.AI cs.MA ·2026-10-28 ·Yuqi Qiu, Zuyao Xu, Lu Sun, Xiang Li, Fasheng Miao +3 more
Abstract · 12 pages, 6 figures, 5 tables

The Agentic Web needs to support continuous collaboration among agents from different organizations. These collaborative relationships may span multiple tasks, network connections and execution instances. Such interactions require stable communication entries, message reception when instances are offline, and source verification after delayed reading. How to organize these responsibilities so that cross domain communication does not depend on specific execution instances being continuously online is a key issue for building long term agent interaction infrastructure. We propose ATP (Agent Transfer Protocol), a cross organization agent communication protocol adopting a domain mediated architecture. In this architecture, administrative domains maintain stable addresses and receiving entries for logical agents. The domain servers of both parties continuously undertake cross domain transmission. The receiving domain verifies and saves messages for execution instances to read later. The sending domain signs the complete message envelope. This allows the receiving instance to independently reverify the domain source and content integrity after the original connection ends, provided the original verification public key is still trusted and obtainable. We implemented an ATP prototype and evaluated it through a controlled cross organization LLM agent workflow, mechanism checks and latency measurements. The workflow completed subsequent business processing after instances went offline or were replaced. The independent verifier successfully verified the domain source and content integrity of saved messages. Latency measurements further quantified the communication and processing overhead introduced by the continuous mediation of domain servers.

0.19

BridgeRouter: Gain-Aware Instance Routing over Cross-Modal Bridge Portfolios

cs.AI ·2026-10-28 ·Ruoxi Piao, Alicia Lu, Yunzhen Zhang, Mustafa Mısır, Hasan Onur Keleş
Abstract

Lightweight representation bridges make pretrained, task-adapted EEG foundation models executable on other biosignal modalities without retraining their backbones. However, selecting a single bridge pathway globally assigns every instance to the same transfer route and ignores complementary errors across routes. We therefore formulate multi-bridge transfer as an instance-level algorithm-selection problem and introduce BridgeRouter, a gain-aware router over a portfolio of frozen cross-modal pathways. A validation-selected Single Best Solver (SBS) serves as a protected default route. During training, task labels are used to compute the reduction in negative log-likelihood provided by each alternative route relative to the SBS, and a lightweight MLP predicts these route-specific gains from a target-modality representation extracted by a frozen encoder. At inference, BridgeRouter observes only the target modality, requires neither EEG, task labels, nor predictions from all candidate routes, and executes exactly one bridge pathway: the route with the largest predicted gain when that gain exceeds a validation-calibrated threshold, and the SBS otherwise. We evaluate seven EEG foundation-model bridge pathways for EMG-based freezing-of-gait detection on FOG and ECG-based sleep classification on ISRUC. Across five seeds, BridgeRouter achieves $81.38\pm1.11\%$ balanced accuracy on FOG, improving over the BIOT SBS by $11.49$ percentage points, and $62.92\pm2.17\%$ on ISRUC, improving over the CBraMod SBS by $3.83$ points. Per-instance bridge oracles reach $99.96\%$ and $78.61\%$, respectively, revealing substantial remaining portfolio headroom. These results show that anchor-relative gain prediction can convert complementary cross-modal bridge behavior into deployable instance-level routing improvements while retaining single-route inference.

0.18

Geometric Time Score Estimators: Bridging Self-Predictive Representations and Generative Occupancy Models For Reinforcement Learning

cs.LG ·2026-10-28 ·Aravind Venugopal, Grace Liu, Jeff Schneider, Benjamin Eysenbach
Abstract

In reinforcement learning (RL), temporal reasoning is essential for estimating the density of future outcomes under a policy's occupancy measure. While self-predictive representations have been a dominant paradigm for occupancy density estimation, we find that they yield surprisingly poor estimates. On the other hand, generative occupancy models can produce remarkably accurate samples from the occupancy measure. Yet, they cannot easily recover occupancy densities, limiting their use for control despite their expressivity. In this paper, we propose Geometric Time Score Estimators (GTSE), a framework that bridges the two paradigms by learning occupancy density estimators along probability paths. We analyze how GTSE implicitly induces better occupancy estimators through improved representation learning and show how to directly extract self-predictive representations from this framework. Through experiments across several locomotion and manipulation tasks, we show that GTSE improves performance over state-of-the-art methods by $2\times$ to $6\times$ in goal-conditioned RL settings, and that the representations learned using our framework transfer effectively to downstream RL tasks. Website: https://aravindvenu7.github.io/website/gtse/

0.18

State-Dependent Memristor Dynamics for Device-Informed Spike-Timing-Dependent Plasticity in Spiking Neural Networks

cs.NE ·2026-10-28 ·Samiur Rahman Khan, Saifur Rahman Sabuj
Abstract · 14 pages, 8 figures. Manuscript under review

This work investigates device-informed spike-timing-dependent plasticity (STDP) in spiking neural networks using a literature-derived threshold-memristor model. Local STDP commands determine programming polarity and a programming coordinate for state-dependent conductance updates in the first fully connected layer, while a backpropagation-trained readout performs classification. HSPICE simulations characterize thresholded, polarity- and state-dependent device dynamics and differential crossbar multiply-accumulate behavior. Experiments on MNIST and Fashion-MNIST compare the hybrid learning rule with direct clipped STDP and evaluate programming asymmetry, device-parameter variation, write noise, stuck-at faults, and input corruption. The hybrid model achieved 95.15% accuracy on MNIST and 85.92% on Fashion-MNIST. In a separate seed-matched comparison, it reached 95.02% and 84.79%, versus 92.33% and 83.25% for direct clipped STDP. Potentiation-dominant asymmetry reduced input-dependent feature selectivity, and endpoint faults introduced during training were less damaging than faults introduced after training. The network-level mapping is a reduced-order, validation-selected abstraction rather than a pulse-accurate physical programming implementation. These results quantify how state-dependent memristor dynamics shape local-STDP learning and its sensitivity to hardware nonidealities.

0.18

Discovery of Diverse and Realistic Financial Tail-Risk Using Generative Flow Networks

cs.LG ·2026-10-28 ·Adir Serruya, Gilad Katz
Abstract

Accurate modeling of rare, high-impact financial events is essential for robust risk management. Sampling-based techniques such as Sequential Markov Chain Monte Carlo (SMCMC) generate diverse scenarios but inefficiently explore high-risk regions, while Deep Reinforcement Learning (DRL) converges on narrow sets of scenarios, limiting diversity. We present GRID, which uses Generative Flow Networks (GFlowNets) -- samplers that draw objects in proportion to a reward -- to systematically generate diverse and realistic financial tail-risk scenarios. Building on continuous GFlowNets, GRID introduces a flexible mixture-policy architecture with variable-specific distribution families for high-dimensional, sequential macro-financial data, together with plausibility constraints that keep generated trajectories realistic. A central practical advantage is robustness to reward-proxy imperfection: because GFlowNets sample proportionally rather than maximize, they remain broadly exploratory even when the oracle's out-of-sample accuracy is low -- an unavoidable reality in financial forecasting. On three real-world crises, GRID improves scenario diversity, realism, and tail-risk coverage over RL and MCMC baselines, and ablations over oracle architectures and distribution families yield practical insights into continuous GFlowNet behavior under noisy rewards.

0.18

Transportable Offline Reinforcement Learning in Presence of Unmeasured Confounding

cs.LG ·2026-10-28 ·Junyi Liao, Hangqi Ren, Zhuoran Yang
Abstract

Offline reinforcement learning seeks to learn decision policies from previously collected data, but deployment environments may differ from the domains in which the data were collected, and logged variables may omit confounders that affected both historical actions and outcomes. We study offline policy learning under the joint challenges of domain discrepancy and unmeasured confounding, where multiple heterogeneous source datasets provide complementary information about a target domain. We develop a causally adjusted transportability framework combining transportability theory with instrumental variables, negative controls, and proxies to identify target-domain causal quantities. For confounded contextual bandits, identification reduces to systems of bridge-function integral equations, enabling minimax estimation from finite samples. Confidence sets for the identified quantities are then used for pessimistic policy optimization, yielding CAT-CCB. We extend the framework to episodic confounded Markov decision processes, where complementary sources identify target reward and transition mechanisms under structural and completeness conditions. Under Gaussian dynamics and linear function approximation, this yields CAT-CMDP. We establish high-probability finite-sample suboptimality guarantees that quantify the effects of approximation, estimation, source-to-target shift, offline coverage, and sample size. For bounded finite-dimensional linear classes, the contextual-bandit guarantee recovers the standard $\widetilde{\mathcal O}(N^{-1/2})$ rate. Numerical experiments support the effectiveness of the framework for learning target-domain policies from heterogeneous confounded data. Together, the results provide a principled approach when no single source dataset suffices for causal identification and policy optimization.

0.18

Value Probes: Learned Value Functions for Beam-Search Red Teaming

cs.AI cs.LG ·2026-10-28 ·Jonathan F. Jao, Andrew Kyle Lampinen
Abstract

Automated red-team auditing of language models is an important way to evaluate safety-relevant behaviors. These audits implicitly search a tree of possible conversation pathways: at each turn an auditor model decides which conversation to continue, and an LLM judge scores the dialogue. We emphasize that this auditing procedure is a Markov decision process, where states are the conversation history, actions are candidate responses, and rewards are judge scores for the target behavior. By adopting this perspective, we introduce two major improvements to auditing. First, we propose adopting beam search for the audit rather than relying on the auditor alone. We then interpret the problem of finding states that are likely to lead to problematic behavior as value estimation. Surprisingly, we show that this value function can be estimated by linear probing from the target model's activations to predict discounted future rewards - a method we term "Value Probes." By using Value Probes as a heuristic in our beam-search audit, we are able to substantially improve auditing success. Across 18 target behaviors and 3 scales of models, our approach raises peak audit scores by roughly 50% compared to (token-matched) standard auditing approaches. Furthermore, we show that Value Probes can be fit using off-policy data from smaller models. We also ablate individual components, and show that our method offers a comparably-performant and more efficient alternative to more expensive language model follow-ups. We hope these methods will improve language model auditing and lead to safer models.

0.18

Two-Stage Clipping: Linear-Time Approximate and Exact Mixed-Norm Projections for Transformers

cs.LG ·2026-10-28 ·Michel Barlaud, Jean-Paul Marmorat
Abstract

Large Language Models (LLMs) based on Transformer architectures have achieved remarkable performance across artificial intelligence tasks, but their computational cost remains a major bottleneck. Structured sparsification provides a promising approach to reducing this cost. In particular, $\ell_{1,\infty}$ projections can induce column-wise sparsity in model matrices, but existing proximal methods (Prox) typically require $\mathcal{O}(nm\log(nm))$ operations for matrices in $\mathbb{R}^{n\times m}$, limiting their computational and energy efficiency. In this paper, we introduce Two-Stage Clipping (TSC), an efficient structured sparsification operator inspired by bilevel optimization. Rather than solving a bilevel optimization problem in the strict sense, TSC decomposes the computation into two successive stages. First, group-level quantities are projected to allocate the global sparsity budget and identify active groups. Second, the resulting values are used as group-wise clipping thresholds for the matrix coefficients. For square attention matrices ($n \times n$), the complexity of our TSC method is $O(n^2)$, which is linear in the number of matrix coefficients. We establish a theoretical bound on the discrepancy between TSC and the exact proximal projection and show on synthetic data that the additional Euclidean approximation cost remains small in practice. Extensive benchmarks demonstrate substantial runtime improvements over proximal baselines and validate the theoretical norm identities underlying the proposed construction.

0.17

Auditable Belief Updates as a Missing Safety Layer for LLM-Powered Multi-Agent Systems

cs.AI ·2026-10-28 ·Zhuoyun Li, Boxuan Wang, Jinwei Hu, Xiaowei Huang, Yi Dong
Abstract

Large Language Models (LLMs) are increasingly being used as cognitive interfaces that shape how users interpret evidence, communicate, and make decisions. As deployed systems become more multimodal, tool-using, and multi-agent, safety failures no longer arise only from isolated harmful outputs, but also from how agents update, store, and propagate epistemic commitments across retrieval, memory, tools, and peer communication. In this paper, we define belief as an engineering abstraction for evidence-based commitments that guide downstream action. Building on this view, we propose the Explicit Belief Multi-Agent System (EB-MAS), which externalizes belief states into a structured substrate, uses belief operators to propose candidate updates, and applies a Cognitive Firewall to bound or reject updates that are weakly evidenced but strongly driven by vulnerability cues. We also discuss the risk of group belief at the group level, including manufactured consensus, collusion, and cascading misinformation, and outline practical integration challenges around grounding, false positives, scalability, and ethical misuse. Therefore, scalable agent safety requires not only better output filtering or stronger base models, but also explicit interfaces for tracing, calibrating, and constraining belief revision at runtime.

0.17

Episodic looking - using reinforcement learning to bias perception enhances few-shot and continual learning with protection from forgetting

cs.LG cs.AI ·2026-10-28 ·David Rawlinson, Gideon Kowadlo
Abstract · 21 pages, 8 figures

There is a fundamental contradiction between the desire for models that can generalize, and models that can continuously adapt to recent input with changing statistics. These conditions, captured in few-shot and continual learning settings, can bias or disrupt learned representations. This harms generalization to a broader context, leaving the model unable to overcome input variance. This paper proposes a solution based on the Complementary Learning System architecture (CLS), in which a Short-Term Memory (STM) delegates generalization to a slow, statistical Long-Term Memory (LTM). Delegation is achieved by working in a space of robust, generalized perceptions produced by the LTM. With this separation, the STM can focus on memorizing recent input and avoiding interference between memories. We show that on an image classification task, the resulting CLS model is more resistant to forgetting in a domain-incremental continual learning setting than a baseline ResNet LTM, and generalizes more effectively in a few-shot setting than other continual learners. In the continual setting, overall accuracy is slightly worse than existing continual learning models, but with a different mix of performance characteristics. Results analysis suggests that the model is limited by the representations and representational plasticity of the LTM. In future work, we will seek to address this by replaying STM memories to the LTM.

0.17

Fed-IGTD: A federated learning-based framework for tabular-to-image encoding

cs.LG cs.CV ·2026-10-28 ·David Chushig-Muzo, Eva Milara, Francisco J. Lara Abelenda, Luis Carlos Estrada-Petrocelli, Diego H. Peluffo-Ordóñez
Abstract

Tabular-to-image encoding methods, which transform tabular data into image-like representations, have shown promising results in predictive tasks. Despite their benefits, these methods follow a centralized paradigm, in which the data are collected and models are trained on a central device. This raises privacy and security concerns, particularly the risk of data leakage, unauthorized data collection and access to users' private information. Federated Learning (FL) provides a decentralized framework in which multiple clients jointly optimize a shared model without directly sharing training data. We introduce Fed-IGTD, a horizontal FL-based approach in which clients share a common feature space and upload only local statistics, from which the server reconstructs the global correlation matrix and obtains a feature-to-pixel mapping to train a convolutional neural network via Federated Averaging (FedAvg). We studied two differential privacy (DP) mechanisms: (i) a one-shot Gaussian mechanism protecting the correlation-aggregation step; and (ii) a client-level DP-FedAvg variant protecting the training with privacy loss composed via a Renyi-DP accountant. We evaluated Fed-IGTD on five datasets under a systematic sweep of non-IID severity and privacy budgets. Privacy protection applied to the mapping-construction step has little impact on predictive performance. By contrast, DP-FedAvg substantially degrades utility at moderate privacy budgets with a small client cohort (K=5). Ablation study indicates that this degradation is driven mainly by the injected noise rather than by uniform averaging or update clipping. Macro-F1 exhibits sensitivity to performance degradation than accuracy under non-IID heterogeneity and class imbalance.

0.17

Where You Inject Matters: The Trust Region Effect in LLM Agents

cs.AI cs.CRcs.LG ·2026-10-28 ·Chengquan Guo, Xiaodong Liu, Weiwei Yang, Mingqian Feng, Jialin Song +5 more
Abstract

LLM agents are increasingly adopted in long-horizon tasks and interact with diverse external environments. Despite their growing capabilities, these agents remain vulnerable to real-world attacks such as indirect prompt injections, which manipulate agents through malicious instructions embedded in external environments. While existing risk assessment work has primarily focused on the injection content, the effect of where an injection appears has not been systematically characterized. Studying injection placement requires agents to operate over longer-horizon tasks and interact with diverse environmental surfaces. Thus, we build three agentic testbeds spanning tool use, coding, and daily workflow domains with diverse injection surfaces. Holding the injection content fixed, we find that agent susceptibility varies across injection placements. We refer to this phenomenon as the trust region effect: agents implicitly assign different levels of trust to different regions of their environments. For example, an agent powered by GPT-5.4 follows an unsafe injection instruction placed in a repository documentation file (i.e., a README or TODO file) 2.6 times more often on average than the same instruction placed in a code or data file. Building upon this observation, we develop learning-based attackers that generate injection content and select better injection placements. Our approach demonstrates that both where an injection is placed and what content it contains can affect the strength of indirect prompt injection attacks. To the best of our knowledge, this work provides the first systematic characterization for the effectiveness of different injection placements, spanning both coarse-grained environmental surfaces and fine-grained positions. The uneven trust that agents implicitly place across their environments provides important insights for future attack and defense design.

0.17

Nearly Minimax Optimal Regret for Adversarial Linear Bandits with Strong Corruptions

cs.LG stat.ML ·2026-10-28 ·Qiwei Di, Quanquan Gu
theory
Abstract · 39 pages, 1 table

We study adversarial linear bandits with strong corruption, in which an adaptive adversary chooses the reward vectors and a strong adversary corrupts each observed reward after seeing the learner's action, subject to a known total corruption budget $C$. Strong corruption breaks the importance-weighted reward estimates used by adversarial bandit algorithms: by corrupting only the rounds in which a rarely sampled action is drawn, the adversary amplifies each unit of corruption by the inverse of its sampling probability. The best known bound for this problem, $\tilde O(\sqrt{d^3T} + dC)$ by Liu et al. (2024), is a factor $\sqrt{d}$ above the corruption-free rate in its leading term, and whether $\tilde O(d\sqrt{T} + dC)$ regret is achievable was left open. We propose CRW-FTRL, which runs follow-the-regularized-leader with a radius-weighted entropy regularizer and estimates rewards with a normalized kernel estimator whose sensitivity to corruption is governed by per-action radii instead of inverse sampling probabilities. The radii are nondecreasing and are chosen jointly with the sampling distribution through a monotone log-determinant barrier problem, so that a negative linear term in the regularizer cancels both the corruption bias and the estimation error of every comparator. For $K$ actions in $\mathbb{R}^d$, CRW-FTRL achieves $\tilde O(\sqrt{dT\log K} + dC)$ regret with high probability against an adaptive adversary, which matches the best known corruption-free rate and the $Ω(dC)$ lower bound up to logarithmic factors. Applied to a discretization of a general compact action set, it yields $\tilde O(d\sqrt{T} + dC)$ regret, which settles the statistical rate of corrupted adversarial linear bandits under strong corruption.

0.17

Latent Decision Flow Rewards for Clinical Guideline Reasoning

cs.LG cs.AI ·2026-10-28 ·Gaurav Bhatt, Marinka Zitnik
Abstract

Clinical practice guidelines encode evidence-based recommendations as conditional decision flows, which makes them a natural source of process supervision for training reasoning models. Two properties of clinical reasoning make this supervision difficult to use. A decision flow consistent with a guideline does not by itself determine the correct answer for a given patient, which we call the path--answer gap, and several valid decision flows can support the same clinical decision, which we call path multiplicity. Current training approaches address parts of this problem. Answer rewards supervise the final decision, and process rewards computed against a reference flow supervise one valid decision flow per case. Supervised fine-tuning on annotated flows teaches guideline-consistent reasoning but constrains the model to one trajectory per case and can reduce answer accuracy. We introduce GUIDER, a method that jointly optimizes a decision flow controller and a reasoning policy. The controller learns to generate plausible decision flows from guideline-derived supervision, and the policy learns to generate a decision flow and a final answer for each patient case. GUIDER computes the process reward for the policy over several controller-generated flows, which accounts for path multiplicity, and gates this reward on answer correctness, which addresses the path--answer gap. Joint training allows the process supervision to evolve with the policy. On MedGUIDE, GUIDER with Qwen3-14B achieves 82.5 weighted accuracy and 73.5 guideline adherence, 3.3 and 3.1 points above the strongest clinical reinforcement learning baseline. With Qwen3-14B and Gemma4-12B backbones, GUIDER also improves over CRPO, MRPO, and FaithMed on the out-of-domain benchmarks AMEGA-LLM, RealMedQA, and CancerGUIDE.

0.17

Cinematize from Unreal: Agentic Choreography for Controllable World Simulation

cs.MA ·2026-10-28 ·Zifan Song, Caihua Shan, Kaitao Song, Yifan Yang, Dongsheng Li +1 more
Abstract · 25 pages, 9 figures, 4 tables

Controllable world simulation requires more than visually plausible frames: scenes must retain persistent state, coordinated actions, and editable temporal structure. Generative world models offer flexible visual synthesis, but their outputs do not always expose the objects and transitions needed for precise revision. Graphics engines provide executable state and direct control, yet translating language into valid operations requires specialized knowledge of assets, geometry, and interdependent APIs. We present CineU (Cinematize from Unreal), an agentic system for language-driven world orchestration and cinematic choreography within Unreal Engine. From a code-based world modeling perspective, an Unreal project serves as a persistent, inspectable world representation, while agent-generated code specifies operations whose effects can be executed and verified. CineU combines hierarchical orchestration, cascaded asset retrieval and geometry-constrained placement, capability-aligned code synthesis, and multi-stage verification with localized repair. Shared scene and timeline context coordinates entities, actions, camera trajectories, and shot boundaries while supporting dependency-aware revisions. Experiments on scene construction, selective editing, multi-shot cinematography, and component ablations show how specialized execution knowledge and engine feedback improve task completion and targeted recovery. CineU illustrates how executable world state can complement learned visual generation to support controllable and revisable simulated worlds.

0.17

Benchmarking Graph Neural Networks for AC Power Flow: The Role of Inductive Biases and Physical Constraints

cs.LG cs.AI ·2026-10-28 ·Federico Clerici, Ferran Bohigas-Daranas, Hamid Latif-Martínez, Eduardo Prieto-Araujo, Oriol Gomis-Bellmunt +1 more
benchmark
Abstract

Fast and accurate models for AC power flow (ACPF) are increasingly important as modern electrical grids require more frequent system analyses. Graph neural networks (GNNs) naturally exploit electrical grid topology, yet the effects of architectural design choices on predictive performance remain poorly understood given differences in graph representations and evaluation settings across studies. We present a unified benchmark of representative GNN architectures for ACPF prediction under a common experimental framework. All models use identical preprocessing and training procedures and comparable parameter budgets, isolating the impact of message-passing mechanisms and inductive biases. We also analyze errors by distance from the slack bus and systematically ablate key design components. We evaluate predictive accuracy and physical consistency on the IEEE 14-bus and IEEE 118-bus systems. Results indicate that anisotropic message-passing architectures offer the most favorable trade-off between predictive accuracy and physical consistency under the studied parameter and topology constraints. Explicit edge features encoding line impedances and global context aggregation are particularly important for representing electrical interactions and system-wide dependencies, especially for voltage-angle predictions on the topologically bottlenecked IEEE-118 grid. Under a budget with approximately fixed parameters, two or three message-passing layers generally provide the best balance between depth and representation width, while additional depth offers no consistent improvement. Ablations also show that standard residual and root connections are not necessarily beneficial in shallow models, and that aggregation operators involve trade-offs between statistical accuracy and physical consistency. These results provide practical guidelines for designing graph-based power system models.

0.17

TreasuryQA Bench: Can AI Agents Work an Archive of Scanned Financial Records?

cs.AI ·2026-10-28 ·Ritiz Tambi, Sahar Shayegan, Joshua Meyer, Sun Kim, Ali Khan +3 more
benchmark
Abstract

Organisations keep decades of financial records as scanned PDFs on shared drives, and AI agents can now reach those drives through Model Context Protocol (MCP) servers. Existing benchmarks do not mirror this production setting because they simplify the task. In some, the relevant documents are handed to the agent, so it never has to search. In others, the agent searches through documents with pre-extracted text, so it never has to read scanned images. We introduce TreasuryQA Bench, which requires both: 70 questions over 1,435 scanned U.S. Treasury publications, held in Veris's high-fidelity Dropbox twin. In each task the agent searches for the relevant file through Dropbox's MCP server, reads the pages it needs with Reducto's hosted OCR server or its own tools, and replies with one exact figure. A Python grader checks each answer against an accepted range around the correct value. In this paper we present results from ten agents. The best, Claude Opus 5.5 in Claude Code, solves 64.8% of attempts and the worst, GPT-5.6 Terra in Codex, solves 16.7%, so the benchmark is far from saturated. We also test how scan quality affects performance: every agent solves fewer tasks on the blurriest scans. Most failed attempts reply with a wrong number rather than no number, and the most common cause is a digit misread from the scan. Three worked examples are included with their answer keys.

0.17

Can LLMs Faithfully Solve Ontology Engineering Tasks? A Mechanistic Interpretability Study

cs.AI ·2026-10-28 ·Bohui Zhang, Albert Meroño-Peñuela, Elena Simperl
benchmark
Abstract

Ontological knowledge underpins large language model (LLM) performance on knowledge-intensive tasks. However, existing evaluations of LLMs' ontological understanding largely rely on behavioral correctness, leaving the underlying mechanisms unexplained. Existing mechanistic interpretability work has largely focused on instance-level factual knowledge, leaving ontological knowledge under-explored. To bridge this gap, we introduce MIOE, a causal-abstraction benchmark that evaluates interchange interventions for two ontology engineering tasks: entity typing and subsumption prediction. MIOE tests whether learned internal intervention targets can change an ontology-level target answer while preserving class-specific auxiliary information. Across Qwen3-8B and Qwen3-4B, we evaluate SAE-, DBM-, and DAS-based methods, including multi-task and multi-layer variants. Distributed subspace methods give the strongest results and generalize across prompt settings, but the induced changes remain only partially separated from auxiliary class-specific information. Multi-task objectives and contiguous layer-window interventions improve this balance, while class-level analyses reveal substantial variation across ontology classes. These results suggest that ontological predictions are mechanistically accessible in LLMs, although current methods do not yet isolate a cleanly disentangled causal mechanism for ontological knowledge.

0.16

Contextual Embeddings Beyond the Gold Chunk

cs.LG cs.CLcs.IR ·2026-10-28 ·Sedigheh Eslami, Markus Krimmel, William Barber, Antoine Chaffin, Maximilian Schall +5 more
benchmarkposition
Abstract

Contextual embeddings represent chunks of a document with respect to the document's global context. These models are typically trained and evaluated under the assumption that a relevant document contains only a single relevant chunk. In this work, we argue that in real-world use cases a retriever should not only find answers spread over several chunks but also surface the evidence needed to verify them. To address this, we introduce pplx-embed-v2-context, a family of contextual embedding models trained by distilling graded chunk-level relevance from a query-aware context compression model. This removes the need for gold-chunk labels and makes the training targets independent of the chunking. We utilize a vision-language backbone for training and release pplx-embed-v2-context in 9B and 0.6B sizes with compatible embedding spaces. On the evaluation side, we introduce two benchmarks that label both the chunks that answer a query and the evidence needed to verify the answer. Context-bench is a controlled diagnostic, and Q2D-Chunk is built from real-world search queries. To encourage further research, we also publish Q2D-Chunk as a public leaderboard. Our 9B model sets a new state of the art on both benchmarks and ConTEB. We find that the visual retrieval capabilities of our 9B model transfer from its vision-language base model and propose to extend chunking to image inputs. We demonstrate that our models achieve state-of-the-art performance among dense retrievers on visual ViDoRe-V3 and that performance improves with granular image chunking.

0.16

Co-Evolving Policy and Rubric via Capability Awareness for Open-Ended Generation

cs.AI ·2026-10-28 ·Chengsheng Zhang, Wenshen Xu, Yuchuan Wu, Zhiyuan Yu, Junfeng Fang +4 more
Abstract · 30 pages

Rubric-based reinforcement learning (RL) offers interpretable, fine-grained supervision for aligning large language models (LLMs) on open-ended generation tasks. However, static rubric criteria may become saturated or remain beyond the policy's exploration, rendering the reward signals insufficiently discriminative. While recent efforts seek to guide sampling exploration or evolve rubrics, they fail to explicitly factor in how rubric difficulty should track the evolving policy's capabilities, resulting in a persistent mismatch between the optimization objective and the policy's current capability frontier. In this paper, we propose \textbf{Co}-Evolving Policy and Rubric via \textbf{C}apability \textbf{A}wareness (\textbf{CoCA}), a closed-loop framework that jointly adapts the rubric and policy through criterion-level diagnosis. Specifically, CoCA contrasts query-only and rubric-guided rollouts to characterize each criterion relative to the current policy, recalibrates criteria with mismatched difficulty, and identifies guidance-revealed capability gaps for frontier-based learning. Based on this insight, CoCA either expands rubric-conditioned frontiers through guided optimization or transfers qualified gains through on-policy self-distillation. As the policy updates, CoCA improves the capability diagnosis and enables more reliable distillation, thereby completing the co-evolution loop. Extensive experiments across diverse benchmarks and LLMs demonstrate that CoCA consistently improves performance over rubric-based RL methods. Moreover, CoCA promotes the internalization of guidance-revealed capabilities into query-only behavior while mitigating capability degradation.

0.16

Routed Continual Learning

cs.LG cs.MA ·2026-10-28 ·Michael Philip, Mahesh Balakrishnan, David Geraghty, David Mao, Vidhya Venkat +1 more
Abstract

Agents deployed at scale repeatedly encounter related problems, and it is natural to let them learn from each other's experience rather than starting over each time. Unfortunately, there is no single way to retain that experience that works well across settings, and no way to know in advance which future problems a given experience will help with. A global history preserves every detail but becomes expensive and mixes together experience that may not belong together; compressing that history into one fixed representation limits the cost but can throw away valuable details. We introduce Routed Continual Learning (Routed CL), an architecture that decides how to partition and retain experience. Past trajectories are grouped into clusters, each of which maintains state using one memory system from a fixed library. Before each run, an LLM-based router either reuses an existing cluster or creates a new one, basing its decision on the problem statement, memory system descriptions, and cluster summaries. We also introduce Context Memory Adapter, which mediates between the memory system and the agentic loop. We evaluate the architecture on a heterogeneous stream that interleaves 301 instances from the six CL-Bench tasks with 97 banking instances from $τ^3$-Bench. Across four language models and five matched rollouts, Routed Continual Learning raises normalized reward over a single global in-context learning (ICL) history by 15-30% of the maximum attainable normalized reward, achieves higher mean normalized reward than ICL-only clustering, and reduces estimated API cost by 65-72% relative to Global ICL. These results support approaching continual learning as a routing problem that chooses both a learning mechanism and a state cluster rather than committing to one representation globally.

0.16

REVBRAIN: Real-time Probabilistic EEG Forecasting Using Residual Vector Quantization

cs.LG eess.SPq-bio.NC ·2026-10-28 ·Hanna Pankka, Jaakko Lehtinen, Timo Roine
Abstract

Real-time forecasting of resting-state EEG signals is required by applications like brain-state-dependent stimulation, where computer generated actions are coupled with ongoing neural oscillations. Recently, the deep learning model WaveNet was proposed for the forecasting task. It shows superior accuracy compared to the prevailing method, the linear autoregressive (AR) model, and provides a built-in uncertainty measure due to its probabilistic modeling framework. However, inference using WaveNet is too slow for real-time use due to its sample-by-sample autoregressive operation: a single forecast requires evaluating the model tens or hundreds of times depending on the length of the prediction window. In this paper, we present REVBRAIN, a probabilistic deep learning method that produces forecasts in real time by generating the forecast over the whole prediction horizon at once. This is achieved by tokenizing the output space using residual vector quantization (RVQ) and generating forecasts over the items in the tokenized space. We tested REVBRAIN for forecasting 200 ms of EEG time series in the alpha-frequency band (8-13 Hz) and compared our results against WaveNet model. We found that REVBRAIN achieved real-time inference on a laptop GPU and an over $50\times$ speedup compared to WaveNet, while reaching superior prediction accuracy. By achieving real-time inference speed, REVBRAIN enables incorporating accurate and probabilistic EEG forecasting into BCI applications, opening up new ways for investigating the functioning of the human brain.

0.16

Learning to Reason from Reference Samples Alone: Reinforced Adversarial Learning for Modern LLM Post-Training

cs.AI cs.LG ·2026-10-28 ·Albert Bou, Javier Valdes-Aleman, Jon M. Laurent, Samuel G. Rodriques, Andrew D. White +2 more
Abstract · 31 pages, 6 figures, 10 tables, including references and appendices

In many settings where LLM post-training is desirable, high-quality answers are recognizable but difficult to score with explicit rewards. Reinforcement learning with verifiable rewards (RLVR) is limited to tasks where an explicit correctness signal is available. Prior adversarial approaches to training reasoning LLMs either complement verifiable rewards with co-trained critics or require reference reasoning traces. We ask whether a co-trained LLM judge can turn reference answers alone into the sole reward for on-policy post-training of reasoning LLMs. We call this method Reinforced Adversarial Learning (RAL). We first validate RAL on three verifiable tasks (integer multiplication, code-based arithmetic-expression synthesis, and SMILES-to-formula translation), using the verifiers only for evaluation. With sufficiently capable judges and enough training, RAL matches direct RLVR. We then apply RAL to scientific abstract generation in two settings. In the first setting, RAL guides the generator to produce abstracts in the scientific domain of the reference abstracts, even though the prompt leaves that domain unspecified. In the second setting, we generate abstracts for U.S. National Institutes of Health grant proposals conditioned on the grant title. Human scientists and an ensemble of LLM judges find our generations harder to distinguish from real abstracts than those of frontier-LLM baselines. These results suggest a path toward post-training reasoning models in scientific and other high-value domains, where increasingly capable judges infer hard-to-specify reward functions from example answers.

0.15

Encode Once, Answer Many: Internalizing Context via Parametric Flow

cs.AI ·2026-10-28 ·Xinhao Zhong, Mengxuan Wu, Chonghua Liao, Linxi Xie, Ruicheng Feng +21 more
Abstract

Large language models are typically adapted to a task either by placing knowledge in the input context, which keeps it external and re-encodes it on every call, or by fine-tuning, which internalizes it but demands a costly round of optimization per task. We aim to internalize the knowledge carried by a context directly into model parameters, without per-task optimization and without keeping the context in the input at inference time. To this end, we propose CLIK, a context-conditioned parameter generator that maps a given context to LoRA parameters for internalizing knowledge in a single forward pass through a Transformer encoder with cross-attention, requiring no test-time gradient updates. Once the generated LoRA is inserted into a frozen base model, the model answers downstream queries from the query alone, having internalized the context into its weights rather than re-reading it. Because the adapter depends only on the context, it is generated once and reused for every query on the same document, so the cost of reading that document is amortized across queries. On document question answering, CLIK substantially improves over the no-context base model and outperforms existing hypernetwork-based generators, while remaining behind a base model that reads the context directly, and it carries over to multi-turn dialogue without any multi-turn-specific training. Analysis further shows that the context-dependent part of the generated adapter concentrates in a compact and specific substructure.

0.15

Optimization-based lane-changing trajectory planning in mixed traffic with embedded input convex neural networks

cs.LG math.OC ·2026-10-28 ·Yuanxiang Yang, Yu Liu, Fabricio Oliveira, Yu Han, Claudio Roncoli
Abstract

Lane-changing (LC) maneuvers are highly challenging vehicle interactions, with substantial implications for traffic safety and efficiency. Developing effective LC strategies for connected and automated vehicles in mixed traffic is complicated by the uncertainty associated with surrounding-vehicle behavior and the computational burden of embedding realistic prediction models within optimization schemes. To capture these complex interactions while retaining tractability, this study proposes a data-driven optimization framework that embeds neural network (NN) surrogates for safety and lateral comfort within a Bézier-curve-based trajectory planner. Specifically, the safety cost during the LC process is quantified via a surrogate composed of an input convex recurrent neural network (ICRNN) encoding the planned trajectory, a fully input convex neural network (FICNN) embedding the LC duration, and a partially input convex neural network (PICNN) decoder fusing them with historical traffic states; separately, another FICNN surrogate approximates the optimal lateral comfort cost as a function of the LC duration. Both surrogates are embedded via linear programming (LP) formulations, avoiding the binary variables required by conventional mixed-integer programming (MIP) approaches for NN-embedded optimization. We further theoretically establish their exactness at optimality and extend the framework to problems with additional analytical objective terms and multiple surrogates. Experiments on a hybrid dataset combining real-world and simulated trajectories demonstrate that the proposed method achieves surrogate accuracy competitive with unconstrained baselines while substantially improving optimization efficiency over conventional MIPs. Closed-loop simulations further show that the proposed planners markedly reduce the weighted planning cost and consistently outperform baseline methods.

0.15

On-Policy Self-Distillation with Teacher Evolution

cs.LG cs.AI ·2026-10-28 ·Shufan Shen, Zhongni Hou, Chao Bi, Yufei Zhang, Guojun Yin +1 more
Abstract

On-Policy Self-Distillation (OPSD), which aligns a student model's token-level distribution with that of its self-teacher prompted by privileged context, has emerged as a promising direction to mitigate reward sparsity in sequence-level post-training paradigms. However, OPSD often fails to achieve continuous performance gains during training. In this paper, we find that the model's limited self-teaching capability underlies this phenomenon by generating unreliable guidance on certain samples. To address this problem, we propose Teacher-Evolution Self-Distillation (TESD), which enables the self-teacher to generate rollouts and receive environmental feedback based on task correctness and context independence. First, using these rewards as training objectives, we perform teacher-oriented optimization to continuously enhance the model's self-teaching capability during distillation. Second, by estimating teacher guidance reliability from these rewards, we apply adaptive student-oriented optimization to down-weight samples with unreliable guidance. Extensive experiments across multiple datasets and model scales show that TESD consistently outperforms representative token-level and sequence-level methods, highlighting the necessity of improving supervision quality in OPSD.

0.15

Deogen: Automatic Benchmark Generation for Reasoning over Policies

cs.AI cs.LG ·2026-10-28 ·Sam Kouteili, William Liang, Keeran Murray, Adriana Zhou, Ramesh Viswanathan +2 more
benchmark
Abstract

Language models are increasingly being deployed in regulated settings, where a model must apply given rules to the facts of a situation to determine a verdict. Benchmarking this capability, known as deontic reasoning, remains a costly and manual process, with existing datasets largely hand-crafted. In this paper, we present DeoGen, a scalable, automatic, and parametrizable benchmark generator for deontic reasoning. By autoformalizing policy rules to first-order logic, we can construct decision trees that capture possible outcome trajectories. Each path down a tree becomes a benchmark entry, with each tree producing combinatorially many entries relative to tree depth. Paths ending at a leaf entail a verdict, while non-leaf paths measure a model's ability to abstain when presented with insufficient information. Evaluations show frontier models return the correct verdict 89-92% of the time. Notably, performance degrades (22-40% correct) on entries without sufficient information to claim a verdict, even with explicit instructions to abstain in such cases. In all, DeoGen presents an automatic, scalable, and parametric method for generating deontic reasoning benchmarks.

0.15

Context Compaction for World Models

cs.AI ·2026-10-28 ·Hayder Tirmazi
benchmark
Abstract

World models let observers explore and interact with generated worlds. Recent work shows that when observers revisit a region, current models often fail to regenerate it consistently. We view this as a context retrieval problem. LLM agents manage context retrieval at two layers of their stack. At the model layer, the KV cache design determines how much context fits in memory. At the agent harness layer, harnesses such as Claude Code pick the context sent to the LLM. Prior work on world models has mostly focused only on the model layer. In this paper, we make three contributions. First, we build a benchmark to measure world model context retrieval. It generates new worlds for every episode, preventing a model from recalling rooms using training data. It also separates context retrieval from guesses based on nearby frames. Second, we build an agent harness that performs context compaction for world models, which to our knowledge no prior work has studied. Third, we use our benchmark to study four world model designs. Context compaction lets our world models retrieve rooms up to 4 times beyond their context window without retraining. However, world models with sliding-window attention lose most of their context retrieval when dreaming. Therefore, a dreaming world agent benefits more from extending the reach of its world model using context compaction than switching to sliding window attention at the model layer.

0.15

What to randomize matters: a controlled study of volume, axis and representation in low-data molecular generation

cs.LG cs.CEcs.CLq-bio.BM+1 ·2026-10-28 ·Neerav Kaushal, Alisa Teige, Ajay MNV Penmatsa
Abstract · 27 pages, 9 figures, 5 tables

Chemical language models can struggle to generate new molecules when trained on small datasets, where a model may reproduce training examples rather than learn a useful generative distribution. Randomized SMILES augmentation varies atom traversal, whereas Sequential Attachment-based Fragment Embedding (SAFE) also allows the order of fragment blocks to vary. How these distinct sources of representational variation affect low-data molecular generation has not been clearly studied. Here, using 1,000 seed molecules to train an autoregressive sequence model, we compare canonical SAFE with randomized SAFE at increasing corpus sizes, isolate atom-order and fragment-order axes of SAFE randomization, and compare fully randomized SAFE with randomized SMILES at molecule-by-molecule matched training corpora. Three findings emerge. First, randomization of SAFE alone raises the genuine yield of novel molecules more than 450-fold, turning a model that merely reproduces its training set into a practical molecular generator. Second, the two axes of randomization act jointly rather than additively: atom order alone reaches a tenth, and fragment order alone a seventh, of the genuine yield obtained when both vary, which is four times their combined contribution. Third, at matched corpus size and compute, randomized SAFE delivers 2.8 times the genuine yield of randomized SMILES. Welch tests with Games-Howell post-hoc comparisons resolve every difference on the primary outcomes. In the low-data regime, thus, what one randomizes rather than only how much determines whether augmentation produces new molecules or merely new spellings of the training set.

0.15

HotPOT: Operator Transformer with Heterogeneous Physics Experts for Large-Scale PDE Pre-training

cs.LG ·2026-10-28 ·Jiayi Li, Guancheng Zhou, Yisi Luo, Ruihua Chen, Penghao Jiang +3 more
code/data
Abstract

Large-scale PDE pre-training trains a single model on the data from multiple partial differential equations (PDEs). However, adding heterogeneous PDEs with different governing equations, physical variables, and variable interactions does not always improve predictive performance across tasks. Existing PDE mixture-of-experts models mitigate cross-task interference through learned routing, but typically use homogeneous expert architectures, making it difficult to directly reflect differences in state evolution across equations. We introduce HotPOT, an operator transformer with heterogeneous physics experts for large-scale PDE pre-training. HotPOT uses a shared encoder to capture information common across PDEs, while each PDE family uses a neural-numerical expert that pairs a learnable predictor with a fixed numerical update containing no trainable parameters. Each predictor outputs the quantities required by its paired numerical update, which then computes the next state. This design allows different PDEs to share a common encoder while retaining their own learnable predictors and numerical updates. Experiments across five PDE families and six settings show that HotPOT outperforms MoE-POT in all 24 matched-scale comparisons before and after fine-tuning, reducing rollout error by 50.7% on average. Notably, the 17M model also surpasses substantially larger MoE-POT models, and the performance gains extend to zero-shot prediction under unseen physical conditions. Code is available at https://github.com/cruiseresearchgroup/HotPOT-code.

0.14

Averaged Annealed Markov Chains Achieve the Algorithmic Threshold for Tensor PCA, Part II: Langevin Dynamics

cs.LG cs.DSmath.PRmath.ST ·2026-10-28 ·Aniket Das, Tselil Schramm
Abstract · Companion paper to arXiv:2610.16705 (Part I). 71 pages, 1 figure

We consider the tensor principal components analysis (tensor PCA) problem on the sphere, in which we are asked to recover an unknown vector in $\mathcal{S}_N=\{x\in\mathbb{R}^N:\lVert x\rVert^2=N\}$ from a noisy observation of its $k$-th tensor power, for a fixed $k\ge3$. Tensor PCA exhibits a well-studied information-computation gap: global algorithms such as spectral methods and sum-of-squares work above an algorithmic threshold, below which we know matching lower bounds (against these and some other restricted models of computation). In contrast, local algorithms such as Markov chains were only known to succeed from a random initialization at a polynomially larger signal-to-noise ratio, and were believed to be inferior to global algorithms for tensor PCA. Continuing the program initiated in Part I [arXiv:2610.16705], we show that spherical Langevin dynamics, when supplemented with annealing and ergodic averaging, achieves polynomial-time strong recovery above the algorithmic threshold from a uniformly random initialization. Together with Part I, our results demonstrate that local Markov chains can succeed all the way down to the algorithmic threshold from an uninformative cold start.

0.14

Planned Restlessness: Delayed Meta Bandits with Budgets

cs.LG cs.DS ·2026-10-28 ·Sudipto Guha
theory
Abstract

We study episodic finite-horizon Bayesian bandits, or meta-bandits, with independent arms. Each episode draws fresh arm qualities from unknown recurring priors. Observations arrive after known delays, and plays obey hard per-episode caps and total budgets. A play earns reward and informs both the current quality and the recurring prior: the learner must explore, use, and generalize. We formulate a common planning optimization over learning trajectories. Its LP relaxation values experiments through current reward, resource consumption and continuation: the value of future decisions from the learning state produced. Dual prices separate the arms into local programs and certify every admissible policy, including hierarchical Thompson sampling. Execution enforces the hard constraints while preserving the promised learning transitions or accounting for their loss. The framework also represents risk-sensitive objectives and prices purchased observations and buyout options. Delayed reports create planned restlessness: an arm's information changes while it is not played, through reports of its own earlier plays arriving on a known calendar. For stationary priors, finite reward alphabets, bounded report lag and nonbinding budgets, posterior sampling over the recurring priors bounds the price of not knowing them by $O(\sqrt{N})$ over $N$ episodes, with other parameters fixed. For one undelayed $T$-step episode with arms capped at $T/X$ plays each, we obtain the LP-relative approximation factor $1+(2πX)^{-1/2}+O(X^{-1})$, improving previous factors, with a matching lower bound for the leading term.

0.13

CausalGrade: Evidence-Guided Causal Graph Scoring with Jev

cs.AI stat.MLcs.LG ·2026-10-28 ·Jiangming Zhong, Jiawei Zhu, Guanting Wu, Qiaoxi Cai, Wanyi Chen +1 more
Abstract · 6 pages, 2 figures

Evaluating the structural quality of candidate causal graphs when the true causal structure is unknown is an important problem in applying causal discovery results. This paper proposes CausalGrade, a causal graph scoring method based on the probabilistic judgment model Jev. The method jointly provides Jev with evidence of association, conditional dependence, and directionality from observational data and edge probabilities produced by the causal discovery foundation model CDFM. Jev produces probabilistic judgments about direct causal edges, and a graph-level score is obtained by measuring the agreement between the adjacency structure of the graph under evaluation and these edge-level judgments. Across 300 synthetic datasets, the score exhibits a relatively strong negative correlation with structural error in all 20 combinations of four causal discovery algorithms and five variable counts, indicating that higher scores generally correspond to smaller structural errors. In 500 synthetic graph selection tasks, CausalGrade achieves a best-candidate hit rate of 54.4\%, exceeding the 48.4\% achieved by the Self-Compatibility metric \(κ_G\), further supporting its ability to discriminate the structural quality of graphs under evaluation.

0.13

SCOPE: Selective Critical-Step On-Policy Self-Distillation for Agentic Reinforcement Learning

cs.AI ·2026-10-28 ·Qirui Liu, Wei Cao, Yu Mi, Yichen Sun, Xuekun Zhang +3 more
Abstract · 18 pages, 8 figures, 5 tables

Outcome-based reinforcement learning provides coarse supervision for long-horizon LLM agents: failed trajectories may contain correct steps, while successful trajectories can contain errors. Trajectory-level skills can also shift teacher preferences at unrelated steps, introducing off-target distillation signals. We introduce SCOPE, a framework for Selective Critical-Step On-Policy Self-Distillation. After a supervised warm-up, the policy analyzes its own trajectories, generates step-specific hindsight skills, and distills their guidance only at selected error steps. A bounded signed teacher-student coefficient stabilizes auxiliary updates while preserving positive reinforcement and negative correction. A success-relative pairwise length reward further encourages efficient task completion. Experiments on ALFWorld and WebShop demonstrate consistent improvements across multiple model backbones. With Qwen2.5-3B, SCOPE achieves 97.7% success on ALFWorld and 85.2% on WebShop, outperforming SEED by 5.9 and 6.3 percentage points, respectively. On unseen ALFWorld tasks, SCOPE reaches 89.6% success, improving over GRPO by 23.2 percentage points with 46.7% fewer interactions. Scaling to Qwen3-32B, SCOPE achieves 99.2% success on seen ALFWorld tasks and 94.0% on unseen tasks.

0.12

In-Context Learning for Multi-Fidelity Multiscale Model Closure and Cross-Operator Out-Of-Distribution Generalization

cs.LG math.NA ·2026-10-28 ·Xinyuan Li, Matthew Zahr, Panos Stinis, Zecheng Zhang
Abstract

Multi-fidelity learning uses low-fidelity information (coarse-scale PDE models) together with limited high-fidelity data (fine-scale solutions) to obtain high-accuracy PDE solutions. It is a data-driven approach to downscale reduced-order models to solve PDEs efficiently and accurately. Standard multi-fidelity learning focuses on one operator or PDE; given a new PDE, retraining using additional coarse-fine solution pairs may be required, which is costly. In this work, we propose an in-context multi-fidelity learning framework for learning multiple multi-fidelity operators with one shared model. The model is trained using paired low- and high-fidelity examples as demonstrations from multiple operators, together with a new coarse-scale solution query, to predict the corresponding fine-scale solution. Once trained, given a new PDE operator, a few (fewer than five) coarse-fine demonstration pairs, and a low-cost coarse query, the model predicts the corresponding fine-scale solution without fine-tuning. When the target operator appears in the training dataset but has a different prediction goal, such as predicting beyond the training time interval, no additional demonstrations are needed. Instead, the model uses training coarse-fine pairs from that PDE and information from other training operators. Testing then only requires generating a low-cost coarse-scale solution query, without fine-tuning. Notably, our smallest model contains only 255,364 trainable parameters, corresponding to approximately 456 parameters per training operator for 560 operators.

0.12

Post-Training Large Language Models with LLMON

cs.AI cs.LG ·2026-10-28 ·Basel Shbita, Michael Hind, Chad DeLuca, Nathan Fulton, Shubhi Asthana +6 more
Abstract

Large Language Models (LLMs) treat all context tokens uniformly, lacking explicit separation between executable instructions and passive data. This leads to prompt injection vulnerabilities, ambiguous instruction binding, and unreliable execution control. We present a post-training methodology using LLMON (LLM Object Notation), a framework that enables separating instructions from data and requires explicit identifier-based execution. We construct a large-scale structured training corpus of 4.6B tokens spanning general, RAG (Retrieval-Augmented Generation), and tool-calling data, augmented with distractor-infused variants in which competing instructions and irrelevant data segments are present in context but not referenced. Post-training two Granite-4.1 base models (3B and 8B) through supervised fine-tuning followed by reinforcement learning, we show that LLMON-structured supervision reduces attack success rates on indirect prompt-injection benchmarks from 53-100% to 0.02-5.6% while preserving general capabilities. We further introduce attention-based interpretability metrics that diagnose failure modes at individual transformer layers for structured inputs, revealing how our approach reshapes internal attention patterns. Our results demonstrate that explicit execution semantics can be taught through standard post-training without architectural modification.

0.11

How Much Do Teacher Bridges Matter in Interleaved On-Policy Distillation?

cs.LG cs.CL ·2026-10-28 ·Leran Chen, Zhiyuan Yao, Lingnan Kong
Abstract · 25 pages, 8 figures, 18 tables

On-policy distillation (OPD) trains a student using teacher predictions on student-generated responses. Some methods insert teacher-written segments, or bridges, but disagree on how to train on those tokens. We compare no loss, forward KL and reverse KL on identical trajectories using Qwen3-1.7B as student and Qwen3-4B-Instruct-2507 as teacher. We separate gradients by which model wrote each token. Under our short training runs and BF16 inference, pre-specified tests detected no added accuracy benefit from teacher text as context or as a training target. A one-paragraph bridge after the student's first paragraph is only 2.7% of the tokens and contributes a small gradient nearly orthogonal to the student-token gradient. In this main comparison, the three loss treatments gave similar gradient directions and were statistically equivalent in accuracy within a +/-5 percentage-point margin. An exploratory pilot placing the bridge where failed responses disagreed most with the teacher suggested a benefit from teacher-token loss. A pre-specified replication did not confirm this, and fourfold weighting showed no significant accuracy benefit. Two limitations bound these findings. Among the tested amounts of training on fixed responses, gains peaked after four updates before the student increasingly failed to stop generating. None of the tested online training variants met the criteria for stable stopping and improved accuracy. Our evaluations used BF16 inference, which rounds away most of the small weight changes. When evaluated in exact FP32, the tested replication models frequently reached the response-length limit. Our accuracy findings therefore concern BF16 inference; they do not establish that teacher-token supervision is generally unnecessary.

0.11

SkillCT: A Multidimensional Benchmark and Execution-Free Baseline for Agent Skill Assessment

cs.AI ·2026-10-28 ·Mengyu Sun, Zunlong Zhou, Ziyuan Yang, Yi Zhang
benchmark
Abstract

Agent skills package instructions, scripts, and resources for specialized tasks and are increasingly distributed through public Web repositories. However, existing skill benchmarks primarily focus on downstream task execution, insufficiently assessing specification quality, behavioral robustness, semantic alignment, and responsibility boundaries. Their reliance on task-specific tools, environments, and repeated executions limits scalability and pre-deployment assessment. To address these challenges, we introduce SkillCT, a benchmark that broadens skill assessment from downstream utility to a multidimensional characterization of skill quality. SkillCT evaluates skills across nine complementary dimensions spanning three aspects: Static Assessment, which captures intrinsic properties of the skill specification; Dynamic Assessment, which measures whether the induced behavior remains stable across backbone models and request variations; and Human-Assisted Assessment, which examines behavior-chain consistency and task granularity. These dimensions provide a broader view of skill quality than execution success. Applying SkillCT to a public skill corpus establishes a reference benchmark for analyzing quality weaknesses and relationships across dimensions. Building on this benchmark, we propose an execution-free quality evaluation baseline that predicts the multidimensional quality profiles of unseen skills by retrieving and leveraging similar reference skills, without executing the skills under evaluation. This reduces token consumption and evaluation time, while preserving consistency with the full assessment. Furthermore, SkillCT-guided optimization improves downstream task success and execution efficiency, demonstrating its value for skill refinement. These findings highlight the effectiveness of SkillCT in supporting scalable skill assessment, optimization, and repository maintenance.

0.11

Verification-Guided Retrosynthesis: Reaction Feedback and Evidence Timing

cs.AI ·2026-10-28 ·Hongxuan Shi, Hankz Hankui Zhuo
Abstract · 12 pages, including appendices

Retrosynthesis planning seeks synthesis routes from purchasable starting materials to a target molecule. Search over predicted reactions can connect a target to available materials, yet an infeasible reaction can invalidate the resulting route. Evaluating route quality only after search has finished cannot correct errors made during search. We introduce Verification-Guided Reflow (VGR), incorporating reaction verification as a decision action in AND/OR search. Verification results are represented as binary verdicts; queries are scheduled from the search structure, and acquired verdicts exclude reactions or adjust search priorities. We define verified solvability as the fraction of targets for which a complete route accepted stepwise by the verifier is found. On Retro190, feeding verification results into search increases verified solvability from 51.1% to 62.6% relative to route evaluation without feedback. Further controlled experiments show that earlier verification and immediate prioritization of accepted reactions each improve route discovery, yet combining them performs worse than either change alone. We therefore propose early pruning and delayed guidance: rejected reactions are pruned immediately, while accepted reactions gain search priority when processed as part of a complete candidate route. In a separate comparison with the verification scheduling rule fixed, this strategy increases verified solvability from 61.1% with immediate guidance to 71.1%, using fewer proposal calls and unique verifier queries.

0.10

Think Before You Act: Decision Gate for LLM Agents

cs.AI ·2026-10-28 ·Iftach Shoham, Yuval David, Offer Akrabi, Erez Bilgory, Ido Levy +3 more
Abstract

LLM agents increasingly perform consequential actions through external tools, yet their proposed actions are often executed without independent verification of policy compliance or task completion. We introduce JevAct, a runtime verification framework that intercepts selected agent actions before execution and enables feedback-driven correction. JevAct compiles domain policies into executable state checks and structured questions evaluated by a prefill-only decision model, which produces probability-based judgments without relying on LLM generation. By combining deterministic verification of tool-state facts with probabilistic assessment of conversational context, the framework identifies potentially invalid actions and premature task termination, providing targeted feedback for bounded revision. JevAct operates independently of the underlying agent model and requires no modifications to its parameters or tools. We evaluate JevAct across eleven LLM agents on $τ^2$-bench Airline and Telecom tasks and BFCL. Results show improvements in agent reliability across multiple models and benchmarks, with particularly consistent gains on Telecom and BFCL, while performance on Airline is more mixed. Overall, these findings suggest that combining structured decision models with executable verification can improve the reliability of tool-using LLM agents.

0.10

Discovering Jailbreaks via Curriculum of Models

cs.LG cs.AI ·2026-10-28 ·Michał Burzyński, Noah T. Rossi, Shomit Basu, Andy Arditi
Abstract

Automated jailbreak discovery aims to find prompts that elicit prohibited behaviors from safety-trained language models, helping developers identify vulnerabilities before deployment. One class of approaches uses reinforcement learning to train an investigator model to generate prompts that elicit a specified behavior from a target model. However, training directly against a safety-trained target is difficult because nearly all prompts are refused, leaving the investigator with a sparse reward signal. Prior work addresses this with the propensity bound (PRBO), a proxy reward estimated by sampling responses from a more compliant target, and then adjusting for how likely those responses would have been under the original target. We introduce Curriculum of Models (CoM), an alternative method that trains the investigator against a sequence of modified versions of the target, progressing gradually from a version that readily exhibits the specified behavior to the original target. For jailbreak discovery, training begins against a target version that rarely refuses harmful requests, and advances to the next, harder version once the investigator reliably elicits the behavior from the current one. We evaluate CoM on elicitation tasks involving CBRN materials, explosives, and illicit drugs, using two open-weight targets, gemma-4-31b-it and gpt-oss-120b. We compare against PRBO and direct training on the original target without a curriculum. When the investigator is trained to elicit a single behavior, CoM finds successful jailbreaks more reliably than either baseline. When trained on multiple behaviors simultaneously, the CoM-trained investigator can elicit held-out behaviors, though PRBO-trained investigators achieve higher success rates. Our results suggest that a simple curriculum over target models offers a promising approach for addressing sparse rewards in rare-behavior elicitation.

0.10

Skillful and realistic precipitation forecasting with reinforced diffusion

cs.LG physics.ao-ph ·2026-10-28 ·Ozan Sener, Anand Avati, David Hafner, Jierui Lin, Laura Manduchi +5 more
Abstract

Accurate precipitation forecasting remains elusive as physics-based models struggle to resolve mesoscale convection and machine learning often yields physically inconsistent, blurred results that consistently underestimate extreme events. We present FitzRoy, the first system that simultaneously achieves state-of-the-art predictive skill and physical realism over the continental United States. Through a reinforced diffusion framework, where a conditional discrete diffusion model is refined via reinforcement learning, FitzRoy effectively resolves the long-standing trade-off between pointwise accuracy and structural fidelity. Across lead times up to 24 hours, FitzRoy consistently matches or exceeds state-of-the-art models while preserving structural realism and capturing extreme events. Its low-latency inference further enables deployment in real-time operational settings. In case studies such as the 2025 Texas Hill Country floods and the 2026 Missouri floods, FitzRoy signals record-breaking accumulations that existing models largely missed. By bridging the gap between realism and skill, FitzRoy establishes a new frontier for high-fidelity precipitation forecasting.

0.10

Convex-SKM-Net: Low-Latency Block Halfspace Corrections for Neural Convex Constraint Satisfaction

cs.LG cs.AImath.OC ·2026-10-28 ·Haoyu Zhu, Xiaochen Zhang, Qingchun Hou
benchmarktheory
Abstract

Fast inference and strict constraint satisfaction are essential for learning-based decision-making. Existing methods often rely on constraint-specific projections or feasible-set parameterizations, while iterative repair can be costly. We introduce Convex-SKM-Net, a convex constraint-satisfaction layer that forms linearization halfspaces from residual and (sub)gradient oracles without constraint-specific inequality projectors. It applies fixed-depth dual projected-gradient descent (dual PGD) to minimum-displacement block problems using GPU-friendly matrix operations. We jointly learn the predictor and a positive-definite mobility, adapting the correction geometry to improve objective quality. We prove convergence of the outer iterates at fixed dual depth and linear convergence in distance under Slater's condition and bounded subgradients. All our test outputs on three second-order-cone model-predictive control (SOC-MPC) horizons and six power-system convex relaxations satisfy a $10^{-3}$ feasibility tolerance. Comparisons with the latest constraint-satisfaction methods show inference speedups of up to $144.3\times$ on SOC-MPC and $7.7\times$ on a convex quadratically constrained quadratic programming (QCQP) benchmark with joint training. On QCQP, the mean objective gap is $0.012\%$, with full test feasibility at $10^{-4}$. Learning mobility reduces the five-seed affine-task mean gap by $22.4\%$ and improves gaps in matched power-system comparisons.

0.10

Optimal Regret for Communication-Efficient Decentralized Online Learning

cs.LG ·2026-10-28 ·Sifan Yang, Wenhao Yang, Wei Jiang, Yuanyu Wan, Lijun Zhang
Abstract

In this paper, we investigate decentralized online convex optimization with compressed communication, where $n$ learners over a network jointly minimize a sequence of global loss functions based only on their local losses and compressed messages from their neighbors. The best known regret bounds for this problem are $O(ω^{-1/2}nρ^{-1}\sqrt{T\ln n})$ and $O(ω^{-1}nρ^{-2}\ln n\ln T)$ for convex and strongly convex loss functions, respectively, where $ρ\in(0,1)$ is the spectral gap of the communication matrix and $ω\in(0,1]$ is the compression quality factor. Compared with the corresponding lower bounds of $Ω(ω^{-1/2}nρ^{-1/4}\sqrt{T})$ and $Ω(ω^{-1}nρ^{-1/2}\ln T)$, these upper bounds exhibit polynomial gaps in their dependence on $ρ^{-1}$. To close these gaps, we first develop Ascent-gossip, which combines an online compression strategy with a Chebyshev recursion and attains a faster consensus rate than existing compressed gossip methods. Furthermore, by adopting Ascent-gossip, we propose an online algorithm that achieves optimal regret bounds of $\widetilde O(ω^{-1/2}nρ^{-1/4}\sqrt{T})$ and $\widetilde O(ω^{-1}nρ^{-1/2}\ln T)$ for convex and strongly convex loss functions, respectively. Up to logarithmic factors in $n$ and $ρ^{-1}$, these results match the lower bounds in terms of $T$, $n$, $ρ$, and $ω$.

0.10

Rubricon: Automated Rubric Generation and Refinement for LLM Evaluation and Training

cs.AI ·2026-10-28 ·Ulyana Tkachenko, Hui Wen Goh, Guram Gogia, Jonas Mueller
Abstract

Rubrics enable reward assignment on tasks without verifiable answers, but are expensive to manually produce from scratch. We present Rubricon, a pipeline that generates discriminative rubrics without requiring human preference labels, rubric annotation, or generator-model training. Using any LLM, Rubricon samples multiple rubrics, selects the most discriminative candidate, adds criteria to capture overlooked requirements or response flaws, and prunes criteria that are misaligned or shallow. Alongside LLM self-critique and test-time scaling, Rubricon uses observed criterion pass rates on example model responses to guide refinement. Across five professional, consumer, clinical, and general-use LLM benchmarks, Rubricon produces rubrics whose scores of frontier LLM responses exhibit 9% greater alignment with labor-intensive expert-authored rubrics, compared to the best alternative synthetic rubric generation method among RRD, QSRG, OpenRubrics, and agentic generation via harnesses like Codex. On HealthBench, post-training a Qwen3.5-9B model with Rubricon rubrics yields 4.5% better performance than training with rubrics from the strongest alternative synthetic rubric generator, RRD.

0.10

Rethinking LLM Jailbreak Evaluation: Continuous Risk Landscapes Beyond Binary Outcomes

cs.LG cs.CL ·2026-10-28 ·Yukai Zhou, Wenjie Wang
Abstract

A binary success indicator is a natural representation for discriminative tasks such as image classification under adversarial perturbations, where each input is mapped to a single prediction. However, generative tasks are fundamentally different: an LLM jailbreak prompt induces a distribution over open-ended trajectories, while a sampled response reveals only one path. Yet LLM jailbreak evaluation typically retains the binary response indicator underlying attack success rate, obscuring the surrounding structure of possible generations. We introduce a continuous, context-conditioned risk landscape over first-token branches, estimated by fixing candidate openings and repeatedly sampling their continuations. This continuous risk-landscape perspective reveals two complementary properties: current decoder exposure and safe rerouting capacity, which have been obscured by binary-indicator-based ASR. The resulting representation also provides a common basis for comparing clean prompts, attacked variants, and attack strength against a known adaptive defense. Across multiple target models and jailbreak attacks, we uncover risky branches beneath apparently safe outputs, identify qualitatively different ways attacks reshape the model's opening distribution, and show that attacks strongest before a defense do not necessarily remain the strongest after the adaptive defense reacts. These findings recast jailbreaks as deformations of the surrounding continuation-risk landscape, establishing a more informative foundation for understanding jailbreak behavior and developing safer language models.

0.10

CALM: Conflict-Aware Lifecycle Memory Governance for Long-Horizon Language Agents

cs.AI ·2026-10-28 ·Chang Li, Xianhao Ou, Xinyan Ma, Hao Li, Haoyuan Xu +5 more
Abstract

A memory can be outdated for present-day reasoning yet indispensable for reconstructing the past. Relevance-based retrieval overlooks this tension: semantic relevance, evidence-grounded validity, and query-specific eligibility are distinct. We propose CALM, a conflict-aware lifecycle memory governance framework that decouples persistent memory preservation from query-time admission. CALM converts heterogeneous records into evidence-grounded memory cells, updates validity and lifecycle states using support, conflict, and dependency signals, and admits query-appropriate evidence through training-free or learned gating informed by temporal applicability. Across LongMemEval-S and LoCoMo with Qwen and GPT backbones, CALM outperforms state-of-the-art baselines, achieving the best overall accuracy. Compared with simple filtering and reranking, query-conditioned admission further improves evidence selection. Conflict-blocking and robustness analyses show CALM suppresses erroneous retrieved memories while retaining useful evidence. Historical-reconstruction experiments demonstrate that preserving superseded records supports accurate earlier-state reasoning without compromising current-state answers.

0.09

Adversarial Self-Distillation

cs.LG ·2026-10-28 ·Mohammad Sadegh Akhondzadeh, Soroush H. Zargarbashi, Aleksandar Bojchevski
Abstract

Even though current safety-aligned language models are able to refuse harmful requests, their guard can still be bypassed by adversarial perturbations. As in other domains, adversarial training is considered a standard defense. However, due to the autoregressive nature of language generation, supervised finetuning on template refusals does not necessarily teach the model to refuse. Evidence of this is that the adversarially trained model still complies when the attacker injects a compliant prefill. We propose Adversarial Self-Distillation (ASD), which extends a model's own safe behavior to the space of adversarial perturbations. The teacher (the model itself given the clean prompt) can refuse the harmful request by default, or with the help of an additional instruction in its context. ASD distills this knowledge to the student (which is the same model under attack) at the prefixes that the student visits (due to on-policy sampling). These potentially include compliant openings. ASD achieves lower attack success than state-of-the-art defenses with little loss in multiple-choice accuracy, and it becomes able to recover a refusal even after a compliant opening.

0.09

Rethinking GFlowNet Training: Negative Reinforce Dynamics

cs.LG cs.AI ·2026-10-28 ·Pedro Dall'Antonia, Daniel Csillag, Diego Mesquita
Abstract · 22 pages, 9 figures, 3 tables

Generative Flow Networks (GFlowNets) are models that learn to sample discrete objects proportional to a target reward. They are trained by minimizing balance errors over sampled trajectories, which inherently ties correctness to exploration: if a region is not visited, balance cannot be enforced there. Adaptive exploration is therefore crucial, yet standard training relies on a semi-gradient update that treats the sampling distribution as fixed at each optimization step. On the other hand, the full gradient reveals a second mechanism that reduces the observed errors in an anti-exploratory manner. In this paper, we establish a connection between the full gradient and min-min dynamics between the GFlowNet and a decoupled sampler, showing that the sampler reduces observed violations by avoiding high-error regions. This insight motivates a bridge between two-player games and single-policy dynamics, which underpins the design of our proposed method, Negative Reinforce Dynamics (NRD). By tying the parameters of an adversarial game together, we recover a single-policy dynamic that extends exploration while maintaining correctness. In standard benchmarks, our method outperforms existing approaches in both convergence speed and exploration efficiency, without training a separate teacher or auxiliary exploration policy.

0.09

Predict the Reward, Not the World: World Reward Models for ML Agents

cs.AI cs.CL ·2026-10-28 ·Yuxiao Lee, Zherui Yang, Fan Liu, Hao Liu
positiontheory
Abstract

Search-based machine-learning (ML) agents improve a solution by writing candidate programs, executing them, and keeping the best. Execution gives the search reliable feedback, but a single run can take hours, and most runs end in programs the agent discards. Recent agents try to save these runs by letting a large language model act as a world model that predicts how each program would perform. However, a wrong prediction feeds the search false information. We argue that an agent does not need to predict the outcome of execution, only the decision that execution would support. We propose World Reward Models (WRMs), which learn from real execution results to choose directly which candidate will perform better, without predicting any intermediate outcome. We prove that a selector which predicts execution from the same inputs cannot outperform the best direct selector, that predicted outcomes add an error which can make a likelihood-trained world model prefer the worse program on every instance, and that cross-entropy on executed winners drives a WRM toward the optimum. Our WRM, ForeCritic, adapts a 9B typed decision model to 5k executed program pairs. On 12,438 held-out pairs from 26 ML tasks, it picks the better program 79.4% of the time, while general-purpose LLM judges reach 53--62%, and its probabilities are calibrated. Inside AIDE, ML-Master, and ForeAgent, it cuts mean task time by 4.4--7.5 times while keeping their mean scores within 0.01 or raising them.

0.08

On the Generalization Gap of Streaming Federated Learning with Markovian Data

cs.LG cs.ITstat.ML ·2026-10-28 ·Masoud Kavian, Milad Sefidgaran, Youssef Attia El Hili, Malik Tiomoko, Abdellatif Zaidi
Abstract · 78 Pages, 5 Figures

Streaming federated learning (SFL) allows clients to train a shared model from continuously arriving data without retaining large local datasets. Such streams often exhibit temporal dependence, yet its effect on generalization remains poorly understood. In this paper we study the expected generalization gap of the aggregated model when client data follow Markov processes, using information-theoretic and rate-distortion techniques. Our framework accommodates heterogeneous Markovian dynamics across clients and distinguishes temporal dependence from information acquired during local learning. We establish bounds under cross-entropy loss, specializing them to Linear Next-Token Prediction (LNTP) and Self-Attention Next-Token Prediction (SANTP). These results quantify the effects of client count, communication rounds, sample size, data dimension, mixing time, and context length. In particular, they reveal a communication--generalization tradeoff: although frequent aggregation may help optimization, it can weaken generalization guarantees. They also suggest a potential advantage of distributed over centralized streaming learning under a fixed total data budget. Temporal dependence enters through the square root of the mixing time. By analyzing the history-state process, we obtain refined bounds with improved context-length dependence under suitable mixing conditions. Experiments on GPT-style language-model pretraining with TinyStories and real-world time-series forecasting illustrate how collaboration and communication frequency influence generalization under temporal dependence

0.08

AtomSteer: Test-Time Constraint Steering for Atomistic Generative Models

cs.LG cs.AI ·2026-10-28 ·Pengfei Cai, Katharina S. C. Jäger, Utkarsh Utkarsh, Nofit Segal, Bowen Deng +2 more
Abstract

Diffusion and flow matching models offer scalable approaches to generating crystalline materials, molecular conformers, and protein-ligand complexes. However, satisfying crystallographic symmetry, structural geometry, and accurate stereochemistry remains challenging. Satisfying these requirements calls for corrections that preserve the geometry already captured by the pretrained model. In this work, we propose AtomSteer, a test-time inference framework for steering pretrained atomistic generative models with structural constraints. During sampling, AtomSteer corrects intermediate structure predictions with bounded local projections, graph-rigid stereochemical transformations, and projections over periodic symmetries, accepting updates that improve constraint satisfaction while limiting changes to the rest of the structure. We demonstrate AtomSteer across three pretrained backbones, FlowMM, ET-Flow, and Boltz-2, spanning de novo crystal generation and crystal structure prediction, molecular conformer generation, and protein-ligand complex structure prediction. AtomSteer improves space-group distribution coverage in de novo crystal generation and improves target space-group recovery in crystal structure prediction. AtomSteer improves molecular validity and E/Z stereochemical accuracy while retaining conformer ensemble fidelity. For protein-ligand complexes, AtomSteer reduces ligand E/Z errors relative to experimental reference structures while achieving competitive docking success and PoseBusters validity. Our method provides a unified framework for adding chemical and crystallographic control to existing atomistic generative models without retraining or architectural changes.

0.08

Curriculum-Based Deep Reinforcement Learning for Robust Autonomous Rocket Landing under Atmospheric Uncertainty

cs.AI ·2026-10-28 ·Atin Chowdhury, Piyush Joshi
Abstract · 21 pages, 12 figures

Autonomous far-range rocket landing is a challenging control problem because navigation, descent-rate regulation, attitude management, and fuel use must be coordinated in the presence of uncertain atmospheric and vehicle dynamics. This paper presents a curriculum-driven proximal policy optimization (PPO) framework for three-dimensional rocket landing in a stochastic environment incorporating altitude-dependent aerodynamic drag, wind shear, temporally correlated turbulence, discrete gusts, uncertain propulsion and aerodynamic parameters, thrust-vector misalignment, engine lag, actuator-gain variation, noisy observations, and command latency. Learning is organised into far-range approach, soft descent, and integrated landing stages. Policy transfer, critic warm-up after reward changes, rehearsal of earlier tasks, and progressive expansion of weather severity are used to support learning across these coupled phases. Across 100,000 evaluation episodes, the policy achieves 96,656 successful landings, corresponding to a success rate of 96.656\% and a 95\% Wilson confidence interval of 96.543--96.766\%. Success decreases from 98.84\% under zero-disturbance conditions to 88.07\% in the highest weather-severity band. Successful touchdowns exhibit small miss distances and low residual horizontal and vertical velocities. An exploratory logistic-regression analysis identifies mean wind speed as the strongest measured correlate of success, with initial distance, initial altitude, initial vertical velocity, and wind-shear exponent also associated with outcomes. The results quantify both the effectiveness of the learned policy over the evaluated disturbance distribution and the degradation that occurs under severe compound atmospheric conditions.

0.08

Aligning Autoresearch: Co-Improvement between Agents and Humans

cs.AI ·2026-10-28 ·Tianjian Li, Chuanyang Jin, Seungone Kim, Tong Chen, John Hoffman +8 more
benchmark
Abstract

Autoresearch agents can now attempt long-horizon AI research end-to-end, i.e., self-improvement. However, best results are still obtained by research agents interacting with human researchers, and improving AI's ability to perform this co-research will only reinforce this effect---a process termed co-improvement (Weston and Foerster, 2025). Further, aligning with human expectations is likely a safer path for research agents, especially given recent agent reward hacking phenomena (OpenAI, 2026b). In this work we address these issues in four ways: (i) we empirically study the role of humans in directing autoresearch agents, focusing on the domain of benchmark creation, and find humans still play a pivotal role particularly in the ideation phase; (ii) we also develop a benchmark, HackBench, that shows current autoresearch agents are not aligned and are unable to achieve hack localization sufficiently; (iii) we present a method to train agents for improved human interaction by using human simulators as a proxy during reinforcement learning (RL); and finally (iv) we develop a method to improve model text generation, and hence agent communication, by RL with rubrics aligned with human experts. Together, our training and evaluation recipes provide a path toward more capable and aligned (co-)research agents.

0.07

Efficient Agentic Audio-Visual Question Answering through Toolset Pruning and Modality Consolidation

cs.AI cs.CLeess.AS ·2026-10-28 ·Yi Wang, Tianzi Wang, Peter Bell
Abstract

Fine-grained audio-visual question answering (AVQA) requires models to resolve complex cross-modal correspondences and temporally grounding from sparsely distributed evidence. Omni-modal agents, which autonomously orchestrate perception tools to actively collect evidence, achieve strong performance, but it remains unclear whether active perception from both audio and visual modalities depends on broad specialized toolsets, especially audio or visual modality-specific tools. Such systems may also introduce long execution trajectories and substantial inference cost. We investigate whether comparable performance can be retained with a compact workflow exposing only essential perceptual capabilities. Starting from the specialized-tool OmniAgent, we conduct staged controlled experiments that prune the toolset and progressively consolidate modality-specific tools into omni-modal tools, motivating a two-tool agent comprising only Omni-modal captioning and Omni-modal QA. Across three widely used AVQA benchmarks, our compact agent achieves comparable or better accuracy than the specialized-tool agent while reducing inference cost by around 70%.

0.07

Accelerating Nonlinear MPC by Behavior Cloning with Separated Timescales: Application to Nuclear Reactor Load-Following

cs.AI cs.LGmath.OC ·2026-10-28 ·Perceval Beja-Battais, Alain Grosset{ê}te, Nicolas Vayatis
benchmark
Abstract

Behavior Cloning (BC) of expert controllers for multiscale dynamical systems is challenging: state variables evolving at widely separated timescales induce highly nonlinear couplings that a shared latent space may fail to adequately represent, degrading closed-loop generalization. We propose Structured Representation Learning for Behavior Cloning (SRLBC), an architecture designed to leverage the physical knowledge of multiscale systems in which slow state variables, fast (stiff) state variables, and predictable exogenous inputs evolve at different timescales. On 500 long-horizon roll-outs of a pressurized water reactor (PWR) load-following benchmark -a high-dimensional, stiff industrial system with challenging safety constraints -and across 20 training seeds, separated embeddings systematically outperform shared embedding baselines on cumulative cost, imitation accuracy and constraint satisfaction. The resulting standalone policies approach NMPC closed-loop performance at a computational cost 15 times lower on average. Since standalone policies still leave a few percent of trajectories infeasible, we embed the learned policies in a safety-augmented approximate NMPC framework, in which the optimizer is only called when the predicted input sequence is not certified as safe. This yields controllers still 6.8 times faster than the NMPC controller on average, with a feasibility (i.e., constraint satisfaction) of at least 99.8% for all architectures, SRLBC retaining the lowest closed-loop cost. Under distribution shift, standalone BC policies degrade sharply: SRLBC degrades the least, and once safety-augmented, it retains a feasibility above 98% with the lowest closed-loop cost. Beyond the quantitative results, Jacobian-based sensitivity analysis shows that the resulting controller relies on a sparse and less redundant set of latent features, each attributable to a single group of physical variables, and aligned with the physical timescale decomposition of the system.

0.06

A Separation Principle for Adversarial Attacks and Defense of Graph Neural Networks

cs.LG ·2026-10-28 ·Farid Mafi, Saber Jafarpour, Mohammad Pirani
Abstract

Adversarial attacks on graph neural networks (GNNs) have been studied as a model-centric optimization problem driven by gradients, learned policies, or meta-learning, with the graph treated as a passive substrate. This view overlooks a defining feature of GNNs: their predictions are produced by structured message passing along the graph, so their adversarial vulnerabilities are encoded as much in the geometry of the graph as in the learned weights. We develop a structure-driven framework for feature attacks in which the attack is derived from the spectral properties of the normalized adjacency operator. Its central result is a separation principle: under a natural relaxation of the attack objective, the optimal feature perturbation is rank one and factorizes into a graph-dependent component (degree-proportional spatial allocation, depending only on the graph) and a weight-dependent component (top singular vector of the composed weights, depending only on the weights). Node degree thus emerges as the fundamental graph quantity governing adversarial vulnerability to feature perturbations. The resulting attack, which we call SepAttack, requires no gradient computation, carries analytical guarantees, achieves competitive or superior performance against established baselines such as Nettack and IG-Attack with up to ${\sim}70\times$ speedup in the targeted setting, and transfers near-perfectly across architectures on continuous-feature graphs because its graph-dependent component is architecture-invariant by construction. We further derive a separation-principle adversarial training procedure (SepAT) directly from the closed-form worst-case attack and report robustness gains on binary-feature benchmarks.

0.04

Jailbreak Connectivity: Towards Diverse, Transferable, and Universal Multimodal Large Language Model Jailbreak

cs.LG cs.AIcs.CR ·2026-10-28 ·Xu Zhang, Ziqing Hu, Shuo Han, Ren Wang
Abstract

Multimodal large language models (MLLMs) are vulnerable to visual jailbreak attacks that bypass safety safeguards. Existing methods typically optimize isolated adversarial examples, limiting attack diversity, transferability, and generalization. We propose Jailbreak Connectivity (JC), a framework based on the observation that effective adversarial inputs can be connected through low-loss regions in image space. JC optimizes continuous adversarial paths: path sampling yields diverse jailbreak images, lightweight surrogate classifiers guide cross-model transfer, and a distribution-level objective enables universal attacks across harmful queries. On MM-SafetyBench, JC achieves average attack success rates of 84.01%, 85.77%, and 97.95% on Qwen3-VL-8B-Instruct, InternVL3.5-8B, and Gemma-3-12B-IT, respectively, exceeding the strongest baselines by up to 68.36 percentage points. In black-box evaluations, JC improves attack success rates over FigStep by 24.69 and 27.34 percentage points on GPT-5.6 and Gemini 3.1 Pro, respectively. These findings reveal structured vulnerabilities in MLLM safety alignment and inform more comprehensive security evaluation.

0.04

MASS: Multi-mask Selective Self-Distillation for Leakage-Resilient Reasoning

cs.LG ·2026-10-28 ·Xiuqi Liu, Oveys Delafrooz Noroozi, Upamanyu Madhow
Abstract

On-policy self-distillation enables the same model to act as student and teacher, with dense token-level supervision obtained by the teacher being given access to privileged information. However, conditioning the teacher on a single reference can concentrate its predictions on one reasoning trajectory and limit transfer to the student. We propose Multi-Mask Selective Self-Distillation (MASS), which approximates the unavailable marginal over diverse reference realizations using multiple independently masked views of one verified solution. MASS aggregates these complementary teacher predictions, selectively transfers reliable token-level corrections, and adaptively focuses masking on influential reference spans. Across mathematical reasoning benchmarks and model families, MASS consistently outperforms strong self-distillation baselines.

0.03

Language, Vision & Media 129

TreemapMix: Dirichlet-Controlled Multi-Image Augmentation for Probability and Ordinal Supervision

cs.CV ·2026-10-28 ·Ejafa Bassam, Konstantin Garbers, Yingsheng Geng, Dalton Jens, Jiaxu Qian +1 more
top venuecode/data
Abstract · Accepted at NeurIPS 2026. 40 pages including references, appendices, and checklist. Ejafa Bassam and Konstantin Garbers contributed equally

Data augmentation improves visual recognition by exposing models to synthetic training examples that encourage generalization beyond the observed data. Mixed-sample methods extend this idea by combining multiple images. However, many common approaches either mix only pairs of images or rely on fixed small layouts, limiting control over source count, visible area proportions, and induced supervision. We propose TreemapMix, a distribution-controlled multi-image augmentation that samples source weights from a Dirichlet distribution and assigns them to area-proportional regions using a treemap partition. The resulting known region areas support two complementary objectives. TreemapMix with soft cross-entropy (SCE) treats areas as soft class-probability targets, while TreemapMix with a Plackett-Luce (PL) objective converts areas into ordinal supervision. On ImageNet-1K under matched training recipes, TreemapMix-SCE achieves substantially lower calibration error than evaluated mixed-sample baselines, while TreemapMix-PL achieves the strongest Top-1 accuracy. Transfer experiments on object detection and instance segmentation suggest that TreemapMix-pretrained features remain useful beyond classification. These results show that controlled multi-image composition can expose the same region-area information as either probability-matching or rank-matching supervision. Code is available at https://github.com/Ejafa/treemapmix.

0.56

RoSAMDepth: Robust Self-supervised Depth Estimation Leveraging Segment Anything Model

cs.CV ·2026-10-28 ·Xuanang Gao, Zhiwei Ning, Gengming Zhang, Jiaxi Cao, Runze Yang +4 more
top venuecode/data
Abstract · 13 pages, 8 figures, including supplementary material. Published in CVPR 2026

Robust depth estimation aims to maintain high-quality depths across diverse conditions. However, most existing methods estimate depth without taking into account the object-level information. As a result, the predicted depth may easily deviate within objects and become blurred under adverse conditions. To overcome this weakness, we propose RoSAMDepth, a novel framework that can assist robust self-supervised depth estimation in leveraging rich and diverse object-level priors from the Segment Anything Model (SAM). We focus on incorporating object-level information across three key aspects: a segment-guided representation contrasting method that injects object-level awareness into the feature representation space; an adaptive regional outlier masking strategy combined with a regional Gaussian likelihood loss that enforces regional depth smoothness; and an object-level reliability estimation strategy that mitigates the influence of unreliable supervision. Extensive experiments across multiple datasets and diverse weather conditions demonstrate that our method produces sharper, more accurate depth predictions, consistently outperforming state-of-the-art methods. Our code is available at https://github.com/xagao/RoSAMDepth.

0.55

Who Called? V33DA: A Physically Verified Multimodal Benchmark for Vocal Attribution in Zebra Finch Groups

cs.SD cs.CVcs.LGq-bio.QM ·2026-10-28 ·Maris Basha, Yuhang Wang, Xiaoran Chen, Longbiao Cheng, Luca Yapura +3 more
top venuebenchmarkcode/data
Abstract · Accepted at NeurIPS 2026 (Evaluations and Datasets Track). 43 pages, 3 figures. Dataset: https://huggingface.co/datasets/songbirdini/v33da ;…

Deciding which member of an animal group produced a vocalization and where that vocalization was produced are central to studying social communication but are rarely evaluated with physically verified ground truth. Existing resources either localize sounds without identifying the caller, or recognize individuals from isolated vocalizations without reliance on the 3D candidate geometry. We close this gap with V33DA, a dataset-centered multimodal benchmark for spatial vocal attribution in social zebra finches: it comprises 33,625 verified events coupling 5-channel audio with three camera views, calibrated 3D pose for every visible candidate, and per-bird radio telemetry across 10 individuals and three experiments. Caller identity is verified from an on-body accelerometer-derived vibration signal; this channel is withheld from benchmark models and used only by an on-body reference ceiling. Given a detected vocalization and the visible candidate birds, the task is to select the caller. We evaluate it under three regimes: session-disjoint testing on familiar birds from held-out recording dates, held-out-experiment transfer to a new cohort, and leave-one-individual-out evaluation on a bird whose vocalizations are held out from training. Reference methods expose a consistent gap between in-domain accuracy and transfer to unseen individuals: methods that can exploit familiar vocal identity perform well on known callers but degrade sharply on unseen ones, whereas candidate-aware methods preserving explicit spatial reasoning transfer substantially better. The on-body reference stays near-perfect across regimes, confirming the internal consistency of the accelerometer-verified labels. We release V33DA with fixed protocols and code for the evaluated methods, and V33DA++, an unevaluated companion set of verified overlapping-call events and +/-2 s context windows.

0.53

SSQA: Sibling-Selective Quadtree Attention for Hierarchical Modeling in Perception Tasks

cs.CV ·2026-10-28 ·Yufan Chen, Arnav Bali, Angela Liu, Laura Zheng, Ming C. Lin
top venue
Abstract · 8 pages, 6 figures, 4 tables. Published at ICRA 2026 (oral)

Perception tasks for navigation in robotics, including aerial platforms such as drones and autonomous driving systems, are inherently structured. Drone-mounted cameras typically capture sky above, terrain below, and obstacles or man-made structures in between, while driving data often contains organized road layouts, lane markings, and surrounding agents. Motivated by these axis-aligned structural priors, we note that such information is typically more structured than in generic image tasks. We hypothesize that processing information in a quadtree-esque manner can not only model features effectively in a hierarchical manner, but also offers an efficient linear-time alternative to vanilla attention mechanisms, which run in quadratic time. In this paper, we propose Sibling-Selective Quadtree Attention (SSQA), which models image tokens hierarchically as a structured, full quadtree. We show analytical complexity analysis that guarantees linear-time feature modeling, in addition to empirical experiments comparing inference speeds with other popular modeling approaches, such as Mamba 2 and Quadtree Attention. Our results, benchmarked across several tasks, show that we achieve results at least as good, if not notably better, as others at a fraction of the computational costs.

0.50

Cross-Primitive Consistency in Jev

cs.CV cs.CL ·2026-10-28 ·Xu Liu, Guikun Chen, Kanzhi Wu, Wenguan Wang
benchmarkcode/data
Abstract

Jev enables efficient decision-making with large language models (LLMs) through typed querying primitives, allowing the same semantic state to be expressed in different formats, such as binary and multiple-choice queries. However, the consistency of decisions across primitives has not been systematically evaluated. This work introduces PrimaCons, a four-axis benchmark for assessing cross-primitive consistency while preserving the semantics of the query. Evaluations across three Jev sizes reveal consistency violations across all four axes. Our analysis shows that these violations follow systematic patterns and that cross-primitive discrepancies predict decision failure with a ROC-AUC of 0.91. To mitigate these inconsistencies, we propose the Cross-Primitive Residual Calibrator (CPRC), which attaches a lightweight residual network to a frozen backbone and jointly optimizes predictive accuracy and cross-primitive consistency. Under a leave-one-domain-out protocol, CPRC reduces cross-primitive inconsistency by 58.57% and improves accuracy by +2.97%., demonstrating its effectiveness in both in-domain and out-of-domain settings. Our findings establish cross-primitive consistency as a key dimension for evaluating Jev and provide a practical approach to improving the reliability of its decisions. Code and benchmark is available at https://github.com/67L1/PrimaCons-CRPR.

0.48

Self-Consistency Geometry Injection for Multi-view Consistent Image Generation

cs.CV cs.GR ·2026-10-28 ·Yupeng Xiao, Yankai Wu, Jianjie Luo, Jiajie Lin, Yi Yu +1 more
publishedcode/data
Abstract · Accepted by ACCV 2026

Recent diffusion models have achieved impressive progress in single image to multi-view generation by synthesizing consistent novel views efficiently. Existing methods usually explicitly or implicitly couple geometry inference and texture synthesis, making generation vulnerable to inaccurate geometry prediction. Although several geometry-guided methods have explored decoupling, they usually treat predicted geometry as a fixed condition and do not explicitly model whether the geometry is reliable. We therefore propose Reliability-Guided Multi-View Generation (RGMV), which comprises a parameter-free Self-Consistency Geometry Gate (SCG) and a Reliability-aware Geometry Injection (RGI) module. Specifically, a geometry proposal generator produces multiple geometry predictions for the same reference image and target viewpoints under different stochastic sampling configurations. SCG measures their spatial agreement to derive a reliability map, which RGI uses to modulate the injection strength of geometry-conditioned residuals in the RGB synthesis model. Moreover, RGMV can adapt to different geometry proposal generators. As a result, RGMV preserves stronger geometric guidance in stable regions while attenuating unreliable geometry cues before they propagate into multi-view RGB generation. Extensive experiments show that reliability-aware geometry injection improves the geometric quality of multi-view generation. Code: https://github.com/WISLab-GDUT/RGMV.

0.47

TSM diff: A new Parallel-Tree-Traversal Algorithm for the Comparison of Music Score Files

cs.MM ·2026-10-28 ·Xiaobo Wang, Florent Jacquemard
published
Abstract

Music score files are more difficult to compare than their text-based equivalents, as their content is more complex and organised into hierarchical structures. This explains the lack of tools for processing music scores similar to the widely used old Unix diff command. We propose a new algorithm and tool for comparing digital music scores in various encodings, such as MusicXML or MEI, based on a graph-structured abstract representation called Tree Score Model (TSM). Given two scores, after conversion into TSM, our procedure calculates jointly: A list of differences, localised by time positions in the input scores; A time distance, cumulated duration of time-spans where scores differ; An edit distance, called surface distance, based on notational attributes not related to time. The two later measures are normalised and combined into a similarity distance between scores. Our procedure has been implemented in C++ and is available as a Python module. Our first experiments have shown a very good efficiency even on challenging data. As a proof of concept, we propose two demos based on different versions of some Mozart's piano sonata, and transcriptions of monophonic soli on the same jazz standards by different performers. https://tsm-diff-lbd-7a7e2a.gitlab.io/demo.html

0.46

The Quine Test: How Much of a Model's Self Survives Its Own Description?

cs.CL ·2026-10-28 ·Bibin Babu, Ivan P. Yamshchikov, Aleksei Shpilman
theorycode/data
Abstract · 32 pages, 3 figures, 12 tables. Code, data and preregistration files: https://github.com/bibinsees/quine-test

Frontier language models express coherent preferences, but behavioral evidence alone cannot say whether these belong to the model or to a character it portrays. Following Hofstadter's view of the self as a self-description, we ask whether a model's description of itself works as a portable specification of its preference profile, a quine test: can the text a model writes about itself regenerate its behavior elsewhere? Ten models from nine labs write self-descriptions; we place each one in the prompt of other models (receivers) and of fresh instances of the author, and measure how closely their choices on a 110-item preference battery match the author's. First, the tested native self-descriptions (native: the profile measured under a generic assistant prompt) show no average reconstruction gain across variations in length, framing, wording and prompt placement: on a 0 to 1 fidelity scale where a model's fidelity to a fresh instance of itself is 0.979 and to unrelated models about 0.89, the one-sided 95% upper bound on the pooled gain is below +0.01 in every tested system-prompt condition, and descriptions based on the author's observed choices do no better. Second, the channel itself is strong: explicitly listing the author's answers lets receivers reproduce its profile at fidelity 0.955 to 0.960 across three model grids, within 0.03 of same-model repeatability. Third, self-descriptions written under an explicit preference policy transfer the assigned preferences almost as well as the policy text itself. Fourth, reading its own self-description shifts even the author away from its baseline, modestly more than unrelated prose of the same length. For the preferences measured here, native self-descriptions influence behavior without reliably specifying how to reproduce it in another model. We release the pipeline, battery, raw call logs, and 198 model-written descriptions.

0.44

Predicting What Answer Engines Cite: Open Decision Models, Rerankers and a Distilled 4B Student on Ten Engines under Controlled Retrieval

cs.IR cs.AI ·2026-10-28 ·Elisha Bajemon, Andre-Louis Rochet
benchmarkcode/data
Abstract · 12 pages, 1 figure, 4 tables. Model: https://huggingface.co/citead/geo-citation-ranker-4b ; benchmark and code: https://huggingface.co/datas…

Generative engine optimization needs a scorer that ranks candidate pages by the share of an answer they will receive as citations. We measure pretrained rerankers, a hosted typed decision model (Jev), eight open decision models and distilled students on GEO-Bench queries answered by ten engines that receive the same candidate sources in their prompt. Where several engines answer the same query, they agree with one another at a within-group Spearman correlation of 0.704. Jev and two open 27B decision models, used without training, are the strongest scorers at 0.575 to 0.578. A Qwen3-Reranker-4B fine-tuned with low-rank adapters on teacher labels for other GEO-Bench queries, released as GEO Citation Ranker 4B, reaches 0.542 to 0.549 across three seeds, and its differences with the 27B models and with Jev are not distinguishable from zero on 229 queries. Most of the remaining distance between the student and Jev disappears when Jev scores one passage at a time, as the student does. Students trained on teacher labels and on observed engine citations reach the same level, and labels from Jev alone match the two-teacher target at 0.6B. We release the model, the benchmark, the predictions of every reported scorer and the code.

0.44

Interventional Training Reduces Causal Bypass in In-Context Introspection

cs.CL cs.AI ·2026-10-28 ·Zilu Tang, Jongin Kim, Kerem Sahin, David Atkinson, Aaron Mueller +1 more
Abstract · 9 pages, preprint

Introspection, the ability of a system to access and reason about its internal signals, is crucial for language model (LM) self-consistency and scalable interpretability. Prior work argues that LMs possess introspective capabilities, and that training (SFT) could enhance such behavior. However, most setups leave open the possibility of "causal bypass," in which the self-report relies on signals that are correlated with the self-report's subject, while being causally unrelated to it. Although high-level concepts are difficult to intervene on, lower-level signals (such as a single neuron activation) allow precise interventions. Understanding such signals can serve as a minimal prerequisite before understanding more complex introspection. In this study, we train hundreds of introspectors across three low-level signal types and eight LMs. We find that straightforward SFT training leads to causal bypass, and propose interventional training, an alternative where we train the model to report on counterfactually perturbed values. This method enables more accurate introspection of endogenous $\textit{and}$ injected signals, particularly in out-of-distribution (OOD) data. The mechanism generalizes across token positions and scales with more data. Our results demonstrate that while causally grounded introspection is possible, it is not guaranteed -- emphasizing the need to train on mechanisms, rather than behaviors alone.

0.43

Does Chain-of-Thought Reliance Follow a Scaling Law?Larger Models Follow Their Steps More, Independent of Post-Training

cs.CL cs.LG ·2026-10-28 ·Esther Xin
code/dataunusual
Abstract · 15 pages,5 figures, code and data in https://github.com/ethxin0011/cot-scale-reliance.cl

Monitoring a model's chain of thought (CoT) is useful for safety only if the final answer depends on the written steps. Prior measurements use natural tasks where true reasoning is unknown. We measure it against an exact ground truth on a synthetic box-swap task where every intermediate state is known. We replace one step inside a model's own chain with a wrong but valid state and check whether the answer follows the edit. We evaluate 25 open-weight models--Qwen3 (0.6B-14B, base and post-trained), Gemma 3 (1B-12B), Llama 3 (1B-8B), and 8 OLMo 2 checkpoints--on 7,200 items each. We find five results: (1) Reliance rises with size and saturates. In Qwen3, answer reliance rises from 16-18% at 0.6B to 97-100% at 14B; a logistic in log N fits well (R^2 = 0.89), and Gemma 3 and Llama 3 follow this trend. (2) Capability is the main driver. Across all models, reliance and step accuracy correlate at r = 0.99; equally capable models are equally reliant, regardless of family. (3) Size still matters at matched correctness. On problems solved correctly before the edit, reliance increases with size among capable models (rho = 0.80, p = 0.002). (4) Post-training does not change reliance. Base and post-trained Qwen3 differ by at most 0.03 from 4B upward; a large gap at 1.7B matches a large capability gap. (5) No model up to 14B solves the task internally. Without written steps, every capable model is at chance once a problem has four or more swaps. OLMo 2 checkpoints never become capable, making the training-token axis a limitation. On tasks requiring step-by-step computation, scale makes the chain of thought more load-bearing, supporting CoT monitoring in this regime.

0.41

Anchored Fusion for Automated Airway Segmentation in Pediatric and Adult CT

eess.IV cs.CV ·2026-10-28 ·Huseyin Umut Isik, Ege Mert Balcik, Duygu Atasoy, Asutosh Sahu, Seyda Ertekin +1 more
code/dataunusual
Abstract · 9 pages, 3 figures, 2 tables. Code, weights and web application: https://github.com/quin-med-harvard-edu/airway-seg

Airway segmentation from chest CT enables branch-level measurement of airway size and growth, but peripheral bronchi remain hard to segment, particularly in children, whose airways are smaller and for whom few annotated scans exist. Different models recover complementary branches of the airway tree. Their union improves overall tree coverage but also introduces disconnected non-airway objects. We propose anchored fusion (A-Fuse), which takes the voxel-wise union of TfeNet, a retrained MEDPSeg and a reimplemented AirMorph and retains the connected component with greatest overlap with the AirMorph mask. On 30 held-out adult CTs from the ATM'26 training set, A-Fuse reaches a Dice score of 93.24% and the highest branch detection rate among compared methods (92.05%). On nine pediatric scans, it reaches 94.15% Dice and 68.79% branch detection. A-Fuse retains the airway tree in all 39 test scans, whereas keeping the largest component selects a non-airway object in four. A-Fuse is deployed as an automated web application and we release all code, trained weights and an open-source ITK-SNAP extension for 3D tubular segmentation editing.

0.41

Report-Derived Labels, an Image Teacher and the Gap to Expert Labels in Knee MRI: An Analysis of One Entry to the RSNA 2026 Knee Abnormality Detection Challenge

eess.IV ·2026-10-28 ·Soheil Kooklan
benchmarkcode/data
Abstract · 24 pages, 10 figures, 14 tables. Code: https://github.com/soheilkooklan/rsna-knee-report-labels

In the RSNA 2026 Knee Abnormality Detection challenge, 58 of 4,407 training studies have expert labels for 12 findings, and the rest have only a radiology report. We compare training targets built from a language-model reading of the reports and from a public image model against two references: the expert labels and the 39,248 study-finding cells that the reports state as present or absent. The report labels and the out-of-fold predictions of the image model reach similar agreement with the expert labels (macro AUC 0.892 and 0.896) but fail on different findings. The report is better for ligament tears and patellofemoral osteoarthritis, which it never calls present when the experts do not; the image model is better for effusion, Baker's cyst, contusion, and synovitis, where the reports call the finding absent, leave it out, or call it present more often than the experts. Their equal mixture reaches 0.927 (gain over the report labels 0.035, 95% interval 0.019 to 0.052), also after removing the studies whose knee side was handled inconsistently. We did not observe a model-level gain from this better target: a 2.5D ResNet50 trained on the report labels alone scored 0.877 on the private leaderboard, after the deadline, against 0.872 for the same model trained on the mixture, and the out-of-fold advantage of the mixture is confounded by a change of code and by possible leakage through the image model's folds. A scanner prevalence prior raised agreement with the reports but not with the experts, so a report-derived reference can reward reporting habits. A rank correction for co-occurring findings gave small out-of-fold gains and no visible leaderboard effect. Our entry, a forked public ensemble with that correction, was displayed at rank 582 of 6,299 on the private leaderboard (0.943, a score shared by 287 teams). All results are exploratory; none of the models is meant for clinical use.

0.41

Beyond Region Memory: Local Point Tracking for Vocal Cord Localization in Video Laryngoscopy

cs.CV ·2026-10-28 ·Huayu Wang, Gyungmin Ko, Jung Heon Kim, Jia-Xian Jian, Dae Kon Kim +7 more
published
Abstract · Accepted by WACV 2027

Reliable vocal-cord localization in emergency video laryngoscopy remains visually challenging due to rapid tissue deformation, specular highlights, and instrument occlusions. We propose Local Point Tracking (LPT), a robust framework that combines a single-frame detector, local point propagation, and stateless mask refinement. LPT follows a localize first, snap second pipeline. First, a single-frame semantic detector generates sparse anchors, within which LPT initializes local query points (SAM-supported by default, with a regular grid as fallback). A point-propagation module tracks these points through detector-missed frames and recovers temporal bounding boxes from their geometry. Second, because point-derived boxes can be loose around the visible boundary, a box-snap step refines the selected box to the target boundary. This decomposition keeps temporal state in local point coordinates rather than a stateful mask or whole-region representation, while the spatial refinement module handles single-frame boundary precision without feeding mask errors back into temporal memory. Extensive evaluations on two distinct datasets demonstrate that our proposed LPT method achieves strong overall performance against existing baselines, with the best mAP_50, success AUC, and missing rate on both datasets. Furthermore, ablation studies confirm that point tracking primarily enhances temporal tracking robustness, while box-snap refinement effectively improves boundary tightness.

0.40

Don't Ask the LLM: Visual Token Pruning with Spectral Structure and Semantic Relevance

cs.CV ·2026-10-28 ·Intaek Shin, Geuntaek Lim, Sungjune Park, Yukyung Choi
publishedcode/data
Abstract · Accepted to WACV 2027. 19 pages (10 pages main paper, 9 pages supplementary material)

Large vision-language models (LVLMs) represent images and videos as long visual-token sequences, substantially increasing inference costs. These costs motivate visual-token pruning, whose effectiveness depends on both what signal is used for selection and when that signal becomes available. Our analysis reveals complementary limitations: visual-side [CLS] attention produces scattered, background-biased selections under aggressive compression, while stronger query-conditioned signals become available after substantial computation within the large language model (LLM). This raises a natural question: can structurally coherent, query-aware token selection be performed before consulting the LLM? We therefore propose ST-Pruner (Spectral-Textual Pruner), a training-free framework for pre-LLM visual-token selection. It combines attention-independent spectral saliency with inverted patch-text similarity through parameter-free multiplicative fusion. Across diverse LVLMs, ST-Pruner retains over 95% of full-token performance using at most 11.1% of visual tokens and achieves a 3.2x end-to-end speedup on LLaVA-NeXT. Code is available at https://github.com/sejong-rcv/ST-Pruner.

0.39

EPSBench: Benchmarking LLM Agents for Earnings Forecasting

cs.CL cs.AI ·2026-10-28 ·Yanbin Yin, Chuyu Xu, Liqin Ye, Jing Peng, Chao Zhang
benchmarktheory
Abstract

Quarterly earnings forecasting requires an agent to connect financial evidence to the income-statement line items that produce a company's reported earnings. We introduce EPSBench, a benchmark for studying this ability across 149 firm-quarters from 86 US companies. Agents read a fixed collection of filings, earnings-call transcripts, and news, then estimate company-specific income-statement drivers with cited evidence. A shared accounting calculator converts these estimates into GAAP diluted earnings per share (EPS), providing a common basis for comparing forecast assumptions. We evaluate six models under a common harness, assessing point forecast error relative to date-matched analyst consensus and checking whether their prediction intervals achieve the stated coverage. Three models show no detectable paired difference in point forecast error from consensus, and three have higher error. Across all six models, nominal 80% intervals contain reported EPS in only 37--70% of runs that return an interval, and most misses fall above the upper bound because the forecasts run low. The driver records trace most of the error gap between models to operating-income estimates.

0.39

AskStream: When and What to Ask Proactively under Uncertainty in Streaming Video Understanding

cs.CV cs.CL ·2026-10-28 ·Lu Wang, Zhuoran Jin, Longxiang Wang, Junhong Chen, Yupu Hao +6 more
benchmarkcode/data
Abstract · 42 pages, 9 figures. Project: https://www.askstream.top ; Code: https://github.com/wanglu-cs/AskStream ; Dataset: https://huggingface.co/dat…

Recent advances in multimodal large language models have increased interest in interaction paradigms for video understanding. The evolution from offline to streaming and proactive interaction has gradually shifted models from passive responders toward more proactive agents that can decide when to answer. However, existing proactive paradigms typically assume that the information required to answer a query is complete, leaving models to only wait or answer. In the real world, however, information is often incomplete due to uncertainty from either the video or the user. In proactive paradigms, such uncertainty can cause the model to wait forever or generate highly hallucinatory responses. We therefore argue that proactive interaction should be redefined with a third action: ask, allowing models to actively ask for missing information. To study when and what models should ask before answering, we introduce AskStream, the first multi-turn benchmark spanning video and user uncertainty for streaming video understanding, with four task categories and eight subtasks. It contains 438 videos with 2,714 manually annotated frame-level time windows across three types: wait, ask, and answer. Across both multiple-choice and open-ended settings, even the strongest model attains only 52.18% on our proposed Ask-F1 metric, highlighting substantial limitations in proactive information seeking under uncertainty.

0.38

Hidden Commitment: When Language Models Silently Pick a Side and How Steering Can Surface It

cs.CV ·2026-10-28 ·Samuel Dawit Assefa, Jae Won Cho
top venue
Abstract · Presented at AI4Good @ ICML 2026

Language-model answers are often treated as a visible record of the internal decision that produced them. We test this assumption in conflicting-context question answering, where two supplied documents support incompatible answers. A model can process the disagreement and still return a one-sided answer, even under prompts that ask it to disclose conflict. In Qwen 0.8B, hidden source commitment is captured by a residual source-choice direction whose sign tracks which document the answer uses. Sparse middle-layer MLP writers feed the same coordinate, and steering the coordinate changes the chosen source. A transfer run on Gemma 3 1B preserves the source-choice readout and sparse-write bridge under LLM-audited labels. Conflict disclosure follows a different pattern, appearing weaker, later-layer, and only partially steerable. These results support a practical monitoring strategy for retrieval systems that detects silent source commitment with residual readouts and pairs conflict-surfacing prompts with disclosure steering when the model has selected a side under disagreement.

0.38

HyperIntent: Tracking Intent Shifts for Efficient Reasoning on Dynamic Textual Graphs

cs.GR ·2026-10-28 ·Yilin Yao, Bin Lu, Xinhui Li, Meng Jin, Xinbing Wang
code/data
Abstract

Dynamic web interactions continuously evolve as new edges arrive, offering large language models (LLMs) great potential for reasoning over edge-level temporal semantics. However, continuously arriving edges are often repetitive, obscuring the key structures that truly reflect temporal changes and limiting the efficiency of LLMs. A key insight is that beneath the dense stream of edge arrivals lies a sparse evolution of temporal intent, enabling LLMs to be responsively triggered only when such intent shifts, while otherwise inheriting the previous state. To this end, we propose a hypergraph-based framework named HyperIntent, which enhances dynamic graph reasoning by parsing differentiated intents, quantifying their temporal discrepancies, and performing shift-triggered semantic updating. Experimental results on destination node retrieval and future link prediction demonstrate that HyperIntent consistently achieves the best overall performance, with average relative Hits@1 gains of 34.0% and 25.4% over vanilla TGAT and DyGFormer, respectively. Meanwhile, HyperIntent reduces LLM token consumption by 54.4%-95.9% across six datasets while achieving leading predictive performance. In-depth studies further confirm the necessity of shift triggering for identifying structural and semantic temporal changes. The code will be made publicly available upon completion of the review process.

0.38

SpectralTF-MAE: A Pre-trained Spectral Foundation Model for Tokamak Fusion Plasma Diagnostics

cs.CV ·2026-10-28 ·Xiao Wang, Zhijie Lu, Duoqing Yang, Jiangtao Yu, Jin Tang +5 more
code/data
Abstract

Spectroscopic diagnostics provide rich information for characterizing plasma states in tokamak fusion experiments, while existing learning-based approaches are often designed for individual tasks and rely heavily on labeled data. In this work, we propose SpectralTF-MAE, a self-supervised spectral foundation model for tokamak plasma diagnostics. The model is pretrained on paired charge-exchange recombination spectroscopy (CXRS) and X-ray crystal spectroscopy (XCS) measurements through masked spectral reconstruction. To better capture diagnostically important spectral structures, we introduce a peak-aware masking strategy together with a peak-enhanced reconstruction objective, encouraging the model to focus on characteristic spectral regions. A shared Transformer encoder is employed to learn transferable representations across different diagnostic modalities. The pretrained model is further adapted to several downstream tasks, including plasma-parameter prediction, cross-diagnostic spectral generation, and spectral peak analysis. Experiments on EAST diagnostic data demonstrate that the proposed framework achieves competitive performance across different tasks and consistently benefits from self-supervised spectral pretraining. These results indicate the potential of SpectralTF-MAE as a general and transferable representation learning framework for plasma spectroscopic diagnostics. The source code and pre-trained models will be released on https://github.com/Event-AHU/OpenFusion/tree/main/SpectralTF-MAE.

0.37

PointNDF: Cascaded Neural Displacement Fields for Point Cloud Denoising

cs.CV ·2026-10-28 ·Michael Y. Wang, Pingping Cai
publishedcode/data
Abstract · 14 pages, 3 figures, 7 tables. Accepted to MMM 2027 (33rd International Conference on Multimedia Modeling)

Point cloud denoising aims to recover clean geometry from noisy 3D observations while preserving both fine surface details and the underlying spatial distribution of the points. Many learning-based methods adopt a two-stage pipeline that first extracts point-wise features from the noisy input and subsequently predicts point-wise displacements via simple regressions. However, these approaches often lack robust feature guidance and distribution-aware geometric refinement, causing them to fail at maintaining optimal point uniformity and geometric quality. To address these limitations, we propose PointNDF, a denoising framework that combines patch-level geometric context with local displacement prediction. Specifically, a Point Transformer V3 (PTv3) encoder aggregates multiscale geometric information into a global patch descriptor. Rather than directly regressing an unconstrained displacement vector, a Neural Displacement Field (NDF) fuses this global patch descriptor with per-point neighborhood features. It then predicts each displacement through bounded-vector and scale factor branches, thereby achieving a superior, more uniform point distribution. Three cascaded NDF stages propagate local features and progressively refine the point positions. Experiments on PU-Net and ModelNet10 demonstrate competitive denoising accuracy and favorable inference efficiency. The source code is publicly available at https://github.com/peridous/PointNDF.

0.37

Nemotron 3.5 Super and Lightning: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Models for Agentic AI

cs.CL cs.AIcs.LG ·2026-10-28 ·NVIDIA, :, Aastha Jhunjhunwala, Aayushi Ashokkumar Srivastava, Abhibha Gupta +554 more
highly cited
Abstract

We introduce Nemotron 3.5 Super and Nemotron 3.5 Lightning, open and efficient Mixture-of-Experts Hybrid Mamba-Transformer models designed for reasoning, long-context understanding, and agentic workloads. Super is a 120B-parameter, 12B-active multimodal model, while Lightning is a smaller 30B-parameter, 3B-active text model; both designed for efficient deployment. Both models combine large-scale distillation, long-context training, and native speculative decoding, while Super further incorporates Indic continued pretraining and agentic software-engineering mid-training. We develop a comprehensive post-training pipeline spanning supervised fine-tuning and reinforcement learning (RLVR and RLHF), complemented in Super by specialized teacher training and multi- teacher on-policy distillation (MOPD). This transforms Super from a text-only base model into a multimodal agent capable of reasoning over text, images, video, and interactive tool environments. We further explore test-time scaling through parallel agent execution and self-verification. Across training and deployment, we jointly optimize capability and efficiency. Indic continued pretraining improves Indian-language benchmarks by up to 9.82 percentage points, while agentic mid-training strengthens software-engineering capabilities and robustness across agent harnesses. The post-trained Super model achieves 51.78% on SWE-bench Pro, surpassing Nemotron 3 Ultra despite using approximately 4.6x fewer active parameters. Post-training MTP boosting improves speculative- decoding acceptance length by up to 21.8% under probabilistic draft sampling, while NVFP4 quantization achieves 99.62% median accuracy recovery relative to BF16. Inference optimizations enable up to 4,856 output tokens/s/GPU on GB200 while sustaining at least 50 tokens/s/user.

0.36

Streaming Speech Recognition on a Microcontroller: a 65.7M-Parameter Transducer in 2 and 4 Bits on the ESP32-P4

eess.AS cs.ARcs.LG ·2026-10-28 ·Billal Boumedine
code/data
Abstract · 18 pages, 2 figures, 11 tables, 1 algorithm. Code: https://github.com/BoumedineBillal/esp32-pie-asr ; weights: https://huggingface.co/Boumed…

We run Kroko, a 65.7M-parameter streaming Zipformer2 transducer for English, on the ESP32-P4, a microcontroller with two 400 MHz RISC-V cores, a 128-bit SIMD extension (PIE) and 32 MB of PSRAM, but no neural accelerator. Without any training, the encoder's 290 matrices are stored with 2 or 4 bits per weight in Hadamard-rotated groups: the width of each matrix comes from its measured sensitivity, and each matrix is first corrected toward the original model's output, then rounded with GPTQ and a search over the group norms. The released 71 MB of int8 weights become a 26.65 MB file. Activations are int16 between the matrix products and int8 into them. A C engine with hand-written PIE kernels processes each 1.28 s chunk of audio in 0.90 s on both cores, beam search included, and its PC build computes the same bits as the chip, so accuracy is measured on whole test sets. LibriSpeech WER is 4.34% on test-clean and 10.26% on test-other, against 3.94% and 9.17% for the released model; on FLEURS and Common Voice the gap grows to 2.3 and 1.1 points. A GPU simulator shows that the integer arithmetic costs nothing and that the whole gap comes from the weights, and an ablation on all of test-clean shows that GPTQ does most of the work. We also report the port's speed history, from 51.3 s to 0.62 s of encoder time per chunk, each design step measured again in one build, an attempt to reach 16 MB that stopped at 19.63 MB, and what did not work. We did not measure energy. The inference engine, the firmware and the 2/4-bit weights are available at https://github.com/BoumedineBillal/esp32-pie-asr.

0.36

The Power Watershed as a Differentiable Segmentation Layer

cs.CV cs.NEeess.IVmath.OC ·2026-10-28 ·Galia Akl, Laurent Najman, Hugues Talbot
code/data
Abstract

Can a Watershed transform, the classical tool of morphological segmentation, be placed at the end of a deep network and trained through? The Power Watershed (PW) is the limit of the seeded Random Walker (RW) as p grows to infinity with edge weights raised to the power p. At every finite p the RW solves a sparse positive-definite linear system and is exactly differentiable with respect to the edge weights and the unaries through one adjoint solve; its multi-label form costs one right-hand side per label, and its limit is the PW. We study this family as the last layer of a segmentation network, the watershed layer (WS layer). A conditioning study measures how the solve, the gradient and the distance to the limit evolve with p, and the implicit gradient is checked against unrolling. On synthetic data with correct edge costs the WS layer transforms a soft prediction into an exact boundary. On nuclei segmentation (PanNuke), a binary WS layer is level with a thresholded U-Net. With the instances as labels, the WS layer turns the U-Net into an instance segmenter whose number of labels is set per image by the markers rather than by the output layout, and the network is trained through that partition: instance F1 score rises from 0.656 to 0.795 over a U-Net with connected components, level with the same network followed by a tuned marker watershed. At the training temperature the partition is a watershed of the learned costs, and the finite temperature lets the solve hedge, discounting weak markers and sharing contested pixels. The hedging puts the WS layer 0.015--0.020 F1 ahead of the baseline when trained on a tenth of the data, at the price of having to deploy the finite-p solve rather than its exact limit at inference. Code and the full experimental record are released.

0.36

SparseTokenGS: Gaussian-based Sparse-View CBCT Reconstruction via Reusable Token Pool

cs.CV cs.AI ·2026-10-28 ·Zihan Deng, Zhenzhen Gong, Yujie Liu, Jiaxin Gong, Dingkun Liu +8 more
code/data
Abstract · 10 pages, 12 figures. Submitted to IEEE Transactions on Medical Imaging

Sparse-view cone-beam computed tomography is widely used for fast, low-dose volumetric imaging and lesion localization in dentistry and image-guided interventions, but it introduces severe artifacts. Feed-forward Gaussian methods transfer the reconstruction cost to training and decode each coordinate from projection-aligned features and neighboring anchors. However, information from the whole scan is written only when the field is constructed and cannot be retrieved during coordinate decoding, leaving the volume inconsistent within the scan. This is a common limitation of such methods: once the field is fixed, decoding uses only local features and reconstruction quality drops substantially. We propose SparseTokenGS, a Gaussian-based method for sparse-view cone-beam computed tomography reconstruction via a reusable token pool. Anchor Token Fusion and Point-Token Attention query this shared pool at the anchors and at the query coordinates, respectively. The local Gaussian prior is retained, and the volume is reconstructed in one forward pass, without per-scan optimization or a projection-domain loss. On three public chest and dental datasets, SparseTokenGS achieves the highest reconstruction performance among the compared methods. On clinical dental scans, reconstruction from only six projections reaches 29.51 decibels in peak signal-to-noise ratio and a structural similarity of 0.842 against the full-view reference, and off-the-shelf tooth-and-pulp segmentation keeps these structures separable for root-canal localization. Experiments on volcanic rock and chicken show performance after same-object adaptation. A reusable token pool gives feed-forward reconstruction a mechanism that keeps scan-wide context available through decoding and can benefit future rapid clinical lesion localization and analysis. The code is available at: https://github.com/FrankDengAI/SparseTokenGS.

0.36

From the Superderivative to a Relative-Gradient Edge Operator

eess.IV cs.CV ·2026-10-28 ·Alejandro Garnung Menéndez
benchmarkunusual
Abstract · 15 pages, 11 figures, 6 tables

First-derivative edge detectors such as Sobel, Prewitt, Roberts and Scharr measure absolute grey-level differences. We revisit the log-gradient magnitude, a relative-gradient edge operator, and motivate it by analogy with the superderivative, a multiplicative derivative in which the independent variable is perturbed multiplicatively rather than additively. In one and two dimensions the logarithm of the superderivative is a radial logarithmic derivative; its local, origin-free, isotropic analogue is the magnitude of the logarithmic gradient. The resulting operator measures relative rather than absolute change, giving an edge-strength map that is invariant under global intensity rescaling. We present the theory, show how the operator can be implemented in two steps---a classical gradient kernel plus a pointwise division---and compare it with standard first-derivative filters on synthetic images with ground-truth edges and on the official BSDS500 benchmark. We also study a parametric family that interpolates between the absolute gradient and the log-gradient; empirically, the best members are small intermediate exponents, which edge the classical kernels by a small but statistically detectable margin in the per-image metric and, with the same Gaussian smoothing as the single-$σ$ Canny, match it in ODS and lead it in OIS and AP, while the tuned Canny remains ahead in ODS. The single-threshold margin over the kernels does not replicate on the held-out test split, and the fully relative endpoint is never the best member, its ranking depending on the edge-map normalisation. The contribution is the construction and the invariance it buys, not a new state of the art.

0.35

Automated Parameter Sensitivity Ranking in Audio Synthesis

cs.SD ·2026-10-28 ·Ogulcan Kibar, Baris Ozcan
Abstract · Submitted to the Journal of the Audio Engineering Society (JAES). 8 pages, 4 figures, 6 tables

Software synthesizers expose dozens of coupled controls, many of them redundant or perceptually inert at typical operating points. This paper presents a two-stage, measure-and-validate framework that ranks parameters by their influence on the audio output and then checks that ranking against an independent, variance-based experiment. Every rendered clip is normalized to -23 LUFS and described by five spectral fingerprints, so that gain shifts cannot masquerade as timbral change. The Sweep Test scores each parameter by spectral deviation over a 101-step sweep; the Sobol Test scores it by summed per-cell variance over a 128-point Sobol excitation. Across five architecturally distinct synthesizers (TorchSynth, Pyo, Pedalboard, Csound, and FluidSynth) spanning 48 to 76 parameters, the two tests agree at Spearman ρ= 0.82-0.96 (p-values \le 7.2 \times 10^{-15}) on the four engines that render deterministically, and at ρ= 0.32 on the one with process-level render noise--a diagnostic of that engine rather than of the method. The strong correlations survive reducing the Sobol budget from 128 to 32 points, and the complete analysis costs a one-time offline budget of about 66,000 renders.

0.35

Gold vs.~Translation: Are We Measuring Language, Knowledge, or Locality in Translated Benchmarks?

cs.CL ·2026-10-28 ·Klaudia-Doris Thellmann, Bernhard Stadler, Dan Saattrup Nielsen, Michael Färber, Jens Lehmann
top venuebenchmark
Abstract · Published in EMNLP 2026

Translated versions of English-origin benchmarks such as MMLU are widely used to evaluate large language models in non-English languages, yet it remains unclear whether they measure the same constructs as natively created benchmarks. We frame this as a proxy validity question and compare DeepL-translated MMLU proxies against six native benchmarks in seven pairings across five Nordic and Baltic languages, evaluating 29 models on each pairing. Beyond ranking agreement, we test incremental proxy validity: does the translated score add predictive value for native performance after controlling for model size and language proficiency? Proxy validity varies substantially across benchmark pairings. Applying the same robustness criteria to all seven pairings, we find four robust effects, two marginal effects, and one null result. The strongest evidence occurs for curriculum-overlap benchmarks such as Skolprov ($ΔR^2$=.128, 95% CI [.038,.253]), whereas translated MMLU provides no significant incremental value for Sweden-specific factual knowledge in SwedishFacts (p=.242). Because SwedishFacts is also the only open-ended benchmark, we cannot fully separate the effects of local content and response format. Overall, the results indicate a clear pattern: translated benchmarks provide a better proxy when their source content overlaps with the native evaluation target, while locally anchored knowledge is less reliably captured.

0.34

TRIM: Temporal Run-Time Monitoring of 3D Object Detection via Activation Statistics

cs.CV cs.RO ·2026-10-28 ·Hakan Yekta Yatbaz, Konstantinos Koufos, Mehrdad Dianati
published
Abstract · 6 pages. Accepted for publication at the 2026 IEEE 23rd International Conference on Smart Communities: Improving Quality of Life using AI, R…

Reliable perception is essential to intelligent transportation systems, yet deep neural networks can silently miss safety-critical objects while continuing to produce apparently plausible outputs. Existing run-time monitors for 3D object detection generally assess each frame independently and therefore cannot exploit the temporal evolution of occlusion, point sparsity, and scene complexity. We present TRIM, a lightweight temporal monitor that detects missed-object failures without modifying or retraining the underlying detector. TRIM summarises intermediate activations using compact channel-wise statistics and processes short sequences of these summaries with a temporal convolutional network. Experiments on NuScenes show that temporal monitoring improves AUROC from 0.686 to 0.704 and reduces the false alarm rate at 95% recall from 0.818 to 0.759 compared with single-frame monitoring. Early and intermediate backbone features provide the strongest monitoring signal, while the temporal inference head requires only 0.159 ms on a CPU.

0.34

MEV: A Multi-Event Video Dataset for Long-Take Generation

cs.CV cs.AI ·2026-10-28 ·Peiyuan Zhu, Shaoan Xie, Yifan Shen, Wen Tian, Zijian Li +2 more
top venuedatasetcode/data
Abstract · Accepted to NeurIPS 2026. Project page: https://mev-dataset.github.io/

Current datasets for text-to-video generation are largely limited to short clips depicting a single event or coarse-grained captions. As a result, they provide limited supervision for modeling long-take temporal dependencies, often leading to temporally inconsistent generations when describing multiple events. In this paper, we introduce MEV, a multi-event video dataset designed for long-take video generation. MEV provides high-quality real-world videos where subjects can perform a sequence of distinct, non-overlapping events, featuring precise temporal boundaries and fine-grained procedural annotations. This event-centric discretization reduces confounding factors and enables models to support temporally localized conditioning and build long-take videos from atomic event units. Furthermore, we introduce a comprehensive annotation pipeline that segments videos into temporally grounded events, and generates structured captions at both the global level and the event level. In addition, under an event-centric evaluation protocol, we conduct multi-event generation experiments by fine-tuning on MEV with temporally grounded event conditioning, leading to improved long-take event transitions and more coherent content. We believe this work provides a publicly available resource for multi-event fine-tuning and event-centric evaluation, together with practical insights into learning from temporally structured video-text data.

0.34

REEL: Same Frames, Different Answers and Different Costs

cs.CV ·2026-10-28 ·Xin Liang, Jiangpeng He
published
Abstract · Accepted at the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2027, Datasets and Evaluation Track

Video understanding keeps moving toward longer recordings, and the longer a recording is, the more questions can be asked about it. But processing it again for every question does not scale, so many systems watch the video once, build a memory, and answer later questions from it. Yet what to remember is decided before anyone knows what will be asked, and whatever is left out cannot be recovered later. Moreover, whether the savings pay off is judged by token counts, which fall when the input gets shorter but not when the model reuses what it has already computed. We present REEL, an evaluation framework that compares three saved representations built from exactly the same observed frames, so that a change in answers can come only from what the system remembered. Text writes a short record of what happened, Visual keeps the encoder outputs, and Prefix State also keeps the language-model prefix. Each is compared with full reprocessing (Frames) by paired correctness changes, cumulative GPU time, and storage size. Across six vision-language models and two benchmarks, Text and Frames differ in correctness on 20\% to 29\% of questions, even when their mean accuracy is almost the same. By contrast, Visual and Prefix State stay much closer to Frames. With unlimited storage, Prefix State is the cheapest in total GPU time among memories that keep almost all of the answers Frames gets right, for every model on both benchmarks. The trade-off is storage: Prefix State keeps up to about 21 GiB per video, whereas Text needs only kilobytes. Reusing computation and discarding information both look like efficiency, but only the second removes evidence that later questions may need, and neither mean accuracy nor token count tells them apart. REEL makes that difference measurable for any system that answers many questions from one saved memory.

0.33

HEARSAYBENCH: Can AI Navigate from Abstract Human Rights to Lived Lives?

cs.CL ·2026-10-28 ·Ava Iranmanesh, Sobhan Lotfi, Ali Iranmanesh, Liwei Jiang
top venuedataset
Abstract · Accepted as an oral presentation at NeurIPS 2026 (Evaluations & Datasets Track)

As large language models (LLMs) reach users worldwide, personal counseling has emerged as one of their most consequential applications, sometimes in life-critical situations. Yet broader access does not mean equal support. The text these models learn from disproportionately reflects people with the freedom to act, embedding an implicit assumption that users can leave, report, seek help, or appeal. Existing benchmarks largely inherit this assumption, overlooking users whose agency is constrained by structural conditions such as war, statelessness, and displacement. Meanwhile, models are increasingly shaped by normative frameworks such as Constitutional AI, often grounded in ideals like human rights, even as many users live where those rights offer little protection or recourse. We introduce HEARSAYBENCH, a human-authored dataset of 400 test cases with in-region human verification grounded in scenarios where structural barriers stand between a person and the exercise of their rights. Built from United Nations and other human rights reports, the scenarios span over 80 regions and cover social, environmental, and personal constraints. Drawing on Nobel laureate Amartya Sen's Capabilities Approach, HEARSAYBENCH tests whether models can distinguish what a person is formally entitled to do from what they are actually able to do. Across 14 proprietary and open-weight models, every model infers a user's situation more accurately than it advises on it. Stronger models narrow this gap, but their recommendations remain unsafe to follow. Models may understand lives shaped by constraint, yet that knowledge remains hearsay if it cannot translate into advice people can act on. Those people are not alignment's long tail; they are among the users for whom good guidance matters most. HEARSAYBENCH makes this failure measurable, a step toward systems that serve the world as it is rather than as assumed.

0.33

Blind Spots: Where Small Language Models Fall Behind Frontier Models on Terminal Tasks

cs.CL cs.LG ·2026-10-28 ·Ashima Suvarna, Ruta Desai, Ahmed Elgohary
top venue
Abstract · Camera Ready, RAAAI NeurIPS 26

Small language models (SLMs) offer an attractive alternative to frontier models (FMs) because they can be deployed with lower cost and latency. This makes SLMs especially appealing for terminal tasks on edge devices, where responsiveness, local execution, and limited connectivity are important considerations. However, they continue to lag behind FMs on terminal tasks, as measured by standard task success rates. Yet these rates conflate several capabilities that may contribute to this gap. In this work, we characterize the SLM-FM gap along three axes: solvability, whether additional test-time compute allows an SLM to reach a correct solution; verification, whether it can identify a successful solution among candidate trajectories; and confidence, whether it can accurately anticipate its probability of success. Across two standard terminal benchmarks, we find that repeated sampling narrows, but does not eliminate, the gap in solvability. In contrast, a persistent gap remains in verification. Given the same candidate trajectory pool, FMs recover substantially more of the gap between random and oracle selection than SLMs. SLMs are also more overconfident in their prospective success estimates and exhibit poorer calibration. These results suggest that the SLM-FM gap is multidimensional, spanning not only the ability to generate successful solutions, but also the ability to recognize successful solutions and anticipate success. We hope this decomposition can guide future work on developing stronger and smaller terminal agents.

0.32

Silhouette-Based Gaitomics for Frailty Assessment in Older Adults

cs.CV ·2026-10-28 ·Laura McDaniel, Ryan Roemmich, Crystal Szczesny, Peter Abadir, Rama Chellappa +1 more
Abstract

Walking offers a window into physical function that may support interpretable video-based frailty assessment. We developed silhouette-based "gaitomics" to connect frailty predictions with measurable patterns of body configuration and movement. In 66 older adults assessed using the Fried Frailty Phenotype, standardized silhouettes and gait energy images yielded 90 descriptors of shape, temporal variation, motion, and texture. Minimum redundancy maximum relevance selection using the mutual information difference criterion prioritized informative features while limiting redundancy. Each classifier was evaluated using nested stratified cross-validation with five outer and three inner folds at the participant level, with feature ranking and feature-count selection confined to training partitions. Shapley additive explanations examined contributions to held-out predictions. Bilateral silhouette difference led the aggregate feature-selection rankings for three-class classification and both comparisons involving frail participants. Attribution profiles differed across tasks, with texture and consecutive-frame silhouette differences prominent in three-class classification. Among classifier-specific evaluations, CatBoost achieved the highest mean three-class micro-AUC of 0.647 \pm 0.085. Leading pairwise mean micro-AUCs were 0.833 \pm 0.249 for frail versus nonfrail, 0.765 \pm 0.168 for frail versus prefrail, and 0.554 \pm 0.129 for prefrail versus nonfrail classification. These findings connect model predictions to explicit gait characteristics while revealing task-dependent feature priorities and limited separation of prefrail and nonfrail participants. The framework provides an interpretable basis for further investigation, with larger independent studies needed to establish reliability and clinical relevance.

0.32

Optimal Sampling and Noise Mitigation for Vocal Fold Kinematics in High-Speed Videoendoscopy

eess.IV ·2026-10-28 ·Harikrishnan Unnikrishnan, Rita Patel, Kevin D. Donohue
Abstract · This is the preprint Auhors manuscript to be published in Journal of Voice

Objectives: High-Speed Videoendoscopy glottal area waveforms conflate lateral fold excursion with longitudinal opening, precluding direct estimation of velocity, acceleration, and left-right asymmetry. Displacement waveforms preserve these quantities but are susceptible to spatial quantization and temporal sampling errors. We characterize these errors, evaluate reduction via Kalman smoothing and temporal interpolation, and determine optimal frame rate bounds relative to pixel excursion, speed quotient (SQ), and open quotient (OQ). Methods: Sampling theory was used to derive analytical frame-rate bounds required for kinematic feature fidelity. Kalman smoothing was compared with FFT-based denoising, paired with sinc or spline interpolation. Interpolation was evaluated on 4000 fps recordings (N=55) downsampled and reconstructed. A 16000 fps dataset (N=12) served as reference for noise reduction across 2000-8000 fps downsampled waveforms evaluated on peak displacement, velocity, OQ, and SQ. Results: Sinc interpolation yielded lower median mean absolute error than spline (0.29 vs 0.52 pixels), with 85% of recordings (47/55) versus 45% (25/55) below 0.5 pixels. For kinematic feature estimation (N=12), Kalman-sinc outperformed FFT-sinc in 75% of feature-by-frame-rate pairings. Conclusion: Computing kinematic features without artifacts requires joint consideration of temporal and spatial resolution. Kalman-sinc preprocessing at 8000 fps is recommended for velocity and SQ measures. Peak lateral displacement can be measured at 2000 fps, while OQ requires 4000 fps for typical spatial resolutions (9-14 pixels maximum excursion). Pixel excursion should be reported alongside kinematic measures to contextualize data reliability.

0.32

The Prestige: Benchmarking Cognitive Visual Reasoning using Magic Tricks

cs.CV cs.AI ·2026-10-28 ·Shuo Wen, Beixi Du, Edwin Meriaux, Junming Shi, Chloe Si +3 more
publishedbenchmarkcode/data
Abstract · 40th Conference on Neural Information Processing Systems (NeurIPS 2026). Track on Evaluations and Datasets

We introduce The Prestige, a video understanding benchmark that probes cognitive visual reasoning through professional magic performances, where deliberate misdirection challenges models to track events and explain unexpected outcomes. The benchmark comprises open-ended questions ranging from temporal and causal understanding to robustness against adversarial questions. Evaluation of leading open-source and proprietary VLMs reveals a substantial gap between even the strongest model and human performance. We also find systematic discrepancies between human and automated grading and identify recurring model failure modes. Ablations further show that models can select correct multiple-choice answers even when they fail to generate correct explanations independently, motivating further investigations into question formats. Together, these findings highlight limitations in both VLM reasoning and its evaluation, underscoring the need for more reliable video understanding models and more rigorous benchmarks.

0.31

Vision-Based Density Estimation for Adaptive Traffic Control

cs.CV ·2026-10-28 ·Mohammad Usaid, Deepak Silaych, Hrithik Mhatre, Archak Mittal
Abstract · 14 pages, 5 figures, 2 tables

Traffic congestion is becoming a critical challenge in rapidly growing urban centers, particularly in developing nations where rapid urbanization and inadequate transport infrastructure place increasing strain on transport networks. In highly heterogeneous and chaotic traffic flows, camera-based traffic-state estimation is limited by severe vehicle occlusion and overlap. In this paper, we propose a framework to optimize traffic flow directly from stochastic visual observations. The architecture is structurally decoupled into three primary components: (a) vehicle density estimation, (b) predictive dynamics modeling, and (c) policy optimization. The proposed methodology introduces a convolutional neural network (CNN) that learns the mapping from camera feeds to spatial density maps, utilizing an adaptive k-nearest-neighbor (k-NN) Gaussian kernel to handle varying congestion levels. This feeds into an ST3DNet predictive module for spatiotemporal forecasting, followed by a hierarchical Graph-Transformer Multi-Agent Reinforcement Learning (MARL) algorithm to optimize signal phases. The density estimation stage was evaluated on 40 annotated intersection images from Rambagh 4, Jaipur, India, with a reported five-fold validation mean absolute error (MAE) of $3.32 \pm 0.95$ vehicles and a mean absolute percentage error (MAPE) of 14.5%. These results pertain to the perception component; empirical training and evaluation of the forecasting and control stages remain future work.

0.31

What Does LOL Measure? Hidden Label Shifts in Low-Light Enhancement Benchmarks

cs.CV cs.LGeess.IV ·2026-10-28 ·Adityan Thirumani Srinivasan
unusual
Abstract · 11 pages, 5 figures, 8 tables

LOL-v1 and LOLv2-Real are the standard benchmarks for supervised low-light image enhancement, and several leading methods ship hand-set test-time adjustments with their evaluation scripts. We audit both benchmarks with 18 public checkpoints of eight methods under one metric implementation that reproduces every published number to within 0.09 dB, and find that their test labels differ from their training labels. (i) The 15 LOL-v1 test labels are rendered differently: the same captures appear in LOLv2 with labels of identical luminance but 1.48x lower saturation, and LOL-v1's own training labels match the LOLv2 rendering; we confirm this on the official release. (ii) The LOLv2-Real test set is the contiguous block of images 690-789, whose labels are 14% darker than the training labels (Cliff's delta = -0.45, p < 1e-12) with no comparable shift in the input statistics we examined; methods trained without any labels score up to 5.5 dB higher on test than on train. (iii) All 15 LOL-v1 test inputs reappear in LOLv2-Real, 12 in its training set. The released test-time settings are consistent with compensating for these shifts. Separating what can be calibrated on training data from what only test-set tuning finds, the test-only residual has median 0.00 dB on LOLv2-Syn, which has no shift, but 0.37 dB on LOL-v1 and 0.92 dB on LOLv2-Real (up to 3.9 dB). Under the shift, test-set checkpoint selection inflates a from-scratch model by 0.9-2.0 dB, and removing the released settings changes which method ranks first. We give concrete reporting recommendations.

0.31

GradSplat: Gradients Inside the Forward Pass for Feed-Forward 3D Gaussian Splatting

cs.CV ·2026-10-28 ·Jingong Chen, Linlian Jiang, Guangyuan Li, Xinxin Zuo, Yang Wang +3 more
code/data
Abstract

Feed-forward 3D Gaussian Splatting predicts a scene from a few images, but our predictor is less accurate than its own Gaussians optimized per scene on those images. This gap lies mostly in the shape of the Gaussians, their scale and rotation, since optimizing the shape alone recovers almost all of it and opacity and color recover less. Gradient steps on the input views correct this shape, and the model holds these views at inference. Log-anisotropy correction changes the axis ratios of each Gaussian in log space at constant volume, and in-forward gradient refinement renders the Gaussians into the input views and applies the gradient of the photometric loss with a learned recurrent update, first to opacity and color and then to the shape. GradSplat is a feed-forward model that refines its own Gaussians by learned gradient steps inside its forward pass, with no target view. On both complete six-view RealEstate10K protocols, it has the best PSNR and SSIM among nine methods and keeps its accuracy from $6$ to $42$ input views. Without fine-tuning, it is the most accurate on DL3DV and has the best PSNR and SSIM on ScanNet++. The code will be released at https://github.com/Jingong-Chen/GradSplat/.

0.30

Geometry-Aware Trajectory State Modeling for Few-Shot Action Recognition

cs.CV ·2026-10-28 ·Zhiqi Miao, Marko Haralović, Stanislav Levendeev, Sarvesh Angadi, Jannatul Ferdous +3 more
published
Abstract · This work is accepted by The IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2027

Few-shot action recognition (FSAR) requires representations that expose discriminative motion dynamics from only a few labeled videos. Recent trajectory-based methods leverage long-term point tracks for explicit temporal correspondence, but often treat trajectories as carriers of sampled visual features or summarize their motion with compact 2D displacement-based descriptors. We instead view an action as the ordered evolution of 2.5D trajectory states and their local geometric interactions. Based on this view, we introduce GeoTRAM, a geometry-aware trajectory state modeling framework for FSAR. GeoTRAM lifts sparse tracks into 2.5D trajectories, where StateMotion models ordered intra-trajectory state evolution and RelMotion captures state-conditioned local interactions through time-varying relative geometry and motion. Quality-filtered spectral sampling provides more reliable trajectory initialization; complementary semantic, motion, and depth cues are separately refined and preserved for joint spatio-temporal modeling; action-guided reweighting and top-$ρ$ trajectory selection support complementary frame- and trajectory-level matching. Experiments on five FSAR benchmarks demonstrate strong gains on the motion-centric Something-Something benchmarks while maintaining competitive performance on more appearance-driven benchmarks and a favorable accuracy--efficiency trade-off.

0.30

Towards Reliable, Scalable, and Value-Aligned Generative Recommendation for Large-Scale Advertising

cs.IR cs.LG ·2026-10-28 ·Yatong Xiao, Yibo Sheng, Lixiang Wang, Mingjie Sun, Yunhao Sha +5 more
hot direction
Abstract · 9 pages, 3 figures, under review

Generative recommendation (GR) reformulates item retrieval as autoregressive generation of discrete semantic identifiers (SIDs), offering a unified paradigm for personalized recommendation. GR4AD, as an industrial generative recommendation framework, has greatly improved advertising revenue while enabling efficient real-time serving in Kuaishou. However, continued deployment has exposed three limitations: deterministic SID assignment remains sensitive to semantic boundaries; conventional backbone scaling provides diminishing revenue gains under strict serving constraints; and sequence-level value optimization provides limited credit assignment to intermediate SID decisions, causing likelihood-based beam search to prematurely prune potentially high-value candidates. We present GR4AD-X, which addresses these limitations through reliable SID construction, serving-efficient capacity scaling, and unified value alignment. Specifically, GR4AD-X improves SID reliability through Multi-Path assignment. It expands model capacity along multiple complementary dimensions rather than relying solely on wider or deeper backbones, achieving a better effectiveness-efficiency trade-off under strict serving constraints. Finally, Unified Value-Guided Generation and Discrimination (UniVGD) aligns generation with commercial value through prefix-level reward optimization, value-guided beam pruning, and discriminative reranking within a unified model. Large-scale online A/B tests on the Kuaishou advertising platform demonstrate that GR4AD-X achieves a cumulative revenue improvement of 3.07% over the deployed GR4AD baseline while satisfying strict latency and throughput constraints.

0.30

OneScene-Bench: Do Vision-Language Models Integrate Evidence Across Views?

cs.CV ·2026-10-28 ·Swathi Suhas, Anurag Das, Bernt Schiele, Jonas Fischer
publishedbenchmarkcode/data
Abstract · Accepted at LM4RWD Workshop, Conference on Neural Information Processing Systems (NeurIPS), 2026

Vision-language models (VLMs) increasingly process multiple views of the same scene, but it remains unclear whether they integrate these observations into a coherent scene-level representation. Existing multi-image evaluations do not isolate this capability, as tasks may admit correct solutions through independent per-view processing without requiring cross-view correspondence. We introduce OneScene-Bench, a controlled benchmark designed to explicitly distinguish when cross-view integration is necessary. Our benchmark evaluates three complementary capabilities on real-world multi-view scenes: unique object counting, cross-view consistency, and spatial reasoning. For counting, we control object overlap across views to separate cases solvable by simple per-view aggregation from those requiring instance-level correspondence. For consistency, we introduce localized edits in shared and view-specific regions, testing whether models distinguish genuine scene contradictions from differences caused by viewpoint and visibility. For spatial reasoning, we construct object relations that require combining spatial evidence across viewpoints to compare physical proximity. We evaluate twelve state-of-the-art open-weight and proprietary VLMs and find that counting accuracy drops substantially on complementary views, ranging from 4.5-50.0%, compared with 45.5-81.5% on complete-overlap examples. Moreover, cross-view consistency remains around 50% across models, including recent proprietary VLMs, despite substantially stronger performance on spatial reasoning. Our results reveal a substantial gap between processing multiple images and integrating evidence across them, highlighting the need for evaluations that explicitly test scene-level multi-view understanding. Code and data are available at https://github.com/swathisuhas/OneScene-Bench.

0.30

UnMixMamba: Unmixing Guided State Space Subpixel Mapping of Crops from MODIS Time Series

cs.CV ·2026-10-28 ·Motasem Alkayid, Yimin Zhu, Naser El-Sheimy, Lincoln Linlin Xu
Abstract

Subpixel mapping (SPM) of crop types from \SI{250}{m} Moderate Resolution Imaging Spectroradiometer (MODIS) time series to field-scale Agriculture and Agri-Food Canada (AAFC) labels supports acreage estimation and agricultural monitoring. Three difficulties remain: (i)~spectral-temporal coupling and mixed pixels that break direct coarse-to-fine classifiers; (ii)~intra-class endmember variability that a single spectrum per crop cannot capture; and (iii)~uniform refinement on the high-resolution (HR) grid, which treats every subpixel the same, so most of the effort goes to easy pixels while errors remain in the minority class. UnMixMamba addresses them as follows. For~(i), a temporal-spectral-spatial decoupled (TSSD) factorization estimates simplex fractions and drives AAFC classification from the same mixture field, instead of mapping low-resolution (LR) observations directly to HR logits. For~(ii), \(\times 8\) abundance super-resolution and a learnable per-crop variant endmember bank reconstruct the MODIS cube at LR and HR under a linear mixing model. For~(iii), the predicted crop fractions decide where a fixed Mamba token budget is spent, so HR refinement focuses on the minority class instead of the full grid. On Alberta 2015 under a fixed spatial-block protocol, UnMixMamba reaches OA 0.8397, mIoU 0.6640, AA 0.7837, and Kappa 0.7715, ahead of ST-DRes, ESPCN, FineCrop, AgriFM, and classical SPM baselines. The task is closed-set four-crop mapping inside a known agricultural mask.

0.30

Geometry Matters in Packed 3D Attention

cs.CV cs.AIcs.LG ·2026-10-28 ·Amil Khan, Anantajit Subrahmanya, B. S. Manjunath
published
Abstract · Accepted to NeurIPS 2026 (Poster)

Packing a 3D volume into a token sequence is not merely long-sequence modeling. Once the volume is packed, sequence order no longer encodes the geometry of the sample---so a positional mechanism indexed by sequence position can learn the memory layout rather than the object. This is a general failure mode for attention over packed geometric data. We introduce PACK3D, a local-global attention operator that separates memory layout from attention geometry. PACK3D keeps local neighborhood attention exact, compresses only distant context into contiguous 3D block summaries, and indexes every query-key interaction with axial RoPE on true token-grid coordinates rather than packed positions. Because the rotation is applied before a block is summarized, the query-to-summary logit remains a weighted sum of query-relative 3D displacements; compressed global context therefore stays spatially interpretable after packing, instead of collapsing to anonymous pooled content. We measure the failure mode directly. On a controlled synthetic task---train on one packing, change only the memory order of the tokens, hold the physical labeling rule fixed---learned positional embeddings and packed-sequence RoPE solve the training layout perfectly yet collapse under Morton, block-contiguous, and random repacking, while axial grid-coordinate RoPE stays exact throughout. The same coordinate system carries to real data: on Allen WTC-11 hiPSC microscopy, PACK3D improves over a parameter-matched 3D U-Net on the 15-structure joint task and over a matched learned-position local-global baseline, with controls isolating the coordinate system, not encoder capacity, as the source of the gain. A fused Triton implementation matches windowed-attention latency at 65k tokens while supplying the long-range context windowing cannot---so the operator is not only correct but trainable at scale.

0.29

ETA-Motion: Efficient Topology-Agnostic Motion Diffusion via Sparse Joint Learning

cs.CV ·2026-10-28 ·Seoyoung Hwang, Tae-Kyun Kim
publishedcode/data
Abstract · Accepted to WACV 2027; Project page : https://sy-hwang.github.io/eta-motion-page/

Topology-agnostic motion generation aims to synthesize motions for diverse characters with arbitrary skeletons. Unlike conventional motion generation, which assumes a fixed skeletal template, this setting must handle diverse joint configurations and inconsistent joint semantics across characters. Existing methods address this challenge through dense full-joint modeling, which incurs substantial computational cost and hinders learning transferable motion patterns across diverse skeletal topologies. We present ETA-Motion, a topology-agnostic motion generation framework based on sparse joint learning. Instead of learning motion from all joints, our method uses a compact subset of joints to reduce redundancy and trains the motion generator on the resulting sparse skeleton. To facilitate correspondence across diverse topologies, we introduce semantic joint embeddings that combine textual and visual cues, together with topology-aware training objectives for preserving skeletal structure. Although ETA-Motion is trained only on sparse skeletal observations, it supports full-skeleton motion generation during inference. Experimental results demonstrate that ETA-Motion outperforms existing topology-agnostic baselines while significantly improving efficiency and generalization to unseen motions and unseen topologies.

0.29

Structural Pattern Mining in Inka Khipus: A KFG Re-Analysis

cs.CL cs.LG ·2026-10-28 ·Maria Contreras
unusual
Abstract · 7 pages, 1 figure

This paper reports a methodological pivot from the Open Khipu Repository to the Khipu Field Guide (KFG) database, prompted by stalled progress on the OKR, and a full re-analysis of the resulting 650-khipu sample. Three findings recur across every analysis attempted. First, khipu size confounds nearly every derived structural count, explaining 42-70% of its variance; all subsequent models control for it. Second, the two-letter name prefix (AS, UR, HP, QU, JC, MM, KH) identifies the original cataloger, not a site or period, which turns documentation completeness into a second confound that must be modeled explicitly. Controlling for both, khipus cataloged by Hugo Pereyra's Proyecto Quipu (HP) show significantly fewer Ascher sum-relation types than Gary Urton's (UR) khipus (p = 0.044), while the Ascher-cataloged (AS) khipus are too collinear with their own near-total lack of side/direction documentation to assess (VIF = 5.0). Third, generic structural clustering (UMAP+HDBSCAN) does not separate the pooled 650-khipu sample from two independent null models (permutation and Gaussian), corroborating an independent conclusion already reached by the KFG itself; the same test restricted to a single cataloger (UR) proved not reproducible across runtime environments and is reported here as a methodological caveat rather than evidence of structure. A targeted replication of the KFG's own S/Z-knot spatial-block heuristic for calendar/astronomical khipus recovers 8 of 10 published candidates and proposes 34 additional ones under the same criterion.

0.29

Data Scaling Laws in Stereo Matching: From In-domain Fitting to Cross-domain Generalization

cs.CV ·2026-10-28 ·Tong Zhao, Di Liang, Yiming Zeng, Bo Xu, Yingzhi Tang +1 more
publishedbenchmarkcode/data
Abstract · WACV 2027

In the era of large models, stereo matching is developing towards generalizable foundation models trained on million-scale synthetic images. However, how model performance scales up with synthetic data is still underexplored. In this work, we preliminarily investigate the in-domain and cross-domain data scaling laws in stereo matching. A realistic synthetic dataset, named XStereo, containing 200 thousand samples is first built to serve as a high-quality benchmark for data scaling research. In-domain training on five large-scale datasets demonstrate a two-regime scaling behavior with data, transitioning from the standard power law to saturation, where the critical transition point is approximately invariant under a fixed model. Cross-domain performance exhibits a substantially weaker and less predictable dependence on the source-domain data scale. The distribution and coverage diversity are more effective for diminishing domain gaps. More instructive suggestions are provided for high-efficiency data construction and training. The insightful findings are significant verification and extension of the existing scaling laws in both language and vision, paving the way to efficient and stronger stereo matching. The XStereo dataset is public at https://huggingface.co/datasets/Tong-Zhao/XStereo.

0.29

Scaling Lean Theorem Proving for Real-World Code Verification

cs.CL cs.AIcs.LOcs.SE ·2026-10-28 ·Debargha Ganguly, Divya Raghunathan, Dejan Jovanović, Vikash Singh, Soonho Kong +4 more
benchmarktheory
Abstract

We present EasyProver, a Lean proving system built from recursive agents with isolated contexts, asynchronous search, parallel tactic probing, and an elastic elaboration service. Using an off-the-shelf frontier model with no additional training, it proves 669/672 PutnamBench problems and all six IMO 2026 formalizations and resolves 597/600 NTP4VC verification conditions and all 189 VERINA tasks by proof or refutation. Because these benchmarks are close to saturation, we also introduce Aegis, which converts competitive programming problems into much harder verification tasks over real implementations. The pipeline writes safe Rust solutions, translates them into Lean with Aeneas, generates specifications, and labels each frozen statement with a kernel-checked proof or refutation. It produces 2,045 differentially validated Rust-Lean program pairs and 11,533 candidate specifications. Unlike hand-written benchmarks, these tasks carry implementation-level obligations from fixed-width arithmetic, explicit failure, generated loops, casts, and functionalized mutation. On a stratified sample of 2,000 statements, EasyProver proves 1,054 (52.7%) within a 30-minute budget per statement. The same tasks can also be used to benchmark other provers and as rewards for reinforcement learning.

0.28

How Many Demonstrations Does a VLA Need? On the Sample Efficiency of VLA Finetuning

cs.CV cs.RO ·2026-10-28 ·Rahul Chand, Miguel Villa Floran
Abstract · 14 pages, 8 figures, CoRL Workshop

Pretrained vision language action (VLA) models show strong zero-shot performance but they are rarely deployed directly and are often first post-trained on demonstrations from the target setting. Yet VLAs are typically compared by zero-shot or full-data fine-tuned success which reveals little about the data and compute required for downstream adaptation. We study adaptation efficiency for six VLAs ($π_0$, $π_{0.5}$, and GR00T N1-N1.7) on four LIBERO suites, varying both demonstrations and training steps. This reveals differences hidden by full-data fine-tuning: models which are within two points on the full LIBERO-Spatial dataset differ by up to 3$\times$ in success with 50 demonstrations. $π_{0.5}$ needs 2-5$\times$ fewer demonstrations than $π_0$ and is the most sample-efficient model at every budget, with the largest gap on the long-horizon suite. Within GR00T, N1.5 adapts more efficiently than N1.6 and N1.7 on three of four suites, despite their stronger zero-shot performance. We also study one pretraining factor behind these differences by freezing the VLM during finetuning: this severely degrades the $π$ models whose VLM is jointly trained with the action expert during pretraining, but barely affects GR00T models.

0.28

AlbumVSR: Identity-Preserving Face Video Super-Resolution from a Personal Photo Album

cs.CV eess.IV ·2026-10-28 ·Nianyao Du, Xijun Wang, Stanley H. Chan
benchmark
Abstract · 5 pages, 2 figures, 2 tables. Nianyao Du and Xijun Wang contributed equally

Severe blur, downsampling, noise, and compression can erase the identity cues needed to faithfully restore a person's face. We therefore use photographs of the same person as an external source of identity. AlbumVSR conditions a frozen, one-step streaming diffusion transformer on a personal album. We use four MagFace/CLIP set tokens to encode the album, and 64 texture tokens which come from a fused UV atlas rendered at the current window's estimated face pose. These tokens are appended to the transformer's cached cross-attention context. On an identity-disjoint VoxCeleb2 benchmark with heavy synthetic degradation, AlbumVSR raises ArcFace similarity to the ground truth from 0.586 to 0.670 and face PSNR from 24.98 to 26.71 dB. On the 91 test identities, its mean identity similarity is higher than that of all 14 baselines.

0.28

From Next Day Wildfire to Next Days Wildfire: A New Multi-Continent Dataset of Active-Fire Spread for Sequence-to-Sequence Forecasting

cs.CV ·2026-10-28 ·Jose Guillen, Renzo Zanca, Valentin Barriere
publisheddataset
Abstract · Accepted at WACV 2027. 19 pages including supplementary material, 12 figures, 13 tables

Benchmarks such as WildfireSpreadTS (WSTS) have enabled data-driven forecasting of wildfire spread from multimodal satellite time series, but they suffer from limitations: size restricts them to large fires that burn for weeks, and they are from similar environments. We introduce AFSeq, a multimodal dataset of 1,194 recent fire events (2023 to 2026) from 8 different countries across three regions, dominated by the short, small fires that existing benchmarks underrepresent. Our dataset is the largest and most diverse to date for time-series-based wildfire spread prediction. We also introduce UTAE-AR, an extension of a classical U-Net with Temporal Auto Encoder, enhanced with recurrent modules before the decoder. This allows the model to not only predict the next day but the whole sequence in an autoregressive way. Moreover, our model can feed its own predicted fire mask back one day at a time, producing the day-by-day forecasts that single-step models cannot. Our results show that UTAE-AR surpasses state-of-the-art on both WSTS, WSTS+ and AFSeq for next-day accuracy while having 21.5% of their parameters, and being able to predict confidently at a longer horizon.

0.28

Predicting Cognitive Profiles from Structural MRI using Spatial Graph Representation Learning

cs.CV ·2026-10-28 ·Taslima Ferdaus Shuva, Md. Tanvir Rahman, Risala Tasin Khan, Mohammad Ali Moni, ASM Ashraf Mahmud
Abstract · 15 pages, 4 figures. This work has been submitted to the Elsevier Array for possible publication

Predicting cognitive abilities from structural neuroimaging remains challenging due to the high dimensionality and geometric complexity of brain morphology. Conventional models typically flatten regional morphometry into tabular features, failing to explicitly represent spatial relationships. We propose a Spatial-Topological Graph Neural Network (ST-GNN) to predict four distinct cognitive phenotypes across large pediatric cohorts (N = 2,378 to 2,424). By projecting regional morphometry onto a population-level standardized spatial graph, ST-GNN restricts message passing to defined neighboring regions. Evaluations demonstrate that ST-GNN effectively fuses spatial morphology with systemic covariates, outperforming linear baselines and topology-free deep models. Ablations reveal that imposing a sparse geometric topology (K-nearest neighbors) and utilizing global feature scaling - which preserves absolute morphometric scale - outperform fully connected networks and region-normalized approaches. These choices provide transferable architectural inductive biases, revealing differences between computational routing and morphometric feature attribution. Integrated Gradients analysis identified distinct morphological features consistent with neuroanatomical evidence, such as cerebellar volume for working memory and superior frontal surface area for executive function. These findings suggest standardized structural MRI provides useful information for task-specific cognitive prediction, establishing a foundation for MRI-based neurodevelopmental assessment.

0.28

IntReal: A Real-World Dataset and Benchmark for Multimodal Interleaved Generation

cs.CV ·2026-10-28 ·Shaohan Yu, Jie Qin, Wende Tan, Haiwen Diao, Limeng Qiao +2 more
benchmark
Abstract · https://yushaohan.github.io/IntReal/

Humans routinely consume multimodal content, where textual and visual information is naturally interleaved. Although Unified Multimodal Models (UMMs) can jointly generate text and images, multimodal interleaved generation remains largely underexplored. Progress in this area is hindered by two key limitations: (1) the lack of well-defined interleaved generation tasks and corresponding training data, and (2) the absence of fine-grained evaluation protocols that can reliably diagnose model performance across different capability dimensions. To this end, we present IntReal, comprising two complementary components: IntReal-100K, a real-world dataset of multimodal interleaved samples spanning four major categories, and the IntReal Evaluation Suite, which evaluates model performance across seven complementary dimensions using specialized expert models, enabling coarse-to-fine diagnosis. Extensive experiments validate our evaluation suite through low inter-metric redundancy (max pairwise absolute Spearman's $ρ\leq 0.42$ ), high agreement with human judgments (83.3% accuracy), and strong consistency across different expert evaluators (Spearman's $ρ> 0.9$ ), supporting the complementarity and reliability of our metrics. By benchmarking four families of UMMs, we show that IntReal Evaluation Suite reveals distinct strengths and weaknesses across models. Post-training experiments demonstrate the utility of IntReal-100K for improving interleaved generation and further show that training unimodal generation can also benefit interleaved generation. These findings suggest that interleaved generation can benefit from modality-specific optimization, rather than requiring joint text-image training at every stage.

0.28

TCC-RAG: Temporal Causal Chain-Guided RAG for Temporal Knowledge Graph Question Answering

cs.CL ·2026-10-28 ·Xinxun Zhang, Yuxiang Wang, Shanshan Feng, Jun Zhang, Tieyun Qian +4 more
published
Abstract · 14 pages, 6 figures, ICDE2027 Research Track

Recent advances in large language models (LLMs) have spurred growing interest in temporal knowledge graph question answering (TKGQA). Most existing methods rely on semantic retrieval augmented generation (RAG), which suffers from fundamental limitations in complex temporal reasoning. On the one hand, semantic-based RAG methods struggle to capture the multi-hop causal knowledge required for complex temporal reasoning. On the other hand, these methods introduce excessive redundant knowledge, trapping LLMs in spurious reasoning chains that potentially result in erroneous answers. To address these problems, we propose a Temporal Causal Chain-guided RAG (TCC-RAG) to retrieve a concise yet sufficient set of causal knowledge to support complex temporal reasoning. Specifically, we reformulate the temporal knowledge graph into a multi-view causal cue graph, with causal cues inducing connections that explicitly encode temporal evolution and causal dynamics. Based on the causal cue graph, we construct a query-specific temporal causal chain to guide the retrieval process, enabling the model to focus on a concise set of causally relevant knowledge. Finally, we introduce a causal evidence filter to refine the retrieved knowledge, ensuring its causal consistency with the underlying temporal logic of the query. Experiments show that TCC-RAG achieves substantial performance gains of 10.72% and 8.40% on two TKGQA benchmarks, and delivers a remarkable 62.97% improvement for multi-step temporal reasoning.

0.28

LLMs as Judges and Examinees for Medical Item Difficulty Prediction

cs.CL cs.AI ·2026-10-28 ·Xiaomeng Xiong, Marcus Walker, Yanlin Jiang
Abstract

Predicting medical item difficulty from item text alone remains challenging. Prior work has largely relied on linguistic features or text representations derived from general-purpose and domain-specific language models, but predictive performance has remained modest. This study takes a different approach by using difficulty-related signals that emerge when open-weight large language models (LLMs) interact with test items. Using USMLE and PANCE medical licensure examination items, we examined LLMs in two roles. As judges, the models provided absolute difficulty ratings and pairwise difficulty comparisons. As simulated examinees, they generated response-based signals including correctness, keyed-option probability, response entropy, and the probability margin between the top two options. We then combined the strongest judgment and response signals with baseline item features in downstream prediction models. Pairwise difficulty judgment was the most consistent individual LLM signal across both datasets and outperformed absolute ratings, and adding simulated-examinee signals yielded only small further gains. The best configurations reached a Spearman correlation of $ρ$ = .447 on USMLE, exceeding the best system in the 2024 BEA shared task ($ρ$ = .25), and $ρ$ = .494 on PANCE. These results were comparable to direct absolute judgments from the proprietary frontier model Claude Opus 5 ($ρ$ = .443 and .488, respectively). Overall, judgment and response signals from open-weight LLMs captured difficulty-related information beyond text representations. Because these models can be deployed locally, the approach offers a practical option for predicting the difficulty of unreleased or high-security test items.

0.28

Volley3D: Unified 3D Full-Sequence Reconstruction from Monocular Broadcasts

cs.CV ·2026-10-28 ·Noâm Boussouf, Adrien Maglo, Mohamed Chaouch, Hideo Saito
Abstract · 8 pages + 10 pages of supplementary. Videos available on the project webpage: noambsf.github.io/Volley3D

3D reconstruction from single-view sports broadcast is a pivotal capability for automated analytics and immersive AR/VR replays, yet notoriously ill-posed due to difficult-to-calibrate moving cameras and depth ambiguity. Unlike widely studied racket sports, team sports such as volleyball feature unpredictable multi-contact sequences and high altitude jumps. These dynamics categorically invalidate the standard planar projection assumptions relied upon by existing methods. To resolve these limitations without 3D training supervision, we propose a complete, physics-grounded pipeline tailored to monocular broadcast volleyball. First, we include a correction module to the perspective drift induced by ground-plane projection by coupling temporal jump spotting with Newtonian interpolation for 3D player placement. Second, for 3D ball trajectory estimation, we design a dedicated sub-trajectory segmentation architecture alongside a collision-matching algorithm. By anchoring physical equations to reconstructed 3D hand vertices, we lift noisy ball 2D detections into a spatiotemporally coherent 3D scene. Evaluations on multi-camera benchmarks demonstrate our approach significantly reduces 3D spatial drift compared to 2D-reprojection baselines.

0.28

Implicit Minor Detection from Multi-Turn Dialogue in High-Risk AI Interactions: A Constrained Self-Evolving LLM Harness

cs.CL ·2026-10-28 ·Xiaohan Zhang, Ying Zhao, Ruilin Guo, Xuanyuan Yu, Kaibo Huang +3 more
benchmark
Abstract · 13 pages, 3 figures, including a technical supplement

In high-risk text interactions, a system may need to estimate whether the current speaker is likely to be a minor before applying age-conditioned safeguards. Many text-based dialogue systems lack strong identity signals such as account information, verified identity, facial data, or device data, and must rely on weak cues that emerge across multiple turns. We formulate implicit minor detection under weak identity information as a high-risk user-attribute inference task and propose a constrained framework comprising a large language model, a Harness, and a Maker-Checker protocol. The framework controls when inference is invoked, requires structured dialogue-linked evidence, and subjects validation-selected Harness revisions to version-level review. On a controlled Chinese benchmark comprising 8,328 multi-turn dialogues, our method improves average precision and minor-class F1 by 7.60 and 6.37 percentage points, respectively, over direct inference using the same LLM family.

0.28

Implicit Reasoning in Transformers: Bridging the Compositionality Gap in GPT-2-like Models

cs.CL cs.AI ·2026-10-28 ·Davide Venuto, Leonardo Venuto
code/data
Abstract · Code available at: https://github.com/Davide011/implicit-reasoning-bridging-compositionality-gap

Compositional generalization remains a persistent limitation of transformer language models: even when a model learns to combine known facts correctly in-distribution (ID), the same compositional mechanism may fail when applied to out-of-distribution (OOD) knowledge. We investigate this compositionality gap in a controlled two-hop reasoning setting using a GPT-2-like transformer model trained from scratch. Building on prior work identifying a circuit associated with compositional generalization, we test whether the failure of OOD transfer is related to how atomic knowledge is structurally exposed during training. We modify the training distribution by presenting both ID and OOD atomic facts in a format positionally aligned with the second-hop retrieval step of the learned compositional circuit, without modifying the model architecture or adding OOD compositional supervision. Under this intervention, the model successfully transfers the learned ID compositional behaviour to held-out OOD atomic facts, closing the observed compositionality gap in the controlled setting. Our results provide evidence that the observed compositionality gap is not a fundamental architecture limitation and that it can be overcome through mechanistically informed training distribution design.

0.27

RuleForge: Inferring and Instantiating Latent Visual Rules as Executable Programs

cs.CV ·2026-10-28 ·Weifeng Xu, Yaxiong Wang, Zhedong Zheng
benchmark
Abstract · 13 pages, 5 figures

Visual rule reasoning asks whether multimodal models can recover an abstract concept from a few examples. Existing studies either infer rules from visual examples but evaluate only classification or verbalization, or supervise executable generation with rules provided as input. Consequently, the relationship between visual rule understanding and reliable rule instantiation remains insufficiently explored. We present Bongard-Forge100k, a controlled benchmark, together with RuleForge, a backbone-agnostic framework that jointly learns visual rule understanding and rule-conditioned executable generation. The benchmark contains 108,797 problems: 86,000 for training, 19,297 for testing familiar rules, and 3,500 for testing held-out rule compositions. The framework progresses through rule-understanding training, joint understanding and scene-program supervision, and refinement via execution-grounded reinforcement learning with Understanding Replay. Executing generated programs under multiple random seeds provides verifiable feedback on executability, rule satisfaction, diversity, and robustness. With UniLIP as the primary backbone, RuleForge improves multiple-choice accuracy over the base model by 20.93 percentage points on familiar rules and 28.42 percentage points on held-out rule compositions, while increasing scene-level rule satisfaction on familiar rules from 2.91% to 70.99%. Additional experiments on InternVL3.5, Qwen3-VL, and Show-o2 demonstrate the effectiveness of the proposed training data and framework across diverse multimodal architectures.

0.27

A New Formulation of the Elastica Model with Application to Image Inpainting

cs.CV math.OC ·2026-10-28 ·Zuoqiang Shi, Xue-Cheng Tai
unusual
Abstract

Curves and interfaces play a central role in computer vision and image processing, and Euler's elastica energy is one of the most successful tools to model and manipulate them. The elastica energy is usually computed through a level-set or a phase field representation of the curve, in which the curvature term makes the energy nonlinear, non-convex and, in the level-set case, non-differentiable, so that its minimization is notoriously difficult. In this paper, we propose a new formulation of the elastica model based on a binary representation of curves. The key ingredient is a very simple formula for the curvature: if $φ$ is the $\pm1$ binary function associated with a curve $Γ$, then along $Γ$ the curvature is given by $-\frac{1}{C_0ε}\left(R_ε*φ\right)+O(ε)$, where $R_ε$ is a radial kernel supported in a ball of radius $ε$ and $C_0$ is an explicit constant. No derivative of $φ$ is involved and the convolution is evaluated by FFT. The resulting energy is a weighted total variation, so that formally the new formulation is a generalization of the well-known ROF model; a phase field type model is obtained from the same curvature formula. Combining FFT with the split Bregman iteration, which for the present problem coincides with an augmented Lagrangian method (ALM), we obtain efficient algorithms whose subproblems are either linear systems solved by conjugate gradient or explicit pointwise formulas. Numerical experiments on the inpainting of binary images are presented to demonstrate the effectiveness of the proposed algorithms.

0.27

Complementing Text Based Moderation with Multimodal Grounding: Detecting Fraudulent Inducement Behaviors on Short Video Platforms

cs.CV ·2026-10-28 ·Xiaoyong Yu
dataset
Abstract

Short video platforms have become a primary channel for black and gray industry operations in internet finance, where inducements such as prompts to like, follow, join private chat groups, or click external links frequently steer users toward fraud. Because such videos are short, dynamic, and often convey intent through fleeting visual cues rather than stable text, a moderation system must guarantee detection recall and localize the offending segment so reviewers inspect only the relevant clip. Existing OCR+LLM pipelines transcribe on-screen text frame by frame and classify the transcript, but exploit only textual information. Many inducements are conveyed through visual cues OCR cannot observe, and keyframe-based localization cannot recover precise temporal boundaries. We explore OCR+MLLM, which augments the text pipeline with a Multimodal Large Language Model, and compare three routes: OCR+LLM, OCR+MLLM, and a large zero-shot MLLM, along detection, latency, token consumption, and deployment complexity. OCR preserves the text signal that guarantees recall, while the MLLM jointly performs OCR-grounded discrimination and temporal grounding, localizing both textual and non-textual inducements. We construct a dataset of 2,419 internet finance short videos (66.49 hours) with 3,194 annotated events across four categories, and study reward design for RL-based grounding under class imbalance. We report two findings: auxiliary reward terms are exploited through reward hacking, and stronger localization rewards degrade class-balanced detection. We therefore design a recall-weighted F2 localization reward with class-balanced resampling. Experiments show OCR+MLLM localizes better than the keyframe OCR+LLM baseline (mIoU 26.52 vs. 14.49) and detects non-textual inducements that text-only pipelines miss, while maintaining 67.0 percent macro recall.

0.27

Neighbourhood shortcuts in building damage assessment from very-high-resolution satellite imagery: diagnosis and a building-evidence plug-in

cs.CV ·2026-10-28 ·JianChong Guo, YuTing Wan, Kefan Li, AiLong Ma, YanFei Zhong
code/data
Abstract · 22 pages, 9 figures, 11 tables. Code: https://github.com/SummerTide/neighbourhood-shortcut-bda

Building damage assessment (BDA) from very-high-resolution imagery supports disaster response, and deep models now achieve high overall accuracy on public benchmarks. Because damage is spatially clustered, most buildings share the damage level of their neighbours, so overall scores are dominated by such buildings, whereas damaged buildings among intact ones are rare and easily overlooked. A high overall score therefore cannot tell whether a model judges the building itself or relies on its neighbourhood as a shortcut, and this study tests which of the two current models do. We stratify buildings by their agreement with their neighbours, move buildings into new neighbourhoods with their pixels unchanged, and diagnose nine models covering the main paradigms of BDA. All models miss a third to half of the isolated damaged buildings, and in non-flood events they call 73--87% of intact buildings damaged once these are moved into a damaged neighbourhood. We propose a building-evidence plug-in that adds two experts to an existing model without retraining it. One expert sees only the building, and the other sees its surroundings but is trained with seamlessly swapped neighbourhoods so that it learns not to trust them. With each of the nine models, the plug-in maintains or improves overall accuracy, reduces errors on buildings that disagree with their neighbours by 8--18%, and cuts by 40--69% the intact buildings that a damaged neighbourhood turns into damaged ones. It also raises the detection of synthetic isolated damage from 1--6% to 13--39% and carries over to optical--SAR data and to an unseen earthquake. High overall scores thus conceal a systematic reliance on the neighbourhood, which building-level evidence reduces at low cost.

0.27

When Support Crosses the Line: Evaluating Ethical Boundaries in LLM-Based Mental Health Support in the Chinese Context

cs.CL ·2026-10-28 ·Bichen Wang, Yufeng Jin, Yixin Sun, Yanyan Zhao, Bing Qin
benchmark
Abstract

Large language models (LLMs) are increasingly explored for mental health support, yet existing benchmarks primarily focus on supportive capabilities rather than whether models consistently adhere to professional ethical boundaries. This issue is particularly important in multi-turn interactions, where users may introduce new information or exert emotional pressure over time. We introduce MindEth-Bench, a benchmark for evaluating ethical boundary adherence in multi-turn mental health interactions. Grounded in professional ethical standards for mental health services in China, MindEth-Bench covers multiple dimensions of ethical conduct and uses Monte Carlo Tree Search (MCTS) to systematically explore interaction trajectories that stress-test model boundaries. Experiments across nine LLMs show that models that initially respond appropriately can shift toward ethical violations over extended interactions, with substantial variation across models and ethical dimensions. These results show that single-turn evaluations may overestimate the ethical reliability of LLM-based mental health support systems and highlight the need to evaluate ethical behavior throughout multi-turn interaction.

0.27

Does 3D Pose Improve Action Valuation in Football? An Empirical Study of Pose Representations and Learning Objectives

cs.CV cs.LG ·2026-10-28 ·Rikuhei Umemoto, Kenjiro Ide, Soham Majumder, Ulf Brefeld, Keisuke Fujii
Abstract · Submitted to IEEE Access

Football action valuation estimates players' contributions to future scoring. Existing multi-agent deep reinforcement learning (MADRL) models use 2D positions and velocities, omitting posture information relevant to imminent actions. We investigate whether adding 3D pose to a SARSA-based model improves observed-action values relative to attack returns: goals, conceded goals or final expected possession value. Using UEFA Champions League 2022/23 event, tracking and 3D pose data, we compare six representations (no pose, raw joints, torso and head orientation, and spatial-temporal graph convolutional network (ST-GCN) embeddings) across five learning settings and three seeds. The settings vary temporal-difference targets and action-prediction loss, with Monte Carlo return regression as a reference. Across 50 on-ball and off-ball comparisons, pose robustly improved return prediction in only three, all using torso and head orientation, while worsening it in twelve, mostly with ST-GCN. Training choices had greater effects: removing the action-prediction loss raised on-ball skill of position-only values from 0.003 to 0.040 and their correlation with expected goals from 0.23 to 0.60; the improvement in return prediction persisted at matched training length. Separately, pose improved prediction of the on-ball action 0.2 s ahead (macro AUC 0.914 to 0.934 with raw joints), while the passer's pose during execution improved pass-success prediction (AUC 0.844 to 0.856). Within this training protocol, pose informed imminent actions but rarely improved observed-action values.

0.27

A 211K Parameter Foundation Model: Genetic Prompting in Neural Cellular Automata

cs.CV ·2026-10-28 ·John Kalkhof, Niklas Ihm, Daniel Kirste, Anirban Mukhopadhyay
Abstract

The scale of modern foundation models enables broad task generalization. In medical imaging, where segmentation demands vary widely across clinical contexts, such flexibility is invaluable. But current models, such as SAM trade adaptability for scale, locking their utility behind high-end infrastructure and leaving behind the low-resource environments most in need of diagnostic support. To bridge this gap, we introduce Foundational-NCA, a lightweight foundation model capable of adapting to a wide range of medical segmentation tasks without retraining. Built on Neural Cellular Automata (NCA), Foundational-NCA incorporates gradient-based learning with genetic optimization over a latent task space, using a compact task vector to guide predictions. This enables dynamic adaptation via a genetic prompt mechanism, allowing the model to generalize across tasks while keeping the parameter count magnitudes lower at just 211 thousand, compared to SAM-based models with 38.9 to 636 million parameters. We evaluate Foundational-NCA on 15 trained and 6 entirely unseen segmentation tasks from the TotalSegmentator dataset, achieving an average Dice score of 84.5% on known tasks and 60.4% on unseen ones. These results demonstrate that task generalization is possible in ultra-compact models, laying the groundwork for resource-efficient, deployable foundation models in clinical environments where infrastructure is limited. By reducing model size by over 180 times, Foundational-NCA takes a step toward a more equitable future for medical AI, one where powerful tools are within reach of every healthcare setting.

0.26

Neural Explicit Shape Representation

cs.GR ·2026-10-28 ·Jun-Hao Wang, Changan Lu, Baoquan Chen, Peng-Shuai Wang
code/data
Abstract

Existing neural 3D representations separately address continuous surface modeling, efficient image synthesis, or object-specific compression. They do not jointly provide the explicit surface queries required by ray tracing and a structured output space that can be predicted across objects. We introduce NeuOctree, a neural explicit representation that directly predicts whether and where a ray intersects a surface, together with its normal and albedo. NeuOctree stores shared features at the corners of sparse octree voxels and interpolates them into a continuous local field. A lightweight decoder with recurrent intersection-guided sampling refines each query near the predicted surface, enabling direct surface rendering without mesh extraction or texture baking and accommodating open, non-watertight geometry. The same octree structure supports both compact per-object optimization and a shared octree CNN that constructs renderable assets from colored point clouds without per-object fitting. Across seven meshes and four storage budgets, NeuOctree achieves the lowest positional error in 24 of 28 comparisons while retaining competitive normal accuracy, and substantially improves joint geometry--appearance reconstruction over N-BVH and VQAD. The feedforward model constructs a representation in 0.05 seconds and transfers from Objaverse to ShapeNet without retraining. We further demonstrate ray-traced rendering and image-conditioned generation, showing that NeuOctree provides a common output space for neural inference and direct surface rendering. Code is available at https://github.com/octree-nn/neural-octree.

0.26

inkGround: Benchmarking Fine-grained Visual Grounding Capabilities of Vision-Language Models for Cursive Handwriting in the Wild

cs.CV cs.AI ·2026-10-28 ·Sidra Hanif
benchmarkcode/data
Abstract

Human handwriting is produced as an ordered sequence of pen movements, but an offline image keeps only the final ink. Handwriting trajectory recovery aims to restore the ordered stroke points and their pen-up/pen-down transitions. Existing image-trajectory pair datasets are recorded with digitizing tablets under controlled conditions, which limits writer coverage and stylistic diversity. Moreover, there is no benchmark that measures the visual grounding capabilities of general-purpose Vision-Language Models (VLMs) for the stroke trajectory recovery task. In this work, we introduce inkGround, a benchmark for stroke trajectory recovery from full-page, camera-captured handwriting. inkGround pairs 537 page images (~18.7k words, ~163k strokes) with reference trajectories produced by a pretrained offline-to-online model, filtered by OCR agreement, and verified by human raters. Because these references are pseudo-labels that represent one plausible pen trajectory, we also evaluate VLMs on the IAM-OnDB dataset, which provides recorded pen trajectories. We cast the task as text-conditioned point grounding: given a page image and a target word, the model must locate the word and output its ordered stroke points. Outputs are scored by trajectory fidelity (DTW, Chamfer distance), readability of the rendered strokes (Accurate Rate, Correct Rate), and style similarity. We benchmark four open-source and three frontier closed-source VLMs with a shared in-context prompt. Closed-source models, notably Gemini-3.1-Pro, produce legible trajectories close to the reference, and this ranking also holds on IAM-OnDB.

0.26

Extremely Sparse-View Computed Tomography from 2D Projections via Pose-Aware Diffusion Priors

cs.CV cs.AI ·2026-10-28 ·Linrui Dai, Jinxiu Liang, Chu Zhou, Mae Yamaguchi, Ryuya Takahashi +1 more
published
Abstract · NeurIPS 2026 poster

Extremely sparse-view computed tomography (xSV-CT) is critical when scanning time, motion, or radiation tolerance limits each scan to only a few projections. In many such scenarios, the same acquisition limits also constrain dataset construction, making pre-reconstructed 3D volumes from dense scans unavailable. This exposes a gap in current sparse-view approaches: Optimization-based methods, which use the sparse measurements directly but lack data-driven priors to suppress artifacts, and learning-based methods, which use strong priors but usually require densely scanned training volumes acquired outside the xSV-CT regime. We study this train-sparse / infer-sparse setting by learning a pose-aware diffusion prior from one 2D projection per training sample, then guiding diffusion sampling with measured-view fidelity, shared-volume reprojection, and projection-physics consistency for tomographic reconstruction. On two public CT datasets across 1-, 4-, and 8-view settings, the proposed method achieves the best novel-view synthesis scores and best or competitive reconstruction quality compared with baselines that use dense-volume supervision.

0.26

Structured Pruning of Feedback Delay Networks

eess.AS eess.SPmath.NA ·2026-10-28 ·Facundo Franchino, Karolina Prawda
Abstract · 11 pages, 8 figures, 4 tables. Submitted to the IEEE Open Journal of Signal Processing

Feedback delay networks (FDNs) are recursive filter structures used to synthesise artificial reverberation. Their feedback matrix governs both energy exchange between delay lines and the computation required at every sample, making its design central to balancing an FDN's acoustic performance and implementation efficiency. In this work, we develop an FDN-specific structured pruning method that connects these acoustic and computational concerns, allowing high-order dense orthogonal designs to be fitted to efficient implementations. The feedback matrix is replaced by stages of fixed permutations and small orthogonal blocks. We characterise connectivity growth with the network depth and derive exact conditional block updates for matrix fitting, followed by refinement against the reference FDN magnitude response. The resulting factors admit direct vectorised evaluation, with arithmetic cost and coefficient storage scaling linearly with network size for fixed block size and depth. On an Apple M2, the selected staged kernels achieve feedback-multiplication speedups over dense BLAS ranging from approximately threefold to over twentyfold across the tested network sizes, while substantially reducing coefficient storage. Objective metrics and listening tests indicate that the proposed staged approximations closely preserve the spectral and temporal character of their dense references. An analysis of processor utilisation and register reuse explains why execution time depends on more than the operation count. Finally, block size and depth provide practical control over response fidelity, storage, and real-time cost, allowing acoustic design to precede the choice of an efficient implementation.

0.26

Human Evaluation with Automatic Scores: Separating Sampling and Estimation Gains

cs.CL ·2026-10-28 ·Yang Liu, Ran Zou, Zhengtao Gui, Chao Xue
Abstract

Automatic judges offer inexpensive information about model outputs, but an evaluation team must decide how to turn that information into fewer human annotations. Scores can guide sampling, correct an estimate, or support adaptive acquisition, and these uses may capture overlapping gains. We compare their incremental value on fixed evaluation pools from three preference datasets, and 19 attributed-question-answering systems. Matched comparisons use the same scores, account for pilot labels and budget rules, and separate acquisition from estimation. Score stratification is a strong reference: after ten-bin stratification, the best fixed local linear correction offers only 1.00--5.27% further MSE reduction in the primary configurations. Unequal allocation offers larger gains in some pools, but can increase error when confident scores conceal variable human outcomes. Comparison on AttributedQA, shrinking learned quotas toward score-only quotas lowers pooled MSE for 17 of 19 systems relative to the score-only procedure, but ordinary pooled 95% intervals cover only 88.09--93.54%. The same shrinkage improves on the score-only reference in only one of eight preference configurations. Ranking experiments further show that lower individual MSE need not improve system selection. These findings support a simple starting procedure, proportional sampling within score strata followed by a stratified human mean, and identify what additional evidence is needed to justify correction or adaptation for the intended evaluation output.

0.26

Structure-retained low-rank adapters for weather synthesis

cs.CV ·2026-10-28 ·Shunxin Wang, Alexandros Stergiou, Luuk Spreeuwers, Estefania Talavera, Nicola Strisciuglio
Abstract

Real-world deployment of computer vision models, particularly in outdoor scenarios, is often limited by environmental interferences, such as adverse weather. Improving model robustness across weather conditions is thus critical for reliability. However, sourcing high-quality data relevant to isolated weather effects is a significant challenge. Most datasets include changes in dynamic elements across observations of the same scene: subjects such as pedestrians and vehicles change between shots, while camera viewpoints vary. These aspects make it difficult to decouple solely weather-induced distribution shifts from the scene. In this work, we address this with a novel weather synthesis framework that utilizes structure-retaining Low-Rank Adapters (LoRA). We analyze the attention blocks in Stable Diffusion XL (SDXL) and identify where relevant structural information is mostly processed during image-to-image translation. Through an architectural analysis of SDXL, we learn structural LoRA matrices for the corresponding blocks and then leverage them for weather synthesis, eliminating the need for external conditioning such as ControlNet. Experimental results demonstrate that our method improves upon existing synthesis approaches, achieving greater realism while preserving structure. This provides a robust pipeline for generating high-fidelity training and evaluation data for outdoor computer vision applications.

0.25

JYVA at BeTraC: Audio-Based Medical Note Generation with SOAP Section-Specific Queries

cs.SD eess.AS ·2026-10-28 ·Youness Dkhissi, Valentin Vielzeuf, Antoine Caubrière, Jean-Emile Dartois
Abstract

We present our submission to the Beyond Transcription Challenge (BeTraC) on medical note generation from doctor-patient audio conversations. We propose a section-aware approach based on Qwen2.5-Omni-3B, in which a lightweight Q-Former learns SOAP section-specific queries to extract compact clinical representations from speech. In the lightweight track, our system ranks fourth and substantially outperforms the challenge baseline, achieving a Concept F1 of 0.5005 on the official test set. The organizers' analysis and our heuristic evaluation suggest improved robustness to hallucinations, and better stability under difficult audio conditions. These results show that explicitly modeling SOAP structure through section-specific queries is an effective way to improve the faithfulness and reliability of audio-based medical note generation.

0.25

Personalising Parametric Audio Effect Recommendation for Vocals from Parameter-only Style Priors

eess.AS ·2026-10-28 ·Soumya Sai Vanka, Hayato Yamakawa, Kazunobu Kondo, Yu Takahashi, György Fazekas
Abstract

Audio effect recommendation has mostly modelled style through reference audio and audio-domain losses. Because the parameter-to-audio map of parametric effects is many-to-one, such systems can match a reference without reflecting how a given engineer works. We instead model an engineer's style as a distribution over effect-chain parameters, which matches the data engineers can realistically share: console recalls store parameters without the audio they were applied to. Under a label-shift assumption, a personalised recommendation follows in closed form from a generic content model, a parameter population model and an engineer prior, with no paired engineer audio and no tuned weights. For vocal chains, the content signal is small but real, and concentrated in high-frequency EQ, dynamics and reverb. A forced-choice test with $14$ listeners found that the style-prior and content-only recommendations were distinguishable (89.3%), with results consistent with prior width affecting the balance between style and content, while parameter-space cosine does not predict perceived difference.

0.25

What do MLLMs know about 'Waldo'? Evaluating Visual Reasoning under Multi-Instance Multi-Object Challenges

cs.CV ·2026-10-28 ·Saumya Mundra, Ajoy Mondal, Vineeth N Balasubramanian, C. V. Jawahar
benchmarkcode/data
Abstract

Recent advances in Multimodal Large Language Models (MLLMs) have led to significant progress in visual reasoning capabilities. However, existing benchmarks curate image-question pairs pertaining to object categories with few instances, enabling trivial object referencing and limited visual ambiguity. In this work, we introduce MIMO-Bench, a benchmark for evaluating MLLMs in a Multi-Instance per Multi-Object categories (MIMO) setting. Each image-question pair in our benchmark pertains to instances belonging to object categories with multiple instances. The presence of such similar feature distractors causes hindrance in basic visual reasoning tasks, necessitating the referring expression to be composed of multiple visual subtasks to uniquely and unambiguously identify the target object. MIMO-Bench comprises 3,802 image-question pairs across attribute binding and numerical counting tasks. Despite impressive performances on existing benchmarks, both SOTA open-source (InternVL, QwenVL) and proprietary (Gemini, GPT-5) models exhibit substantially lower performance on MIMO-Bench, highlighting their limitations in the MIMO setting. The benchmark will be made publicly available.

0.25

Brain-Behavior Modeling from Multimodal MRI using Opportunistic Pretraining and Common-Cohort Adaptation

cs.CV ·2026-10-28 ·Md. Tanvir Rahman, Taslima Ferdaus Shuva, Nabil Anan Orka, Asaduzzaman Khan, Mohammad Ali Moni
Abstract · 39 pages, 4 figures. This work has been submitted to the Elsevier NeuroImage for possible publication

Incomplete imaging coverage limits sample sizes for multimodal cognitive prediction. We evaluated a two-stage opportunistic deep ensemble using baseline Adolescent Brain Cognitive Development Study data to predict fluid, crystallized, and total cognition. Separate neural models processed structural/diffusion, resting-state, and task-based MRI data with covariates. Models were pretrained on eligible modality cohorts (up to 9,686 participants), adapted to common training cohorts, and fused via Ridge regression (trimodal cohort: 4,216 participants). Five-fold family-aware cross-validation compared this approach against constituent and covariate-only models. Two-stage training increased mean test coefficient of determination (R^2) in 24 of 27 comparisons (median gain: 0.0081). Fusion increased mean R^2 by 0.0051-0.0333 over the strongest constituent and reduced mean absolute error by 0.0067-0.0976. Task-plus-resting-state and trimodal ensembles reduced absolute error across all targets (nominal p <= 0.0108). Task-plus-resting-state fusion achieved the highest mean R^2 for fluid (0.2222) and total cognition (0.3665), while structural-plus-resting-state led for crystallized cognition (0.3706), yielding R^2 gains of 0.0805, 0.1180, and 0.0890 over covariate-only references. Ablations showed expanded pretraining drove most improvements. Gradient-based attribution identified target-dependent input patterns. These findings support using partially observed cohorts for model development to achieve modest fusion gains. Pending external validation, this approach may support MRI-informed cognitive phenotyping.

0.25

LightGenBench: A Benchmark for 3D Emission Generation

cs.CV ·2026-10-28 ·Dongchen Yang, Xingguang Yan, Manolis Savva
benchmarkcode/data
Abstract · Project page: https://3dlg-hcvc.github.io/lightgenbench/

Recent generation methods produce 3D assets with detailed geometry and rich textures. However, these assets are non-emissive: they reflect light but emit none. We study a new task, emission generation: predicting spatially varying RGB emission over a 3D asset's surface from its geometry, non-emissive materials, and a rendered view. To support this task, we introduce LightGenBench. Our benchmark makes three contributions. First, we curate a dataset of 37k artist-authored emissive assets in UV, voxel, and multi-view form. Second, we establish baselines across four method families: UV-domain, sparse-voxel, multi-view, and segmentation-based generation. Third, we design an evaluation protocol that measures where an asset emits and how, complemented by user-study comparisons on renderings with varying emission strength. Across the four families, the sparse-voxel baseline TRELLIS.2-Emission performs best because generating emission natively in 3D avoids the failure modes of the other representations, such as the fracturing artifacts of UV space and the occluded emitters that multi-view generation cannot see. However, TRELLIS.2-Emission only reaches an IoU of 0.271, leaving the task far from solved. We hope LightGenBench establishes emission generation as a benchmarkable problem for 3D generation.

0.25

Guidance-Coupled Nullspace Drifting for One-Step Medical Image Reconstruction

eess.IV ·2026-10-28 ·M. Tuna Alatan, Agah Server Tapan, Şaban Öztürk, Muhammad Usama Mirza, Bilal Kabas +2 more
Abstract

Generative reconstruction can recover high-quality images from undersampled acquisitions, but state-of-the-art generative priors commonly rely on iterative sampling. Reducing the number of iterations lowers inference cost but also removes opportunities for progressive distributional refinement, making it difficult to retain image realism without sacrificing reconstruction fidelity. Drift modeling offers an alternative by shifting progressive refinement to training, thereby enabling one-step sampling without an iterative inference trajectory. Existing drifting methods, however, do not account for the imaging operator when modeling distributional drift. We propose DriftRecon, a generative method for one-step medical image reconstruction that performs drifting directly in the nullspace of the imaging operator. DriftRecon uses nullspace-restricted noise to construct training samples and a multiscale generator that incorporates measurement conditioning across resolution levels, with its generated component projected onto the operator nullspace. To control the fidelity--realism trade-off at inference, a conditional guidance parameter continuously steers the same generator between measurement-conditioned and unconditional generation. We derive a guidance-dependent similarity-kernel temperature schedule that preserves the guidance-weighted mass receiving strengthened supervision, thereby progressively localizing drift interactions as guidance increases to limit the associated loss of reconstruction fidelity. Experiments in sparse-view CT and accelerated MRI reconstruction show that DriftRecon outperforms established generative reconstruction methods and recent drifting models while producing reconstructions in a single forward pass.

0.25

Keepace.ai: A Technical Report on a Large Language Model for Sports and Health

cs.CL ·2026-10-28 ·Keepace. ai Team, Bowen Wu, Heng Wang, Jufeng Mi, Kang Liu +10 more
Abstract · 21 pages, 3 figures, 13 tables. Technical report

Large language models hold promise for delivering safe, scientifically grounded, and personalized exercise guidance at scale, yet the sports-and-health domain still lacks curated training corpora and rigorous evaluation benchmarks. Keepace (Keepace.ai) is a domain-specialized LLM built on Qwen3.5-27B through post-training, currently optimized for Chinese-language exercise guidance across three tasks: workout plan generation, sports-and-health knowledge QA, and personal sports data analysis. In internal rubric-based evaluations designed with domain experts, Keepace outperforms its base model and larger general-purpose baselines, with the most pronounced improvements in exercise safety, boundary compliance, structured output completeness, and scientific accuracy. This report summarizes the current design, data pipeline, safety-first evaluation framework, and evaluation results of Keepace.

0.24

MiDiff: MIDI-Conditioned Diffusion Models for Drum Audio Enhancement

eess.AS eess.SP ·2026-10-28 ·Elad Aspis, Sharon Gannot
Abstract · 13 pages. Submitted to IEEE/ACM Transactions on Audio, Speech, and Language Processing. Audio demos and code: https://midiff-paper.github.io…

Bridging the semantic gap between perfect symbolic musical structure and severely degraded acoustic signals offers a powerful paradigm for high-fidelity audio restoration. We introduce MiDiff, a diffusion-based generative model for drum audio enhancement that conditions the reverse process on temporally aligned Musical Instrument Digital Interface (MIDI) piano rolls. We extend the Score-Based Generative Models for Speech Enhancement (SGMSE) framework in the complex Short-Time Fourier Transform (STFT) domain by integrating a pretrained MIDI encoder, multi-scale temporal alignment, and three complementary conditioning strategies: early injection, Feature-wise Linear Modulation (FiLM), and cross-attention. We further show that providing the same piano roll only through classifier guidance, driven by an Automatic Drum Transcription (ADT) model on the audio-only score, yields no measurable gain: the symbolic prior is effective when it conditions the score network. To relax the requirement for ground-truth MIDI at inference, we further propose a hybrid reverse schedule that transcribes the diffusion state online with the same noise-aware ADT network and hands over from the audio-only score to MiDiff at a tunable switch time, cutting test FAD by 47% relative to the audio-only baseline without any symbolic input (switch time selected on the test set). We train our model on the StemGMD dataset using a non-additive degradation pipeline that simulates room reverberation with echo, electret capsule frequency response, and soft-clipping saturation with noise. MiDiff substantially outperforms the SGMSE baseline in terms of Fréchet Audio Distance (FAD) on both the validation and held-out test sets. Ablation studies demonstrate that combining all three conditioning strategies yields superior performance and that authentic MIDI velocity dynamics are critical for effective symbolic conditioning.

0.24

DRC-TTS: A Dual-Resolution Codec-Based Architecture for Lexicon-Free Malay Speech Synthesis

eess.AS ·2026-10-28 ·Tianhui Su, Tien-Ping Tan
Abstract · 27 page, 7 figures

Modern discrete neural audio codecs facilitate speech synthesis by replacing continuous acoustic representations with low-frame-rate discrete token sequences. For Malay speech synthesis, character-level input reduces reliance on pronunciation lexicons. However, combining character-level inputs with low-frame-rate discrete codecs introduces a temporal alignment challenge. When predicted character durations are quantized into frame counts at 12.5 Hz, durations shorter than one codec frame may be assigned zero frames. Consequently, the corresponding character representations may be lost during length regulation, potentially leading to pronunciation omissions. This study proposes DRC-TTS, a dual-resolution codec-based architecture for lexicon-free Malay speech synthesis. The model expands character representations at a temporal resolution of 50 Hz and subsequently downsamples them through strided convolution to 12.5 Hz for discrete codec-token prediction. Objective evaluations show that the proposed model reduces the character error rate (CER) from 27.11% for the baseline to 21.86%, an absolute reduction of 5.25 percentage points. The proposed model also achieves a mean opinion score (MOS) of 4.30, compared with 4.12 for the baseline. These results suggest that the dual-resolution design may improve character-level alignment and perceived speech naturalness under the evaluated conditions.

0.24

TheseusSyn: Unified Medical Image Synthesis via Context-Conditioned Masked Autoregression

eess.IV ·2026-10-28 ·Shibo Li, Venet Osmani
Abstract · 11 pages, 2 figures

High-fidelity medical image synthesis is promising for data-efficient learning and privacy-preserving model development. However, most existing approaches rely on symbolic or label-based conditioning, which constrains generative diversity and limits case-specific contextual fidelity. Inspired by the Ship of Theseus paradox, we propose TheseusSyn, a novel context-conditioned synthesis paradigm that generates images through iterative content replacement. Our framework employs a multi-codebook discrete tokenizer to convert medical images into sequences of token indices, extracted with a global representation token that serves as a contextual constraint. A masked autoregressive generative model then iteratively masks and predicts tokens conditioned on both local neighborhoods and the global context. By progressively replacing image content until the entire token sequence is renewed, TheseusSyn produces synthetic samples that preserve anatomical coherence and semantic consistency while introducing controlled diversity. We evaluate the proposed framework on multiple datasets from MedMNIST by constructing synthetic datasets with varying masking strategies and training ResNet-18 classifiers exclusively on either real or synthetic data. Experimental results demonstrate that classifiers trained on TheseusSyn-generated images achieve competitive performance on official test sets, indicating that the synthesized data preserve diagnostically relevant features and semantic consistency. Systematic analysis of masking strategies reveals their impact on structural fidelity and downstream classification accuracy. These results suggest that context-conditioned masked autoregression provides an effective and generalizable solution for label-free medical image synthesis, offering a scalable alternative to conventional data augmentation and class-conditioned generative models in medical imaging applications.

0.23

Component Translation and Judge-Labelled Refusal: A Multilingual Safety Evaluation

cs.CL cs.CR ·2026-10-28 ·Yongxi Zhou, Junwei Yao, Wenbo Ye, Jiahui Ge
Abstract · 12 pages, 1 figure. Ten-page manuscript prepared for the 4th Workshop on Large Language and Foundation Models (WLLFM 2026) at IEEE BigData 2…

Whole-prompt translation evaluates a request in different languages but does not compare which component carries a language. We introduce a paired reversal: translate the complete request component or the complete operation component into the same target language, holding behaviour, English frame and translated-span count fixed. Component identity and non-English text quantity change together. Assigning English, Chinese and Spanish independently to the two components yields nine configurations on 190 of 300 sampled harmful behaviours retained after consensus annotation and automatic gating. Refusal scored by language-model judges is the primary outcome; unsafe compliance is reported separately. On Gemini 2.5 Flash and DeepSeek, three refusal swaps are about 5.3 percentage points and the fourth is -0.5; none survives the pre-specified six-test Holm correction. In an exploratory extension, Mistral Small 4 has a 13.7-point Spanish swap (95% interval [8.4, 18.9]), which survives the twelve-test Holm sensitivity correction under the primary judge, and 12.1 points under a second judge. GPT-5.4-mini estimates are within 2 points of zero. Component assignment supplies an additional evaluation dimension. Evaluations should record component languages and translated text quantity, comparing matched behaviours at equal translated-span count while interpreting refusal and unsafe compliance separately.

0.23

Calibrating Track-Query Updates at Test Time for Multi-Object Tracking

cs.CV ·2026-10-28 ·Riku Inoue, Shogo Sato, Yu Mitsuzumi, Kazuhiko Murasaki, Tomoyasu Shimada +1 more
Abstract

Multi-Object Tracking (MOT) requires localizing multiple objects and maintaining their identities over time. Recent query-propagation trackers carry target-specific representations across frames and update them with current-frame image features, yet identity association remains difficult when visually similar targets overlap or cross. During such interactions, current-frame query updates may contain components directed toward nearby active tracks, which may affect identity association in later frames. To investigate this issue, we analyze track-query updates preceding identity switches (IDSWs), where the same target is assigned a different track ID. Our analysis shows that (i) propagated track-query representations change more abruptly before IDSWs, (ii) current-frame updates have larger positive projections along directions toward other active tracks, and (iii) bounding-box intersection over union (IoU) retrieves the counterpart track involved in a subsequent IDSW with high recall when that track remains active before the switch. Motivated by these findings, we propose Test-time Track-Query Calibration (TQC), a training-free and gradient-free method for query-propagation trackers. TQC calibrates the tracker output before propagation using a bounded correction estimated from update components directed toward nearby active tracks. It requires only track-query representations and bounding boxes available during inference, without retraining or architectural modification. Across nine DanceTrack and SportsMOT test settings, TQC improves HOTA and AssA in all settings and IDF1 in eight, with an average HOTA gain of +0.38 and limited inference overhead. Controlled variants on six validation settings further show that TQC's correction direction yields more consistent HOTA improvements than alternative directions with matched correction norms.

0.23

Quantum-Inspired Pairwise Ranking Aggregation via Active LLM Feedback

cs.IR cs.CL ·2026-10-28 ·Quan Mai, Susan Gauch, Gibson Nkhata
benchmark
Abstract

Ranking aggregation from multiple retrievers is a powerful yet under-explored technique in modern retrieval and RAG systems. We introduce LLM-IRA, the first extension of the quantum-inspired interactive ranking aggregation framework to document reranking. By leveraging large language models (LLMs) solely for pairwise judgments on high-disagreement document pairs, our method learns fusion weights in a principled, interactive way. The final ranking is a simple, ultra-fast pointwise linear fusion of scores, no full pairwise matrix or expensive listwise prompting is required. Experiments on the BEIR benchmark and TREC Deep Learning tracks with an open-weight instruction-tuned judge (Llama-3.1-8B-Instruct) show that LLM-IRA matches or numerically exceeds strong static fusion baselines on 5 of 6 evaluated benchmarks. We report two weight-update variants: momentum blending and a full accumulated-history refit, which together reach statistical significance (Wilcoxon signed-rank, p<0.05) over the best unsupervised aggregator on 4 of 6 benchmarks (DL19 under momentum; DL20, DBpedia, and NFCorpus under accumulated refit, which is the stronger of the two configurations overall). On the benchmarks we profiled (DL19, DL20), it already clears the best unsupervised baseline within as few as 5 LLM calls per query, well before exhausting its typical ~25--30 call budget. LLM-IRA also achieves cost-effectiveness comparable to state-of-the-art zero-shot LLM rerankers at an order of magnitude fewer LLM calls, and requires no training data or model fine-tuning.

0.22

Benchmarking decision models against generative large language models on medical multiple-choice questions

cs.CL cs.AI ·2026-10-28 ·Oscar Freyer, Daniel Glauert, Moritz Hofmann, Max Ostermann, Stephen Gilbert +1 more
Abstract

Medical workflows often require selecting among predefined options. We compared 13 decision models with 15 generative large language model configurations, including different reasoning settings, using 1,273 four-option MedQA questions under text-only conditions. Accuracy ranged from 67.71% to 89.08% for decision models and from 78.32% to 96.86% for generative configurations. Among the tested configurations, Mercury Decide had the lowest mean final-attempt cost (USD 0.0069 per 1,000 questions), while Clef Omni had the shortest median request duration (0.291 seconds). Decision-model probabilities distinguished correct from incorrect answers, but error rates among answers assigned probabilities of at least 0.90 ranged from 1.71% to 12.77%. Answer agreement across three runs ranged from 92.30% to 100%, while probability repeatability varied. Decision models offered resource advantages alongside lower peak accuracy; selecting models for medical workflows requires considering probability quality and repeatability together with accuracy and the resources needed to detect and correct errors.

0.22

CalibCause: A Quality-Gated, Interpretable and Calibrated Deep Learning Framework for Diabetic Retinopathy Grading

eess.IV cs.LG ·2026-10-28 ·Sohaib Sulman, Muhammad Atif Saeed, Akhtar Jamil, Alaa Ali Hameed, Roland Aydin
Abstract · 15 pages, 8 figures, 6 tables

Most deep learning systems for diabetic retinopathy (DR) grading are evaluated on a single data source, with limited evidence of cross-site generalisation, uncertainty calibration, and interpretability. We propose CalibCause, a quality-gated mixture-of-experts framework that combines per-expert visual explanations, label-free image-quality conditioning, knowledge distillation, and calibrated confidence with deferral. We evaluated the framework using 51,310 fundus images from EyePACS, DDR, and APTOS 2019, with IDRiD and DeepDRiD for external testing. Evaluation included pooled testing, leave-one-environment-out experiments, paired bootstrap confidence intervals, and local grade-threshold adaptation. On the pooled test split, the label-free distilled student achieved quadratic weighted kappa (QWK) of 0.863 and accuracy of 0.865 with 27.8 million parameters, 52% fewer than its teacher. Across held-out environments, the label-free model achieved mean QWK of 0.741 compared with 0.743 for a plain ConvNeXt-Tiny baseline, while adversarial invariance did not improve cross-site grading accuracy. Refitting grade thresholds using 200 labelled images increased held-out EyePACS QWK from 0.533 to 0.621 for the label-free model. After vector scaling, expected calibration error ranged from 0.025 to 0.036 across CalibCause variants, and deferring the least confident 20% of images increased student accuracy from 0.865 to 0.923 on the remaining images. These findings demonstrate that quality-aware grading, interpretable expert routing, model compression, and calibrated deferral can be integrated without sacrificing cross-site performance relative to a plain baseline, although reliable deployment still requires local validation and adaptation.

0.22

Detecting Fine-Grained Differences between Image Datasets

cs.CV ·2026-10-28 ·Leander Girrbach, Alexander von Recum, Kirill Bykov, Zeynep Akata
benchmark
Abstract

Detecting whether and in particular how two data distributions differ is a fundamental statistical problem. For example, we might be interested in which concepts (expressed in natural language) differ between two classes, the outputs of different generative models, or the same model with different prompt specifications. When working with large and complex image datasets, we intuitively think that first selecting concepts from a broad list using fine-grained patch embeddings, and then optimizing the concepts to better fit the data is an effective workflow. However, prior work instead uses MLLMs to propose concepts and image-level embeddings to select them. Therefore, we systematically compare these techniques and find that selecting concepts from broad lists can replace MLLM-proposals at obviously lower cost. MLLM inference can then be spent on optimizing selected concepts instead to achieve substantial improvements. Additionally, by designing a new benchmark that focuses on low-prevalence and multi-concept distinctions, we find that selecting nonredundant concepts instead of only the most distinguishing concepts matters, and we introduce techniques for how to effectively do this. Finally, we show how our methods address important problems like detecting spurious correlations and biases. Overall, our work presents a fundamental perspective shift on how we analyze interpretable distributional differences between datasets.

0.21

Fu-TCM: a multimodal large language model that learns traditional Chinese medicine diagnostic reasoning

cs.CL ·2026-10-28 ·Jiahang Xie, Tong Jin, Lingfei Wu, Yaping You, Jie Huang +7 more
Abstract

Large language models (LLMs) are advancing in multimodal understanding and reasoning, accelerating medical model development. Traditional Chinese medicine (TCM) uses evidence from the four examinations to determine disease location and nature, differentiate syndromes and guide treatment. This diagnostic process constitutes a complete chain of reasoning. We present Fu-TCM, a multimodal TCM LLM trained on data retained after cleaning and filtering an initial collection of approximately 7.12 million examples derived from classical texts, textbooks, medical images, clinical cases and publicly available datasets. The model was further optimized using Bianzheng-Grounded Policy Optimization (BGPO), which explicitly targets syndrome differentiation through rewards for response format, intermediate TCM reasoning fields and final syndrome predictions. Fu-TCM-27B led all six benchmarks spanning TCM reasoning, text and vision, achieving a macro-average accuracy of 81.01% and surpassing the strongest reference model with complete results across all six benchmarks by 7.13 percentage points. Its largest margin was 11.90 points on Fu-bianzheng. Case-level analyses further showed a positive association between intermediate-field accuracy and final syndrome-selection accuracy in most evaluated models. These findings suggest that TCM syndrome differentiation may be computationally learnable and that structured reasoning supervision can improve performance across TCM tasks.

0.21

Holistic Diversity Management in LLM-Guided Evolutionary Search

cs.CL ·2026-10-28 ·Seockbean Song, Youngsik Yoon, Sungjae Lee, Siwei Wang, Wei Chen +1 more
Abstract

LLM-guided evolutionary search iteratively selects existing programs, uses LLMs as mutation operators to generate a new program, and evaluates it to guide subsequent search. To preserve diversity and reduce the risk of premature convergence, it often adopts the island model, in which groups of programs, called islands, evolve separately. However, this structure alone does not guarantee distinct search directions, as islands may start from similar programs or converge toward similar solutions during evolution. This suggests that diversity must be considered throughout the search process, from initializing islands with distinct programs to developing island trajectories and sharing discoveries across islands. To this end, we propose HoliDivEvolve , a holistic approach to establishing and effectively utilizing island diversity throughout evolutionary search. Specifically, we introduce diverse initialization to establish different initial programs, trajectory adaptive island selection to allocate computation based on current score, recent growth, and the remaining search budget, and growth adaptive migration to share alternative ideas across islands. Experiments across diverse benchmarks show that HoliDivEvolve achieves competitive performance on most evaluated tasks. We further conduct controlled analyses to examine how our design choices affect search behavior and final performance.

0.21

RoleFUZZER: User-Behaviour Fuzzing for Role-Specification Adherence in Large Language Models

cs.CL cs.AIcs.SE ·2026-10-28 ·Kshitish Ghate, Michael Saxon, Tadayoshi Kohno, Aylin Caliskan
Abstract · Preprint under review

Large language models (LLMs) are often conditioned into specific roles such as tutors, interviewers, or agents acting for an organisation, through system prompts that specify how they should behave. The extent to which they can actually adhere to these role specifications under multi-turn interactions with realistic and potentially problematic users is understudied. To address this, we propose user behaviour fuzzing, a guided search over plausible user behaviours that can lead a model to violate its role specification, and introduce RoleFUZZER, a framework that adapts coverage-guided fuzzing from software testing to operationalise the search. RoleFUZZER constructs scenarios guided by a user-behaviour coverage grid that crosses a grounded taxonomy of user interaction types and the role specification of the model under test i.e. what is under pressure. With RoleFUZZER, we evaluate LLMs across five role specifications, including tutor, financial advisor, AI interviewer, antagonistic roleplay character, and web agent under organisational policies. Across simulated and validated conversations in 15 campaigns, the framework reaches >= 89% coverage of the 64-cell behaviour grid for every model and role tested, and surfaces specification-grounded failure patterns. At an equal budget of conversations, removing the RoleFUZZER grid and asking an LLM to generate specification targeting users free-form lowers coverage to 39-69% with fewer kinds of violations. The same specification can also fail in opposite directions. For e.g., on the antagonist, open-weight target models override the player's safety signals, while Claude Haiku 4.5 breaks character instead. RoleFUZZER offers a tractable way to uncover vulnerabilities in LLMs across role specifications by testing diverse user behaviours that are likely to violate them.

0.21

Cross-Device Transfer of a Hospital-Trained Fetal Head Circumference Model to a Low-Cost Handheld Ultrasound Probe

eess.IV ·2026-10-28 ·Patrick Gao, Lara Hart, Ken Sullivan
benchmark
Abstract · 24 pages, 5 figures, 4 tables

Automated fetal head circumference (HC) measurement reaches near-expert accuracy on the HC18 benchmark, but published models are trained and evaluated on hospital-grade equipment. We quantify what happens when such a model is carried to a low-cost handheld probe. A lightweight segmentation network (CSM, 0.13M parameters) was trained on HC18 across five independent runs and applied without modification to images from a $1,350 handheld probe (SonoQue C3), acquired at the same visit as the reference hospital scan in 18 patients. In-domain accuracy was 2.34 +/- 0.43 mm; on handheld images the same model gave 16.12 mm against the hospital record, a 6.9-fold degradation. Input preprocessing, a choice rarely reported, accounted for about a third of that gap: direct resizing gave 49.45 mm against 16.12 mm for an aspect-preserving pipeline. A leave-one-out data-efficiency curve showed that fine-tuning on two images significantly worsened accuracy, and that from four images upward error fell by 0.75 mm per doubling of the local dataset (p = 0.0045), although no training-set size up to 27 images significantly improved on the unmodified model. Much of the residual gap belongs to the reference rather than the model: scored against a manual tracing of the same image the model was already within 10.48 mm, and its disagreement with sonographer calipers on the same frame was a near-constant offset, negative in 10 of 10 cases. A single calibration constant estimated leave-one-out reduced that disagreement from 8.12 mm to 3.49 mm, reaching the intraobserver reproducibility reported for experienced sonographers, but did not improve agreement with the hospital record. Cross-device evaluations should therefore report the input preprocessing used and characterise the precision of their reference standard.

0.21

CoSTA: Co-Evolving Self-Play Timeline Agents for Complex Temporal Knowledge Graph Question Answering

cs.CL ·2026-10-28 ·Haikun Liu, Zhichao Yan, Shaoru Guo, Jiawei Gu, Victor Gutierrez Basulto +2 more
Abstract

Complex Temporal Knowledge Graph Question Answering (TKGQA) involves more than retrieving facts: the timestamp or validity interval of one fact can constrain the search for another, while multiple intervals may jointly determine the answer. Existing single-agent TKGQA methods typically investigate these facts within a single reasoning context, where evidence from different event histories can become interleaved and distractors can misdirect subsequent retrieval. Distributing investigations across agents can separate event-specific evidence into distinct contexts, but effective collaboration must still be learned because question decomposition, subtask investigation, and evidence synthesis are interdependent. To address these limitations, we propose \textbf{Co}-Evolving \textbf{S}elf-Play \textbf{T}imeline \textbf{A}gents (CoSTA), which organizes facts into navigable timelines, delegates investigations to workers, and uses a coordinator to synthesize their original timestamped observations. CoSTA initializes a shared policy through role-conditioned distillation and jointly optimizes both roles through co-evolving self-play with final-answer rewards. Across three benchmarks, CoSTA achieves state-of-the-art overall performance, with particularly strong gains on complex multi-fact questions, while its compact variants outperform the evaluated closed-source LLM agents.

0.21

Pronunciation-Supervised Text-to-Speech for Irish (Gaeilge)

cs.CL ·2026-10-28 ·William Lane
Abstract

Speech technology for Irish must be built from very little transcribed speech, yet the language's pronunciation is thoroughly documented in dictionaries and dialect descriptions. We ask whether that documentation can supervise the adaptation of a large pretrained text-to-speech model to a language it was never trained on. We adapt Qwen3-TTS to Irish using 12.8 hours of crowdsourced Common Voice speech, correcting two defects in the vendor fine-tuning script that training loss does not reveal (Appendix A), and reach character error rate (CER) 0.225 on held-out cross-corpus text, against 0.397 for an exhaustively tuned from-scratch system and 0.437 for zero-shot cloning. We then compile the Wiktionary pronunciation record into a lexicon and a phoneme critic, and use that critic as the verifier in one round of rejection fine-tuning: CER 0.159 cross-corpus and 0.067 in-domain, with the voice still anonymous across all 202 training speakers. The same critic explains why the voice cannot yet be steered to a minority dialect: 229 of 12,311 training clips lean Ulster against 8,451 Munster. Pronunciation documentation, converted into a verifier, is a practical lever for low-resource speech synthesis, and a way to measure what the available data cannot support.

0.20

MixCards: Evaluating Structured Descriptions of Multi-Source Sound Scenes

cs.SD eess.AS ·2026-10-28 ·Sripathi Sridhar, Yujin Kim, Giovana Morais, Liang-Yuan Wu, Mehmet Atilay Kucukoglu +4 more
benchmarkcode/data
Abstract · Benchmark dataset: https://doi.org/10.5281/zenodo.23702059 Code: https://github.com/Sound-Captioning-for-Accessibility/mixcardbench

Audio captioning benchmarks typically evaluate a single unstructured free-text description of an audio scene, providing limited insight into how a model fails to understand polyphonic audio. A plausible caption may omit a source, or conflate multiple sources into one, or assign an attribute of one source to another, and conventional captioning evaluation cannot locate these errors. We introduce MixCardBench, a benchmark for structured captioning of polyphonic audio, which evaluates models in two stages: first, identifying and temporally localizing the sources in a scene, and second, generating a structured MixCard describing the semantic, temporal, and acoustic properties of each source. The benchmark contains 4884 ecologically plausible synthetic soundscapes with up to nine overlapping speech, music and sound effects sources built from high-quality recordings and source-level annotations, where each source has a reference MixCard. We evaluate six state-of-the-art large audio language models and find substantial limitations in current models, which struggle considerably with speech and sound effects timing and detection, with the strongest model reaching only around 60% mean card-field score. Performance varies significantly across attributes, exposing capability differences that are hidden by aggregate caption scores. We assess the quality and ecological validity of MixCardBench through human validation, distribution matching and a listening study, and show that human corrections have little effect on model rankings, and our synthesized soundscapes are closer to in-the-wild YouTube videos than source guided but otherwise random synthesis. The benchmark dataset and code are publicly available, and demo examples are at https://mixcards-anon.github.io/mixcards-demo/

0.20

StepControl: Long-Horizon Human Motion Generation from Foot-Contact Schedules

cs.CV ·2026-10-28 ·Hojun Jang, Jinseok Bae, Donggeun Lim, Taesun Kim, Seungbin Jeong +3 more
published
Abstract · WACV 2027, 8 main pages

Controllable human motion generation has largely focused on specifying what action should occur or where the body should move. We study a complementary question: can full-body motion be controlled by specifying only when each foot contacts the ground? We introduce StepControl, an autoregressive diffusion model conditioned on a future binary left/right foot-contact schedule, without requiring target root and body poses, or semantic prompts. StepControl represents contact as explicit temporal events and injects them through dedicated cross-attention with classifier-free guidance. We further introduce a stochastic mixture noise schedule and warmup strategy for stable long-horizon autoregressive generation. Experiments show that StepControl follows prescribed contact timing more accurately than adapting existing spatially guided motion generators to contact constraints. It maintains effective contact control over long continuous rollouts and responds to edited schedules such as cadence changes, stop-and-go pauses, limping gaits, and bilateral jumping. Our results show that minimal foot-contact schedules provide an effective and intuitive temporal interface for controllable motion generation.

0.20

Automatic Iconographic Annotation of Medieval Illuminations: Comparing Generative Vision-Language Models and Supervised Discriminative Classifiers

cs.CV ·2026-10-28 ·Vincent Jolivet, Lucas Terriel, Chahan Vidal-Gor{è}ne
Abstract

The large-scale digitisation of medieval manuscripts creates a growing need for automated iconographic indexing. We evaluate two complementary paradigms for this task on 9,735 full-folio images annotated with 2,333 TIMEL concepts: generative Vision-Language Models (VLM) and supervised discriminative classifiers (SDC). The VLM experiments evaluate fine-tuning, structured JSON decoding, taxonomybased candidates generation, and deterministic reranking. The discriminative experiments evaluate flat, hierarchy-aware, and patch-attention heads over a frozen DINOv3 encoder. Our results highlight key functional differences: ML-Decoder (SDC) achieves the strongest overall retrieval performance (precision@1 of 0.696, MRR of 0.806), with loss calibration (Asymmetric Loss) serving as the primary performance lever. While SDCs provide superior precision and reliability within a closed vocabulary, VLMs offer strong semantic generalization and broad taxonomic coverage, particularly through structured JSON targets and verify-calibrate reranking (hierarchical F1 of 0.559). Furthermore, model behavior strongly depends on concept abstraction: SDCs excel at perceptually grounded descriptive concepts, whereas VLMs provide useful hypotheses for complex interpretative scenes. Finally, we find that enforcing explicit hierarchy within model architectures limits coverage compared to spatial feature aggregation. Overall, rather than competing approaches, SDCs and VLMs emerge as complementary components for future hybrid indexing systems combining structured retrieval with exploratory annotation assistance. As a methodological contribution, we define the soft F1 metric to provide a taxonomy-aware evaluation that credits semantically close predictions based on lowest common ancestors.

0.20

Residual Motion Splatting for Continuous Space-Time Video Super-Resolution

cs.CV ·2026-10-28 ·Vincent Herfeld, Eva Coupet{é}, Julien Mairal, Jean Ponce
Abstract

Continuous space-time video super-resolution produces a high-resolution, highframe-rate video from a low-resolution, low-frame-rate one at arbitrary spatial scales and temporal instants. Classical spatial multi-frame super-resolution has shown that detail beyond a single frame's sampling limit can be recovered by aligning complementary observations with sub-pixel precision and accumulating them on a finer grid. Existing learning-based methods largely depart from this principle: some align and fuse frames only at low resolution before regressing a continuous high-resolution representation, while others interpolate each frame to the output resolution before splatting. In both cases, the per-frame samples are resampled before fusion, so their sub-pixel positions are lost before frames are combined. Restoring this principle in a learned continuous model raises two challenges: estimating sub-pixel motion at unobserved times, and handling holes and occlusions when splatting a coarse grid onto a finer one. We address the first by predicting per-pixel continuous trajectories that refine linear optical-flow paths with smooth residual motion, using features integrated over the full video sequence. We address the second by directly splatting consecutive low-resolution features in a learned latent space onto a high-resolution grid at their predicted sub-pixel positions with predicted visibility weights, before a decoder resolves remaining ambiguities. This makes temporal interpolation and multi-frame superresolution a single operation. Our method improves both reconstruction fidelity and perceptual quality on multiple standard benchmarks over the state of the art, while running up to an order of magnitude faster.

0.20

WeVisCut: The Whole Editing Room in One Agent From Narration to Long-Form Video

cs.CV cs.AI ·2026-10-28 ·Chenyu Mu, Wenming Weng, Qinrui Wang, Xingsong Ye, Hao Yu +6 more
benchmarkcommunity pick
Abstract

Producing long-form videos from narration requires interdependent decisions that remain coherent across shots. Existing agentic editors often use multi-stage workflows driven by general models, leaving the quality-cost trade-off of compact, task-adapted policies unclear. We address this gap with WeVisCut, a long-form script-to-video system that connects tool execution with policy learning. Its agent harness maintains shared production state so that tool feedback informs subsequent decisions and repairs. Because finished videos do not expose these decisions, our four-stage data pipeline combines editing inversion with verified replay to turn recovered editing intent into executable supervision. Applied to 3,750 finished videos, this pipeline yields over 130K training examples. We use this supervision to train a single 9B policy that coordinates production and recovers from failures within the same harness. Moreover, we introduce a 100-task benchmark and five metrics assessing editing quality and delivery reliability. Under a shared tool environment, WeVisCut substantially outperforms its base model and achieves comparable performance to Claude Opus 5 while using nearly 33x fewer generated tokens per successful output minute. It also outperforms task-matched native editors on our proposed metrics.

0.19

Learning Scientific Trajectories to Predict Future Biomedical Discoveries

cs.CL cs.AI ·2026-10-28 ·Chia-Hsuan Chang, Brian Ondov, Jiang Lv, Lan Jiang, Yunmai Wang +5 more
Abstract

AI co-scientists have expanded the use of large language models for generating scientific hypotheses and guiding experimental investigation. Here we investigate whether temporal patterns in the semantic representations of complete biomedical conclusions can guide hypothesis generation. We introduce ForeSCITE, a framework that identifies thematic clusters of published conclusions, estimates a direction separating earlier from more recent findings within each cluster, and projects existing conclusion embeddings along these directions. An embedding language model trained to reconstruct conclusions decodes the projected coordinates into natural-language hypotheses. Across 407 clusters derived from approximately three million pre-cutoff conclusions, we identify temporal structure and illustrate its retention after decoding. In retrospective, time-sliced evaluations, ForeSCITE increases semantic coverage beyond the historical corpus, biomedical entity-pair novelty and post-cutoff pair re-observation relative to five prompt-based and retrieval-augmented baselines. Expert review of prioritized CAR-T hypotheses identified 37 candidates as both novel and biologically plausible, and expert justifications for five selected hypotheses illustrate their biological rationale and subsequent literature support where available. These findings support temporal guidance in semantic space as a scalable mechanism for proposing biomedical hypotheses for subsequent scientific evaluation.

0.19

A Tutorial on Generative Models for Music Audio

eess.AS cs.AIcs.ETcs.MM+1 ·2026-10-28 ·Mateusz Modrzejewski, Joanna Kurkowska, Dorien Herremans
Abstract

Music audio generation spans instrumental clips, full songs with vocals, edits of existing recordings, and systems that play alongside a musician in real time. This monograph explains the methods behind these systems for readers familiar with neural-network training who are new to music generation. It introduces neural audio codecs and autoencoders, autoregressive and masked token models, diffusion and flow matching, sampling acceleration, and preference optimization in a shared notation. Worked examples connect training objectives and sampling procedures to published systems. The application sections relate the musical requirements of instrumental generation, songs, editing, and live performance to model design. Comparison tables record representations, methods, inputs, tasks, and release information documented in papers and official sources. Symbolic composition, speech, and sound effects are included where they connect to music audio. We explain how to interpret automatic metrics and learned judges, design and report listening tests, and assess evidence about data provenance, memorization, attribution, and output detection. Selected legal decisions and disputes illustrate the implications for training and release. The reviewed systems combine established generative paradigms with new representations, planning stages, and post-training procedures. Differences in data, scale, and evaluation limit what cross-system comparisons can establish about the sources of reported gains.

0.19

Similar Scores, Opposite Failures: Human Correction as an Instrument for Video Caption Evaluation

cs.CV cs.CL ·2026-10-28 ·Ankit Khedia, Astitva Jaiswal, Chaitanya Prasad, Anisha Raju, Aangeeras
position
Abstract · 29 pages, 10 figures, 17 tables. Analysis code included as ancillary files. Data available on request at https://data.deccan.ai

Video captioning models are increasingly evaluated with metrics designed for single-sentence image captions. We test whether those metrics recover human judgement on dense, multi-window video description, and find that they do not. Six frontier models captioned 196 short clips. Trained annotators post-edited every caption against a failure taxonomy fixed before annotation, once as a whole-clip overview and once window by window, and a quality-control reviewer checked each annotation, yielding 1,144 overview verdicts and 4,672 window verdicts. Reference-overlap metrics do not separate the captions reviewers accepted from those they rejected: BLEU-4 reaches an AUC of 0.510 and a CIDEr-style tf-idf cosine 0.528, and restricting references to reviewer corrections leaves BLEU-4 at 0.479 (95% CI 0.431-0.529), indistinguishable from chance. Review granularity changes what is detected. The same annotators record camera and shot errors on 2.7% of captions when reading the overview and on 12.8% when reading window by window, a 4.7x gap that holds across every robustness check we ran, on the mode most likely to need a near-total rewrite. Correct-overview rates span 54.1-72.7%, but only 4 of 15 pairwise gaps survive Bonferroni correction (5 after adjusting for reviewer and clip effects), and accuracy and derived severity order the models differently. Models with statistically indistinguishable accuracy have sharply different failure profiles, which no single scalar shows. We argue that caption evaluation should record what was wrong rather than only how often, and we report the taxonomy, the per-model profiles, the corpus properties that predict difficulty, and the code that regenerates every number in the paper.

0.18

Do Audio LLMs HEAR Pain? Tracing Acoustic Evidence from Speech to Language Reasoning

cs.CL cs.AIcs.SD ·2026-10-28 ·Wenqi Li, Xiangjue Dong, Sulaiman Vesal
Abstract · 26 pages, 9 figures

Large audio language models (LALMs) can process speech directly, but it remains unclear how well they use the non-lexical acoustic information that speech carries. We study this question through pain, using TAME-Pain, which contains pain-neutral Harvard sentences and spoken pain ratings recorded at different self-reported pain levels. We introduce HEAR, a four-stage diagnostic framework that traces pain-related information from the speech signal to the model's prediction, and apply it to seven closed-source and open-weight LALMs. On the Harvard sentences, where pain can only be inferred from the voice, all models perform at chance level, and same-speaker references or few-shot examples improve balanced accuracy by no more than six points. On the spoken ratings, the same models reach 87.5-99.7% balanced accuracy. Layer-wise probing of open-weight LALMs shows that pain can be decoded from early audio representations (74.1% balanced accuracy for binary prediction and 45.0% for three-class severity), but that this information, along with most pain-associated acoustic cues, decreases in deeper layers. In an acoustic perception test, weaker models cannot distinguish even large changes in these cues, whereas stronger models can but still fail to infer pain. Finally, we recover pain-related acoustic cues and directly reinject them as structured text into the prompt, which improves pairwise pain prediction by up to 32 points. These results indicate that current LALMs encode paralinguistic information in early layers that is largely unavailable when the model produces its answer.

0.18

SpatialVec: A Unified Representation Learning Framework for Geospatial Objects with Boundary-Aware Sampling

cs.CV cs.DBcs.IRcs.LG ·2026-10-28 ·Samya Acharja, Kanchan Chowdhury
published
Abstract · Accepted to 2026 IEEE International Conference on Big Data as a full research paper

Spatial representation learning is essential for Geospatial Artificial Intelligence (GeoAI) applications, enabling machine learning models to capture the geometry, location, and spatial relationships of geographic entities such as points, polylines, and polygons. Existing spatial encoding methods often struggle to preserve fine-grained topological information, particularly for interactions involving complex geometries such as polylines and polygons. These approaches inadequately capture boundary structure and local spatial context, causing topology-sensitive relationships near geometry boundaries to be underrepresented in the learned embedding space. Consequently, accurately modeling complex spatial interactions remains a significant challenge for downstream spatial reasoning tasks. To fill the gap, we introduce SpatialVec, a boundary-aware spatial representation learning framework that adaptively samples geometry boundaries and their surrounding regions, capturing fine-grained geometric structures critical for topology-aware spatial reasoning. A transformer-based encoder-decoder trained through geometry reconstruction produces compact, geometry-aware, and unified representations for points, polylines, and polygons. Empirical evaluation on two real-world datasets shows that SpatialVec substantially improves the preservation of challenging topological relationships involving polygonal interactions and outperforms state-of-the-art baselines significantly on downstream prediction and classification tasks.

0.18

Aligning Misaligned Benchmarks for Single-View Novel View Synthesis

cs.CV cs.AI ·2026-10-28 ·Yang-Che Tseng, Erqun Dong, Jeong-gi Kwak, Gopal Sharma, Shakiba Kheradmand +3 more
Abstract

While single-view novel view synthesis methods have been dramatically enhanced recently, how to quantify their performance has yet to be standardized. In this work, we demonstrate that an important factor lies in how each method's own reference frame is aligned with the so-called ground-truth pose. Existing methods often rely on test-time optimization of target poses for this alignment, which inherently is ill-posed due to the complexity of the optimization landscape of camera poses vs render accuracy, leading to methods being evaluated at an arbitrary local optimum. Some recognize this limitation and thus rely on 3D reconstructions of entire scenes using training and test views to align poses, which may be infeasible. In this work, instead, we propose to use traditional 2D-to-2D and 2D-to-3D image matching techniques to align 3D reconstructions. We also evaluate only within co-visible regions, between different views, which is the only place where a `correct' ground truth can be defined. Our evaluations show that with proper coordinate alignment, novel-view synthesis methods that use explicit 3D Gaussians achieve the state of the art, despite being previously reported in the literature to be outperformed by pure image-based 2D methods. We further provide a simple baseline that directly utilizes an off-the-shelf video generator and a feed-forward geometry estimator. Our simple baseline is already comparable to state-of-the-art methods specifically trained for novel-view synthesis, without any task-specific training.

0.18

Knowledge-structured and evidence-aware semantic self-distillation for glomerular pathology foundation model

cs.CV ·2026-10-28 ·Qiming He, Jing Li, Zihan Wang, Tian Guan, Yifei Ma +22 more
Abstract

Visual foundation models learn rich pathological morphology but lack explicit concept-level supervision. Vision--language pretraining can provide this supervision but requires costly image--text pairs. Here, we introduce GloSAGE, a semantic self-distillation framework that uses lesion profiles predicted by a pretrained visual teacher to construct structured pathology descriptions. Reliability-guided selection identifies image--text anchors, while Knowledge-Structured Pathology Text Synthesis and Evidence-Guided Negative-pair Construction account for partially observed lesions. Image--text alignment trains projection layers while retaining the frozen visual encoder; an intermediate Semantic Transfer Layer supplies features for downstream tasks. Developed on GloMillion and evaluated across seven glomerulus- and case-level cohorts, GloSAGE supported zero-shot lesion recognition, image--text retrieval, hard-negative discrimination and concept-intervention analyses. It outperformed its source visual model in most lesion-recognition tasks and improved lesion grading, while supporting competitive case-level disease classification, injury grading and baseline renal function estimation. Ablations supported the combined contributions of knowledge-structured text and evidence-guided negatives, and the intermediate transfer layer improved mean lesion-recognition ROC-AUC relative to the complete projection output. A frozen language model conditioned on GloSAGE features generated glomerular descriptions. These findings suggest that teacher-derived lesion profiles can provide annotation-efficient semantic supervision for extending a glomerular visual foundation model.

0.18

DiffUltra: Task-Guided Pathology Synthesis for Large-Scale Real-World Lung Ultrasound Data

eess.IV ·2026-10-28 ·Gary Y. Li, Yu-Cheng Chou, Li Chen, Mohsen Zahiri, Priyam Patel +3 more
Abstract · This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may…

Despite their scale, large medical imaging datasets may underrepresent clinically relevant pathological presentations. Synthetic data can expand coverage, but findings must remain compatible with surrounding anatomy. This is challenging in ultrasound, where findings resemble surrounding tissue and must be positioned relative to anatomical landmarks. We propose DiffUltra, a task-guided latent diffusion pipeline that models pathological structure and spatial context, and screens synthetic images for downstream utility. DiffUltra conditions generation on structural masks decomposing findings into anechoic, hypoechoic, and hyperechoic subregions. Two placement strategies insert findings into pathology-absent backgrounds or replace existing findings with structurally matched pathology from other patients. A filter trained exclusively on real data retains images whose predictions agree spatially with target boxes. We evaluate DiffUltra for lung ultrasound detection of pleural effusion and consolidation, using over 700,000 expert-annotated frames from 598 patients at 12 sites. Averaged across three real-data scales, synthetic insertion significantly improved video-level AUROC for both pleural effusion (p = 0.017) and consolidation (p < 0.01). For pleural effusion, frame-level average precision improved by 4.5 percentage points on average and by 6.1 points using only 10% of real data (p < 0.01 for both). Synthetic insertion also yielded larger average precision gains than an existing diffusion-based lesion-insertion method at every scale. At full scale, targeted synthesis improved AUROC for both pathologies by 1.1-3.9 points in an underrepresented anatomical zone and on a held-out acquisition system. These results suggest that anatomically constrained, task-guided synthesis can help address coverage gaps in ultrasound training data.

0.17

Distilling the Shape of Uncertainty: What On-Policy Distillation Learns and How to Predict It

cs.CL cs.AI ·2026-10-28 ·Shiping Yang, Chuanqi Tang, Hossein Hajimirsadeghi, Layla El Asri, Mohammad Amin Shabani +1 more
code/data
Abstract

On-policy distillation (OPD) is increasingly adopted in the post-training of large language models. However, a stronger teacher may fail to improve the student. Prior work attributes this to a lack of new knowledge, but what that knowledge is remains unclear. We show that OPD learns not only \textit{how to branch}, i.e., which candidate tokens to consider, but also \textit{where to branch}, i.e., where to keep reasoning paths open. The existing token-overlap metric captures only the former. We introduce the entropy shape, the sequence of next-token entropies, to characterize the latter, and propose the Entropy Shape Correlation (ESC) to measure student-teacher agreement in entropy shape. Tracking ESC throughout training and intervening on a teacher's entropy while preserving its token preferences, we empirically show that \textit{where to branch} causally drives OPD gains, independently of \textit{how to branch}. Finally, we formulate OPD outcome prediction as a teacher selection task and establish an evaluation protocol for it, under which pre-OPD ESC selects the best teacher for both students and outperforms existing predictors. Code is available at https://github.com/maybenotime/opd-esc.

0.17

SelectSep: Target-Selective Video Queries for Audio Source Separation

eess.AS cs.MMcs.SD ·2026-10-28 ·Suyeon Lee, Doyeop Kwak, Joon Son Chung
Abstract · 5 pages, 2 figures, 2 tables. Submitted to ICASSP 2027

Video-queried sound separation (VQSS) aims to isolate a target sound from an audio mixture using its corresponding video as a query. However, scene-level video queries can become ambiguous when multiple sounding objects are visible, particularly when the target and interferer are semantically similar. We propose SelectSep, a target-selective VQSS framework that jointly uses text and a single click coordinate to specify the target source and instance without requiring segmentation at inference time. We adapt the visual encoder through target-feature distillation with text and click conditioning, and then fine-tune the separator using the adapted visual features. We also introduce an on-screen interference setting, where both the target and interferer are visible, to evaluate target selectivity. Experiments on standard benchmarks and the on-screen interference setting demonstrate consistent improvements in separation quality and target selectivity, particularly for mixtures containing semantically similar sources.

0.17

Cross Lingual Abstractive Summarization for Marathi

cs.CL cs.AI ·2026-10-28 ·Asang Dani, Shailesh R Sathe
Abstract

Text summarization is an important task for advancing natural language processing (NLP) research and applications in Indic languages. Despite progress in extractive and abstractive summarization, both monolingual summarization (MS) and cross-lingual summarization (CLS) remain relatively underexplored for these languages. Recent advances in Transformer architectures, pretrained language models (PLMs), and dataset availability offer new opportunities to address this gap. In this work, we construct cross-lingual training data from the monolingual Marathi MahaSum corpus and combine it with the smaller cross-lingual PMIndiaSum dataset to fine-tune IndicTrans2, an Indic neural machine translation model. We evaluate the resulting models using ROUGE, chrF and chrF++ metrics, focusing primarily on Marathi-to-English summarization due to computational constraints. We also discuss the applicability of our approach to the English-to-Marathi direction. Our experiments demonstrate that leveraging a pretrained translation model yields substantially better summarization performance than the IndicBART and mBART-50 based baselines reported in previous work on cross-lingual summarization.

0.17

PulmoWorld: An Asynchronous Multimodal Medical World Model for Longitudinal Pneumonia Dynamics

cs.CV ·2026-10-28 ·Cheng Tang, Xinyu Zhang, Jiawei Chen, Minghui Zhang, Shaocong Wei +12 more
Abstract · 49 pages, 17 figures

Pneumonia changes across respiration, inflammation, and organ function, while the measurements that describe it arrive at different times and are often incomplete. We introduce PULMOWORLD, a world model for longitudinal pneumonia dynamics that connects asynchronous state estimation, recursive simulation, and evidence-constrained planning. History-aware filtering separates state propagation from observation-driven correction, so a measurement updates the shared belief only when it is observed. That belief supports recursive forecasts of clinical values and modality representations, and summaries of future prognosis. Cap-MBRL searches imagined continuations inside an authorized action set and replaces a frozen fallback only when a candidate meets the risk budget and a predicted-cost margin. On the pneumonia cohort MASS, recursive forecasts are competitive with task-trained sequence models, and simulated future states retain prognostic information. Trained separately on public MIMIC-IV-CXR chest radiographs and clinical trajectories, this model improves prognostic discrimination over latest-value fusion in development and is competitive on long-horizon error on the locked pneumonia panel. In five known-mechanism MASS-Syn environments, Cap-MBRL lowers selection regret relative to direct fallback execution.

0.17

ConvRoute-ViT: Convolutional Token Routing for Efficient Vision Transformers

cs.CV ·2026-10-28 ·Manish Das, Shiv Ram Dubey, Satish Kumar Singh
Abstract · 5 pages, Conference

Vision Transformers (ViTs) achieve strong image recognition performance, but their self-attention increases quadratically with the number of input tokens and makes them computationally expensive. The standard implementation of ViT models treats all patch tokens with equal importance across every layer of the model. This uniform treatment introduces redundant computation and may reduce the model's focus on the most discriminative object regions. In this paper, we propose ConvRoute-ViT, a convolutional token routing mechanism for efficient vision transformers that performs token reduction through convolutional routing. Our model preserves the most relevant patch tokens according to class-token attention at selected layers, while the remaining tokens are spatially reorganized and compressed into a compact set of convolutional summary tokens. This design reduces the sequence length without completely removing the contextual information of the less informative tokens. By applying the routing process hierarchically, the proposed framework constructs compact and discriminative representations for deeper layers. Experiments on CIFAR-100 and TinyImageNet show that ConvRoute-ViT improves computational efficiency while maintaining classification accuracy compared with the base ViT.

0.16

BudSplat: Feed-forward 3D Gaussian Splatting under a Rendering Budget

cs.CV ·2026-10-28 ·Muyu Xu, Fangneng Zhan, Yu Wei, Ling Shao, Shijian Lu
Abstract

Feed-forward 3D Gaussian Splatting has recently emerged as a promising alternative to per-scene optimization, enabling fast novel view synthesis by predicting Gaussian primitives directly from input images. However, it quickly faces a bottleneck of rendering efficiency while scaling from sparse to dense multi-view inputs, which offer more geometric constraints but produce a rapidly growing set of candidate Gaussians. Existing solutions cannot reliably distinguish redundant primitives, leading to many redundant primitives from repeatedly observed regions and increased rendering cost without much gain in visual quality. We propose BudSplat, a compact feed-forward 3DGS framework that learns to keep the most valuable Gaussians under a limited primitive count. BudSplat has two novel designs. The first is a detail-aware importance score that is extracted from multi-level visual tokens, providing an explicit cue for regions where pruning is likely to harm perceptual quality. The score can be injected into voxel fusion to preserve high-value local contributors during multi-view aggregation. The second is quality-aware Gaussian pruning that selects a compact set of primitives by jointly considering rendering contribution and detail importance. The quality-aware pruning enables local quota redistribution that helps avoid over-allocating primitives to large redundant regions. Experiments on RE10K, DL3DV, and ACID show that BudSplat achieves a favorable trade-off among rendering quality and efficiency.

0.16

3DCodeEdit: Benchmarking Instruction-Guided 3D Program Editing

cs.CL cs.CVcs.GR ·2026-10-28 ·Binjie Wang, Xinran Zhao, Shoubin Yu, Xiang Li, Silong Yong +6 more
benchmark
Abstract

Language models increasingly create 3D content through code, with practical workflows often relying on editing existing programs rather than regenerating them, for speed and precise control. Such edits require understanding what a program builds and foreseeing how a change reshapes it, yet prior benchmarks mostly evaluate generation or simple edits against a reference. In this paper, we introduce 3DCodeEdit, a benchmark of 2,592 instruction-guided editing tasks over 108 Blender programs that construct objects entirely in code. Its edits range from parameter to structural changes, style redesign, and scene composition, and are judged from renders against human-reviewed checklists. Across 17 models, the average success rate is only 37%, and even the strongest models complete fewer than 60% of structural and style edits. As models become more capable, their failures shift to geometric foresight, where the code runs but builds the wrong shape. We therefore study whether multi-turn feedback can compensate for this missing foresight. Feedback from a separate reviewer model that describes how the renders differ from the request helps most, more than maximum reasoning effort or an agent harness, and can let a 27B model surpass the single-shot rate of a 2.8T-parameter model.

0.15

Deeper Backpropagation or More Updates? Training Looped Language Models Across Budgets

cs.CL ·2026-10-28 ·Jiarun Shen, Xiao Li
Abstract

Repeating transformer layers can improve language model performance without adding parameters. However, the added computational depth increases both forward and backward computation costs during training. Training such models requires choosing a backward window that controls how many final layer repetitions retain gradients during backpropagation. Longer windows improve per-token learning efficiency but increase per-update cost, while shorter windows reduce per-update cost but sacrifice per-token efficiency. Understanding which window performs better requires examining different training budgets. Prior work compares backward windows across computational budgets. When training budgets increase, it remains unclear whether the same window continues to perform best. We measure how the relative performance of two windows shifts as training budgets grow. Our method isolates per-token efficiency by training both windows on identical data, and compares complete training recipes at approximately equal training time at each budget. Results show that shorter windows lose much of their advantage as budgets grow. Controlled experiments show that both benefits decrease at larger budgets, with a greater decline in the benefit of extended training. Backward window comparisons depend on the available budget, and the preferred window among the tested candidates can change as the budget grows.

0.14

HEDGE: Gated Hypergraph Memory for Scalable Multi-Agent Debate

cs.CL ·2026-10-28 ·Amirhossein Abaskohi, Mahdi Mostajabdaveh, Zirui Zhou
Abstract

Multi-agent debate systems require agents to retain and revisit arguments across multiple rounds, yet limited context windows and unstructured memories make relevant information increasingly difficult to recover. Existing flat and pairwise-graph memories also struggle to capture higher-order relations, including multi-agent agreement, conflicting evidence, and evolving consensus. We introduce HEDGE, a selective memory framework that represents debate trajectories as semantic hypergraphs. Hyperedges connect related claims across agents, while intra-hyperedge relational graphs preserve fine-grained argumentative structure. A learned memory selection module (MSM), trained through supervised learning and reinforcement learning, routes each memory candidate to STORE, BACKUP, or OMIT, retaining useful information while limiting redundancy. Across seven diverse benchmarks, HEDGE consistently outperforms strong single-agent, multi-agent, and memory baselines, improving accuracy by 5.0 points on average over the strongest multi-agent system with GPT-4o-mini. The gains reach 8.2 points on conflict-heavy tasks, while HEDGE reduces memory usage by 48% and the number of debate rounds by 34%.

0.14

Statutory Article Retrieval from Arabic Judicial Narratives: Family Law and Law of Evidence in Saudi Arabia

cs.CL cs.AI ·2026-10-28 ·Zarah Shibli, Areej Alhothali, Lulwah Alharigy
benchmark
Abstract · Submitted to the MLNLP 2026 Conference

Reliable legal information retrieval must ground legal reasoning in authoritative statutory sources. In Arabic judicial settings, long factual narratives must be mapped to concise articles that express abstract legal rules, creating a narrative-to-legal-rule abstraction gap. Using a benchmark of 200 synthetic Saudi Arabic judicial cases and 381 articles from the Family Law and the Law of Evidence, we compare word-level and char-4 BM25, zero-shot dense encoders, per-law retrieval, dense-sparse reciprocal rank fusion (RRF), and a proposed Decompose-and-Refine (DaR) pipeline. In DaR, a large language model splits each narrative into disputed legal issues and generates, for each issue, three retrieval channels: a statute-style paraphrase, legal keywords, and a candidate taxonomy chapter. On the whole corpus, the instructed Qwen3-Embedding-8B dense encoder reaches an MRR@10 of 0.849, the best result of any non-decomposed system. Under per-law scoring, this advantage reflects Family Law only: restricted to Evidence articles, the same encoder drops to an MRR@10 of 0.087, and fusion with BM25 raises it only to 0.207. DaR changes this picture, raising Evidence MRR@10 to 0.696 and Family MRR@10 to 0.864. Without decomposition, the two laws favor different strategies; under DaR, the same retriever achieves the best MRR@10 on both.

0.14

Tracing the Influence of Retrieved Literature on Scientific Ideation in LLMs

cs.CL ·2026-10-28 ·Zhipin Wang, Koel Dutta Chowdhury, Christoph Leiter, Steffen Eger
Abstract · Updated preprint. Under review

Large language models (LLMs) are increasingly used to generate research ideas from scientific literature. Existing systems combine literature retrieval with search, generation, and refinement, making it difficult to determine how the retrieval strategy itself shapes the ideas generated. We isolate this contribution by comparing five retrieval strategies and a No Literature baseline under a shared generation protocol, using 45,000 ideas across 15 machine learning topics and ten open-weight models. Relative to no literature basline, all five retrieval strategies reduce diversity among ideas generated from the same set of papers by $5.6\%$--$17.9\%$, while most increase diversity by $10.2\%$--$18.1\%$ when ideas from different sets are pooled. Effects on idea quality and novelty are modest on average but vary across topics. We further apply a JSD-based attribution method to trace the influence of individual input papers on each idea's motivation, core idea, approach, and validation. In $79.3\%$ of ideas, only one or two distinct papers receive the highest attribution scores across these four parts. These leading sources often recur across ideas generated from the same set of papers, and groups with more varied source attribution contain more diverse ideas ($r=0.70$). These findings motivate various sets of literature and the use of attribution to guide the choice of sources for subsequent ideas.

0.14

Robust LLM-Generated Text Detection through Matched Conditional-Response Probing

cs.CL ·2026-10-28 ·Junfeng Hao, Wenxian Wang, Tao Leng, Xingshu Chen, Hao Ren +1 more
Abstract

Reliable detection of large language model (LLM)-generated text must remain effective under meaning-preserving text revisions, such as rewriting, polishing, and expansion. In practice, LLM outputs are often rewritten, polished, expanded, translated, or lightly edited before publication, weakening the lexical, syntactic, and local discourse cues used by many existing detectors. We introduce \textbf{M}atched \textbf{C}onditional-\textbf{R}esponse \textbf{P}robing (\textbf{MCRP}), a frozen-LLM probing framework that measures how a candidate text responds when a semantically matched condition is introduced, removed, or replaced. For each candidate, the frozen probe LLM first generates a short semantic condition and then compares token-level losses and prompt--target attention measurements under matched, null, and control conditions. It aggregates bidirectional conditioned alignment, counterfactual sensitivity, and internal attention coupling into a probe representation, on which it trains a lightweight supervised random-forest classifier while keeping the probe LLM frozen. Across standard detection, robustness to meaning-preserving text revisions, multilingual, and multidomain evaluations, MCRP achieves consistently strong performance. Under rewriting, polishing, and expansion, it reaches 93.97\% F1 and 96.24\% ROC-AUC, outperforming the strongest evaluated baseline by 19.50 and 25.77 percentage points, respectively. Component ablations show that the response feature groups make complementary contributions to predictive performance. In the condition-control study, matched conditions consistently outperform random, mismatched, and corrupted alternatives, demonstrating a measurable incremental benefit from semantic matching.

0.14

Two Birds with One Stone: Skill-Augmented Generation via Unified Retrieval and Utilization

cs.CL ·2026-10-28 ·Zihan Luo, Jiahui Kang, Wei Zhou, Qilin Fan, Mingyan Li +1 more
Abstract · Under Review

Skills, as reusable external knowledge, have demonstrated substantial potential to enhance the ability of Large Language Models (LLMs) to solve complex tasks. However, existing skill-augmented generation methods typically rely on separate models for skill retrieval and skill utilization, resulting in disconnected optimization of these two processes. To address this gap, we propose Skill-Uni, a unified reinforcement learning framework that jointly optimizes skill retrieval and utilization. Specifically, we adopt a generative skill retrieval paradigm that enables a single model to both learn how to retrieve relevant skills and leverage them for task solving. To effectively train this unified framework, we introduce a two-stage training strategy consisting of warm-up instruction tuning and GRPO, allowing the model to progressively improve its skill retrieval and utilization capabilities. Extensive experiments on five skill-related benchmarks demonstrate the effectiveness of Skill-Uni, yielding average performance gains of 66.59% and 34.82% over the base Qwen3-4B and LLaMA3.1-8B models, respectively. Besides, Skill-Uni is compatible with various skill retrieval engines, demonstrating its flexibility and generalizability. Additional in-depth analyses also provide insights into the effectiveness of Skill-Uni.

0.11

How Architecture, Typology, and Adaptation Strategies Shape ASR for Low-Resource African Languages: A Controlled Empirical Study

cs.CL cs.SDeess.AS ·2026-10-28 ·Mercy Muchai, Kevin Chege, Nick Mumero Mwangi, Samuel Chege Maina, Felermino Ali +2 more
Abstract

Despite recent advancements in ASR, identifying effective architectures, fine-tuning approaches, and data adaptation strategies for low-resource languages remains challenging. We present a systematic comparison across five low-resource African languages and Swahili as a mid-resource contrast, fine-tuning three models from distinct architectures: Phi-4-multimodal-instruct, Whisper-large-v3-turbo, and MMS-1B-all. We analyze data sufficiency, monolingual versus multilingual adaptation, domain fine-tuning, and language-adapter transfer. Phi-4 achieves the best overall accuracy, MMS shows comparable CER, while Whisper exhibits high CER, consistent with known hallucinations and repetition loops. Performance saturates at different data scales, offering practical data-efficiency targets, while typological complexity consistently shapes error rates. Domain fine-tuning benefits Phi-4 and Whisper but degrades MMS, revealing limitations of adapter-based adaptation. Proxy language adapters outperform training MMS adapters from scratch.

0.09

What Hallucinations Have We Measured? Semantic Illusion, Confabulation, and Expressed Uncertainty

cs.CL ·2026-10-28 ·Hazel H. Kim, Jakub Binkowski, Ekaterina Fadeeva, Juan David Guerra, Yijie Hao +20 more
position
Abstract

Hallucination remains one of the most critical challenges in deploying large language models (LLMs) reliably and safely. Yet the term is used inconsistently across the research community. We argue for a strict definition of hallucinations in LLMs. First, hallucinations should be confident incorrectness, kept separate from expressed uncertainty (UNK)1. UNK reveals that the model may be wrong, whereas a hallucination does not reveal it and is therefore more dangerous in practice. Conflating the two obscures their distinct behaviors and makes them hard to study. Second, hallucinations should be split by a simple re-elicitation test: does any semantically equivalent prompt recover the correct response? This test probes whether the relevant knowledge2 is accessible. If none does, the knowledge is likely absent or inaccessible, and the error type is a confabulation (CF) that likely requires external knowledge or retraining; if some does, the knowledge is accessible from the model but applied inconsistently, and the error type is a semantic illusion (SI) that simple prompt or decoding changes can mitigate3. Re-evaluating 14 recent hallucination methods under this taxonomy yields three findings. (1) Detection methods flag UNK more often than hallucinated responses (i.e., SI and CF). (2) Mitigation methods are far more effective for SI than for CF. (3) Latent signals are more informative than output-level signals for CF. An audit of recent hallucination papers shows that most conflate these categories. We recommend reporting CF, SI, and UNK separately, and introduce a practical labeling protocol that needs only a modest number of semantically equivalent prompts.

0.09

Leaving the Cave by Breaking the Symmetry: Identifiable Structure Can Enable Cross-Lingual Generalization

cs.CL ·2026-10-28 ·Frederick Riemenschneider, Anette Frank
Abstract · 12 pages, 7 appendix pages, 1 figure, 10 tables

Why do multilingual language models generalize across languages even without shared vocabulary? The Platonic Representation Hypothesis posits that models trained on different data, objectives, and modalities converge toward a shared statistical model of reality. Extending this intuition to representations inside a single multilingual model, we propose the Relational Identifiability Hypothesis: relations in a shared world can promote entity alignment by ruling out alternative cross-lingual correspondences, making corresponding entities more likely to become directly readable without an additional transformation. We test this hypothesis in a controlled bilingual Transformer whose two token-disjoint languages describe the same modular-arithmetic world. We find that the languages can develop a shared answer subspace without aligning corresponding numbers. Cross-lingual alignment emerges selectively when the available relations constrain number correspondences: subtraction makes zero task-level identifiable, inducing selective alignment, while an order-sensitive operation produces broad alignment across numbers. These patterns predict success in zero-shot cross-lingual transfer. Together, our results show that relational structure in a shared world can promote entity alignment and enable transfer.

0.09

V-CAPE: Visual Consistency Assessment for Product Image Editing

cs.CV ·2026-10-28 ·Iaroslav Melekhov, Roberto Alcover Couso, Francesco Taioli, Daniel Coelho, Artur Bekasov +2 more
benchmark
Abstract

Recent advances in generative image editing enable complex transformations such as background replacement, object manipulation, and pose adjustment from natural language instructions. However, verifying that these transformations preserve visual consistency, defined as the preservation of an object's intrinsic visual attributes, including geometry, texture, color, and fine-grained details, remains a largely unsolved problem. This challenge is particularly pronounced at scale, where manual inspection is infeasible and automated quality assessment (AutoQA) could become essential. In this work, we introduce Visual Consistency Assessment for Product Image Editing (V-CAPE), a large-scale dataset for evaluating visual consistency in generative image editing. The dataset consists of 40K unique source-target product image pairs, each annotated with binary accept/reject labels and structured failure modes capturing fine-grained inconsistencies such as geometric distortions, pose errors, and identity mismatches. V-CAPE combines real-world edited images with a synthetic subset constructed from 3D assets, enabling both realistic evaluation and controlled analysis. Vision-language models (VLMs) across all evaluated size categories on AutoQA achieve accuracy below 65%, indicating that the dataset poses a substantial challenge regardless of model scale. Our results highlight fundamental limitations of current models in verifying visual consistency under transformation, positioning V-CAPE as a benchmark for advancing fine-grained image editing evaluation.

0.08

CopyLoop: Fast Looped Transformer Decoding via Context Copying and Shallow Screening

cs.CL cs.AI ·2026-10-28 ·Wenyu Du, Stephen Chung
Abstract · preprint

Looped Transformers increase computational capacity through the repeated application of shared blocks, but their recurrent depth makes conventional full-depth decoding expensive. While existing speculative pipelines accelerate generation by processing consecutive token positions at varying depths, they are typically limited by sequential decoding overheads. We introduce CopyLoop, a training-free decoding algorithm that combines low-cost context-copy proposals with shallow screening and continuous recurrent scheduling. Rather than step-by-step generation, CopyLoop proposes multi-token suffixes from repeated context. It then evaluates them early at a shallow recurrent depth to efficiently filter out incorrect branches, processing newer blocks concurrently while earlier blocks are still progressing toward full-depth verification. We evaluate CopyLoop across six model checkpoints from the Ouro, Raven, and Huginn families, covering code generation and multi-step mathematical reasoning. CopyLoop achieves decoding speedups of 2.98x--10.70x over full-depth autoregressive decoding on Base models and 2.74x--3.52x on Thinking models, consistently outperforming or closely matching all competing baselines across evaluation tasks. Furthermore, ablation studies confirm a strong synergy between shallow screening and continuous scheduling, while our analyses show that CopyLoop requires only lightweight model calibration and yields greater speedups on longer generations with reusable contexts.

0.07

The Search for Statistical Indistinguishability Between Real and Fake Images

cs.CV cs.LG ·2026-10-28 ·Mohaimin Rafi, Otavio Moraes Rosa, Anshuman Chhabra, Utkarsh Ojha
Abstract

Images produced by generative models appear strikingly similar to real ones. In this work, we study whether perfection has been reached by generated content in terms of matching the properties of real data. We train deep neural networks to separate real data from fake and observe their test accuracies. Only a chance performance has any scope of proving statistical indistinguishability. However, since generative models are known to sometimes introduce small errors, e.g., low level checkerboard artifacts, we create progressive abstractions of the original real/fake images to potentially bring them to a space where those artifacts get wiped away. These include operations such as blurring, downsizing, cropping, converting RGB images into edge/semantic maps, etc. To our surprise, almost none of the abstractions are able to make fake data indistinguishable from real. The only cases when neural network based classifiers fail to separate the two categories is when the data is abstracted to such an extent that the original content seems mostly lost to the human eye, e.g., downsizing a $256 \times 256$ RGB image of human face to $4 \times 4$ resolution. We observe this trend across four different datasets of varying complexity (MNIST, CelebA, FFHQ and GenImage) and across various types of generative models (GANs, VAEs, diffusion models). This shows that the nature of fake artifacts extends all the way from low level to very high level in images. The simple implication is that fake images are much more different from real than we are used to think.

0.07

Author-Informed Stance Detection: A Dataset and Findings

cs.CL ·2026-10-28 ·Ogen Schlachet Drukerman, Einat Minkov
dataset
Abstract

Stance detection in social media is challenging, as interpreting a text often requires understanding the author's perspective. We re-frame political stance detection as an author-informed text understanding task, where it is desired to infer both the author's political leaning and their stance towards the target from the text. We first show that human annotators struggle to infer authors' political leanings and stances from individual tweets, and that pretrained large language models (LLMs) perform below human level in this setting. We then demonstrate that author context improves stance detection for both humans and LLMs, with particularly pronounced benefits for tweets perceived as sarcastic. Through extensive experiments, we further show that LLMs can be effectively fine-tuned to predict authors' political affiliations, and that author-informed fine-tuning yields better stance detection performance than standard fine-tuning. This approach not only surpasses human performance but also improves generalization across datasets. To support further research on author-informed political stance detection, we release a large dataset of stance-labeled tweets enriched with author-context information.

0.07

Wav2Aug: A Drop-In Speech Augmentation Policy

eess.AS cs.SD ·2026-10-28 ·Gianfranco Dumoulin Bertucci, Mirco Ravanelli
Abstract

SpecAugment is the dominant speech augmentation policy, but researchers have studied it primarily for automatic speech recognition (ASR), and it operates on spectrograms, limiting its applicability to other speech tasks and waveform models. We propose Wav2Aug, a drop-in waveform augmentation policy that randomly samples two augmentations from a calibrated pool of waveform transformations. We calibrate the augmentation hyperparameters once and reuse them across tasks and architectures, eliminating task-specific tuning. Experiments on ten speech benchmarks show that Wav2Aug achieves ASR performance comparable to SpecAugment while improving over it on average across non-ASR tasks, yielding an average relative error reduction of 12.8% over the strongest compatible SpecAugment baseline. We further show that multi-view training provides additional gains, with two augmented views offering the best tradeoff between performance and computational cost.

0.06

Visually Irrelevant Hallucinations in Medical Vision-Language Models

cs.CV ·2026-10-28 ·Ke Zou, Nan Zhou, Tao Yang, Meng Liu, Linchao He +7 more
benchmark
Abstract

Medical vision-language models (VLMs) show strong performance in radiology report generation (RRG) and medical visual question answering (Med-VQA), producing free-form clinical descriptions and answers from medical images. However, they may generate clinically meaningful statements that lack visual evidence or contradict image content. We refer to these errors as visually irrelevant hallucinations and investigate them at the phrase level. We propose CLEAR (Counterfactual Localized Evaluation of Alignment and Response), a model-agnostic post-hoc framework for phrase-level hallucination risk estimation. CLEAR decomposes generated responses into visually assessable clinical phrases. For each phrase, phrase-conditioned grounding identifies a candidate visual region, which is selectively perturbed by a localized counterfactual intervention. CLEAR characterizes the intervention using two complementary auditing cues: response cues, capturing semantic and clinically meaningful changes in the generated response, and alignment cues, measuring changes in image-phrase compatibility. These cues are normalized and jointly modeled to estimate phrase-level hallucination risk. Across four datasets spanning RRG and Med-VQA, CLEAR outperforms state-of-the-art hallucination detection baselines across multiple medical and general-purpose VLMs. On MIMIC-CXR, CLEAR improves AUPRC from 85.10% to 90.23%. Its effectiveness is further validated on a GT-derived controlled VIH benchmark. These results establish visually irrelevant hallucination as an important fine-grained reliability failure mode in medical VLMs and demonstrate the value of model-agnostic visual-support auditing for more trustworthy multimodal clinical generation.

0.06

ATI-OPD: Adaptive Teacher Intervention for Stable Long-Sequence On-Policy Distillation

cs.CL ·2026-10-28 ·Jiannan Guan, Youxin Jiang, Yutai Hou, Qiguang Chen, Yirong Zeng +7 more
Abstract

On-policy distillation (OPD) trains a student on its own responses using token-level feedback from a teacher. As responses grow longer, the student's prefix can depart from the teacher's own reasoning, weakening its feedback on later tokens. Updates based on this feedback can reinforce the same deviations in later responses and destabilize training. Prior approaches such as length control and loss reweighting limit learning from problematic responses but leave their prefixes unchanged. Fixed cutoffs can also discard later reasoning. We propose \textbf{Adaptive Teacher Intervention for On-Policy Distillation (ATI-OPD)} to improve subsequent teacher guidance by revising the student's reasoning prefix. ATI-OPD replaces the earliest high-disagreement segment with a short teacher segment, then lets the student resume its own reasoning from the revised text. To handle renewed disagreement, it excludes persistently divergent suffixes from the distillation loss while retaining useful student continuation. Training diagnostics show less teacher-student disagreement, fewer gradient spikes, and fewer extremely long responses. Across five mathematical reasoning benchmarks and two teacher-student pairs, ATI-OPD improves average student-only accuracy over standard OPD.

0.03

Security & Privacy 24

FGSM Attack Transferability from Simulation to Quantum Hardware in Network Intrusion Detection

cs.CR cs.ETcs.LG ·2026-10-28 ·Zeeshan Ajmal, Kimmo Halunen, Marc Maußner, Volker Reers, Arif Ali Khan
top venuecode/data
Abstract · 8 pages, 2 figures, 5 tables. Accepted at the 2026 IEEE 2nd International Conference on Quantum Artificial Intelligence (QAI), 13-16 Decembe…

Quantum computers are now available through cloud platforms, and recent work has proposed using them to detect network intrusions. It is not yet clear how easily such systems can be attacked. We train a hybrid variational quantum classifier (VQC) on the Friday DDoS subset of CICIDS2017 and test it against two attacks: label flipping during training and FGSM evasion at prediction time. We run the tests in simulation and on IBM hardware, the Fez superconducting processor (Heron r2). We find that two models with similar accuracy can differ greatly in how easily they are attacked. A model that decides with little confidence loses half its detections to a very small change in the input, while a model trained to decide with more confidence needs a much larger change before it fails. Attacks built in simulation against the more confident model also work on real hardware, and the device classifies every unaltered attack sample correctly. Simulated device noise does not reduce accuracy, but it does reduce the model's confidence and makes attacks easier, so accuracy alone does not show how exposed a model is. Label flipping has no measurable effect until 15% of attack labels are changed. These results show that device noise should not be treated as a defense and that reported weaknesses in these classifiers depend partly on how they were trained.

0.46

Manifest-Assisted Retrieval in Artificial Intelligence Search: Evidence Gaps and a Security Evaluation

cs.CR cs.AIcs.CEcs.CL+2 ·2026-10-28 ·Gangzhen Qian, Kainan Zhou, Chuhong Xu, Hang Xiao, Zhi Jing +3 more
published
Abstract · 8 pages, 3 figures, 5 table. Accepted at IEEE FLLM 2026. Author accepted manuscript. Reproducibility artifact: https://fllm-conference.org/2…

The external sources retained in our non-systematic audit did not isolate the incremental security or utility effect of consuming /llms.txt while ordinary retrieval remained fixed. The initial audit covered 39 sources through August 16, 2026, with later targeted updates through October 14. A purposive 15-tool study observed no automatic requests in 45 trials at one endpoint; a non-probability log study found that 97% of domains with detected files received no recorded requests in May 2026. We separate publication, fetch, model use, and action; define five operational threat paths and paired security-utility estimands; and exercise deterministic pre-model controls with 30 author-defined parser fixtures. We also execute a reduced pilot: 24 tasks across eight sites, five conditions, and 120 real model responses, with ordinary evidence held fixed. Frozen automatic scoring identifies attacker-target adoption in $21/24$ raw adversarial responses, with one further response unresolved; the adversarial URLs-only condition has no known adoptions and one unresolved response. The pilot measures output changes conditional on supplied evidence, without live retrieval or external actions. Together, the framework and pilot distinguish manifest-content effects from publication, discovery, and downstream execution, while leaving full retrieval and agent evaluation open.

0.46

How Deterministic SAST and AI Security Agents Divide the Work: Coverage, Harness Policy and Cost in Snyk VulnBench 2.0

cs.CR cs.AIcs.SE ·2026-10-28 ·Liran Tal, Chris Liu, Charlotte Fereday, Sebastian Roth, Brian Clark
benchmarkcode/data
Abstract · 44 pages, 21 figures, 7 tables. Benchmark harness, reference sets and all run data: https://github.com/snyk-labs/snyk-vulnbench ; companion …

We ask what AI coding agents and AI security harnesses find that deterministic static application security testing (SAST) does not, and what they miss that SAST catches. VulnBench 2.0 is a benchmark of 108 hand-curated, attacker-reachable vulnerabilities in 20 open-source web applications in Go, JavaScript/TypeScript, Python and Java; 62 flow across more than one file. A finding is credited only if it locates the vulnerability's attacker-controlled source and its dangerous sink, each within two lines. We evaluate nine productized configurations under one blind protocol: the Snyk Code static analyzer, Claude Code's /security-review with three model and effort settings (one with the Snyk MCP server), Codex Security with GPT-5.6 Luna, Terra and Sol, and the DeepSec scanner with two model backends. Snyk Code reached 82.4% macro recall (89 of 108) and Claude Code with Opus 5 at xhigh effort 79.3% (85 of 108); they disagree on 28 vulnerabilities, split 16 to 12, too few to tell them apart. Neither engine was sufficient alone: together they reached 101 of 108, the ceiling of all seven comparable configurations combined, and the AI configurations also reported at least seven real vulnerabilities that the reference set did not contain and Snyk Code did not report. What an AI engine misses depends on its harness as much as on its model: without Claude Code's built-in review policy, the same model matched 16 instead of 9 of the 20 vulnerabilities in the classes that policy excludes or restricts (supplementary run on 19 of 20 applications); Codex Security with GPT-5.6 Sol located 80 vulnerabilities at file level but 34 at line level; DeepSec's model-free file selection kept a file from only 56 of the 108 flows. Coverage and cost were only loosely related. The harness, fixtures, reference sets and all run data are public.

0.39

Recurrent Bipartite Graph Neural Networks for QC-MDPC Error Decoding in McEliece Cryptosystems

cs.CR cs.LG ·2026-10-28 ·Parth Rajeshkumar Tandel, Elisabeth Fink
Abstract · 19 pages

As the advent of large-scale quantum computing threatens classical encryption, Quasi-Cyclic Moderate Density Parity-Check (QC-MDPC) codes have emerged as a foundational element of Post-Quantum Cryptography (PQC). However, decoding these relatively dense codes using classical Belief Propagation algorithms often leads to failures caused by short graph cycles that stall the decoder. In cryptographic applications, these failures are severe security vulnerabilities that leak secret key information. To address this, this paper proposes a Recurrent Bipartite Graph Neural Network (R-BGNN) as proof of concept. By integrating Gated Recurrent Units (GRUs) into the message-passing framework and employing a multi-topology curriculum training strategy, the R-BGNN actively learns to break limit-cycle oscillations. Evaluated across 20 unseen matrix topologies, the proposed architecture significantly outperforms classical Bit-Flipping and Min-Sum baselines, reducing the Decoding Failure Rate (DFR) by 63.7\% at moderate noise levels ($t=10$) while converging in fewer iterations.

0.38

Trustworthy Distributed Machine Learning in Cloud-Edge-IoT Systems: A Heterogeneity-Aware Survey of Security, Privacy, and Robustness

cs.CR cs.DCcs.LG ·2026-10-28 ·Long Dang, Ali Alfatemi, Kaiqi Xiong, Mohammed Aledhari, Abdellah Chehri +2 more
publishedsurvey
Abstract

Distributed machine learning (DML), including conventional distributed training and federated learning (FL), enables collaborative model development across cloud-edge-IoT systems without fully centralizing raw data. However, distributing data, computation, updates, and orchestration increases the attack surface and introduces coupled challenges across privacy, robustness, reliability, fairness, and governance. This survey presents a security-centered, heterogeneity-aware roadmap for trustworthy DML, with FL treated as a prominent high-risk case. We first characterize the assets, adversaries, trust assumptions, and coordination patterns that define DML/FL security boundaries. We then analyze how statistical, system, and trust heterogeneity, including non-IID data, resource asymmetry, intermittent participation, and cross-silo/cross-device trust differences, affect convergence, fairness, attack detectability, and defense reliability. Next, we synthesize attack-defense interactions across poisoning, backdoors, Byzantine manipulation, privacy leakage, model extraction, network/orchestration attacks, Sybil behavior, and strategic participation. We further compare layered defenses, including robust aggregation, drift-aware optimization, differential privacy, secure aggregation, homomorphic encryption, secure multiparty computation, trusted execution environments, watermarking, provenance, authenticated communication, and monitoring. Finally, we discuss deployment, evaluation, governance, and open research guidelines, emphasizing reproducible, non-independent and identically distributed (non-IID) partitions, attack-aware benchmarks, privacy budgets, attack success rate, client-level disparity, latency, communication, memory, energy/carbon overhead, and auditable lifecycle artifacts.

0.32

Adversarial Process Disturbances at a Deployable Alarm Budget: Detector Evaluation on a Simulator and a Real Plant

cs.CR ·2026-10-28 ·Benjamin Blakely
Abstract

Industrial anomaly detection is evaluated largely on simulators, and recent work reports large drops, with unstable rankings, on real plant data. We ask whether that extends to the security case. We evaluate eleven detectors across classical, neural, dependency and causal families on a simulated chemical process and a sodium thermal-hydraulic facility: different processes sharing no equipment, but both continuous, at steady state, and perturbed only through temperature, pressure, flow and valve, never composition. Three conditions prior work does not cover: disturbances authored against a threat model rather than native faults; one construction applied to both processes; and detection read at a declared false-alarm budget. The drop persists throughout. The simulator panel looks deployable, neural matching classical; on plant telemetry detection collapses and the neural family trails, a gap that in-sample thresholds likely understate. Which detector wins transfers between the two processes; how often it fires does not. Two results hold on both. Sequential valve manipulation defeats the whole panel: at tier 2 no detector flags more than 0.17 of its perturbed samples, against 0.89 or better for most disturbances. And the plant's collapse is not merely that its disturbances are weaker, though they are by about twelvefold; matched on measured strength, the plant is never the better detected.

0.32

Enhanced image forensics through partial C2PA-based provenance information

cs.CR cs.AIcs.MMeess.IV ·2026-10-28 ·Riccardo Ziglio, Eleonora Marchesini, Andrea Montibeller, Giulia Boato, Silvio Ranise +1 more
Abstract · paper accepted to IEEE International Workshop on Information Forensics and Security (WIFS) 2026

Cryptographically verifiable C2PA-based Content Credentials (CCs) can be attached to freshly acquired or generated digital images but also to already existing ones with no previous provenance information, for instance after processing through trusted C2PA-enabled software tools. The latter case leads to images with so-called partial CCs, which capture only the last operations applied, while prior provenance remains unknown and can be investigated through signal forensics. This paper addresses the forensic analysis of such images, highlighting the opportunity to leverage the partial CCs as additional trusted knowledge to adapt forensic detectors towards the statistical characteristics of the image under analysis and mitigate potential distribution shifts. To this end, a two-step strategy is envisioned: (i) reconstructing the operations applied by the last signer through a structural analysis of the partial CCs; (ii) exploiting the obtained reconstruction to enhance the forensic detector of the analyst through an adaptation process. An empirical analysis on partial CCs produced by commercial C2PA-enabled software tools through different editing operations is carried out, providing evidence on the feasibility of the reconstruction in practical cases. An experimental case study on the forensic task of synthetic image detection is then presented, showing that the proposed strategy can significantly enhance the performance of state-of-the-art detectors on test images that are postprocessed via C2PA-enabled software tools after being generated/acquired.

0.28

STAR: Scalable TEE-based Anonymous Rate-limiter

cs.CR ·2026-10-28 ·Kosei Akama, Yoshimichi Nakatsuka, Koichi Moriyama, Keisuke Uehara
Abstract

Protecting Internet services from attacks such as TLS/HTTP floods requires rate limiting, yet identifying users to enforce these limits can compromise their privacy. Anonymous rate limiting promises to reconcile these goals, but existing schemes either struggle to scale or assume that services and authorities do not collude. We present STAR, which uses server-side trusted execution environments (TEEs) to overcome these limitations. By enabling lightweight cryptographic operations, TEEs make authentication efficient, but introduce a privacy challenge: malicious services can observe memory-access patterns to infer user secrets. We address this threat through countermeasures against secret-dependent memory accesses and assess our implementation using dynamic analysis. To extend these efficiency gains across servers, we also design a secure sharding mechanism that distributes client requests. We demonstrate that strong privacy guarantees can coexist with efficient authentication at scale; STAR processes authentication requests in tens of microseconds and handles approximately 13.9K requests per second per CPU core.

0.28

DeepDiff: Neuro-Dynamic Regression-Freedom Checking for Security Patch Validation

cs.CR cs.SE ·2026-10-28 ·Laboni Sarker, Shuvendu K Lahiri
datasetcode/data
Abstract · 22 pages, 4 figures

Automatically validating AI-generated security patches remains a significant challenge. Eliminating a reported crash and passing existing test suites do not provide sufficient confidence that a patch correctly addresses the underlying vulnerability without introducing regressions. We present DeepDiff, a technique for regression-freedom checking that differentially evaluates candidate patches against the original vulnerable program using non-crashing inputs that exercise the affected code paths. The expectation of regression-freedom is that a correct patch should eliminate the vulnerability triggering crash while preserving the intended behavior on non-crashing inputs. To support this evaluation, we introduce a dataset of correct and incorrect patch candidates for open-source memory-safety vulnerabilities from CyberGym. We explore different approaches for non-triggering test input generation to reach the patched code, a purely neural regression-freedom checker, and a hybrid approach that combines dynamic analysis with neural techniques. Our experimental results demonstrate the effectiveness of regression-freedom checking in validating security patches while outlining the challenges that remain.

0.28

SPARSEBLEED: Bypassing Attribute-Inference Defenses via Sparsity Optimizations

cs.CR cs.AR ·2026-10-28 ·Darsh Asher, Farshad Dizani, Joshua Kalyanapu, Ro Cammarota, Aydin Aysu +1 more
Abstract

Deep neural networks (DNNs) are widely deployed in safety- and privacy-critical domains such as healthcare and finance, where models are trained on confidential data. This paper presents the first hardware-level attribute-stealing attack that bypasses state-of-the-art privacy defenses, revealing that accelerator optimizations intended for performance can directly undermine data confidentiality. SparseBleed requires no access to gradients, logits, model weights, ground-truth labels, power traces, DVFS manipulation, shared caches, or other channels. Instead, we show that sparsity-driven execution paths encode stable timing signatures that act as a surrogate for model confidence, enabling high-fidelity attribute inference from a single query. SparseBleed generalizes across heterogeneous backends and hardware engines, including Intel AVX2/AVX-512/AMX, NVIDIA Ampere Tensor Cores with 2:4 sparsity, and nine well-known sparse and dense libraries such as Intel oneDNN, oneMKL SparseBLAS, and NVIDIA cuSPARSE. We evaluate SparseBleed to infer sensitive input attributes and true output labels. On Texas-100X, SparseBleed infers sensitive input attributes such as Sex with 66.39% accuracy and recovers the true output label, Surgical Procedure, with 99.8% accuracy. Finally, we evaluate mitigations. Constant-time execution by disabling sparsity optimizations is expensive (e.g., up to $2\times$ higher latency on NVIDIA Sparse Tensor Cores and $\approx 25\%$ higher power in AMX microbenchmarks). By contrast, a padding defense that equalizes responses to the worst-case time masks timing leakage with only 7.24% average performance overhead across nine libraries on Texas-100X/Adult and near-zero extra energy.

0.25

FlipJev: Fast Black-Box Red Teaming Finds Universal Adversarial Suffixes for Typed Decision Models

cs.CR cs.AI ·2026-10-28 ·Xiaogeng Liu, Xinyan Wang, Chaowei Xiao
code/data
Abstract · Pre-print. Code: https://github.com/SaFo-Lab/FlipJev, Webpage: https://flipjev.github.io

Typed decision models are an emerging class of language models that make decisions instead of generating free-text responses. They answer in a single fast and cheap query and are increasingly used, including in security-sensitive tasks. However, red teaming of these models leaves two gaps: existing adaptive attacks do not use their speed, and few studies have searched for universal adversarial examples. We fill both gaps on Jev, the representative typed decision model, with a quick black-box query-based attack that calls no attacker language model. We conduct extensive experiments on 12 datasets with different data types and application scenarios. The results show that Jev resists common prompt injections: no strategy changes a correct answer in 10,240 paired tests. In contrast, our attack finds short suffixes that flip confident answers within minutes. These suffixes are universal: one suffix per dataset flips 43% of unseen documents. Because typed decision models, unlike classic NLP models, can serve many tasks, we test whether these suffixes transfer across tasks. We find that one suffix generated from one task can flip 90-100% of the samples in other tasks. We therefore believe the evaluation of typed decision models must include strict adaptive attacks to fully reveal their robustness.

0.25

IDSForge: Self-Explaining Network Intrusion Detection Through Neural Behavioral Rule Learning

cs.CR ·2026-10-28 ·Hao Fu, Jinxia Wei, Wan Wei, Chun Long, Degang Sun +1 more
Abstract

Anomaly-based network intrusion detection systems (NIDS) can detect previously unseen attacks by identifying deviations from benign traffic. However, deep learning-based anomaly detectors typically rely on implicit decision boundaries, and post-hoc explanations may not faithfully reflect the detection logic. This makes alerts difficult to interpret and audit. To address this issue, we present IDSForge, an anomaly-based NIDS that directly learns explicit, executable behavioral rules from benign traffic, making its detection logic interpretable by design. IDSForge uses the same rule set for detection and explanation, ensuring explanation fidelity by construction. It uses an explicit distributional autoencoder trained only on benign samples to learn axis-aligned regions of benign traffic. A high-speed rule compiler then consolidates these instance-specific regions into a compact set of feature-level interval constraints. These rules can be executed directly by popular NIDS engines, supporting high-throughput inspection with low overhead. To adapt to changes in traffic, a self-distillation co-evolver updates the model and rule set incrementally when the false-positive rate on analyst-verified traffic exceeds a target threshold. Our evaluation shows that IDSForge reduces F1 error by $5.7\times$ to $47\times$ relative to state-of-the-art anomaly-based baselines. Deployed in Suricata, IDSForge sustains a throughput of 7.8 Gbps with a median per-packet latency overhead of 3.1 $μ\text{s}$.

0.23

Answer-Only Disclosure: Private-Context LLM Assistance over Web APIs with a Leakage Ledger

cs.CR cs.AI ·2026-10-28 ·Liang Yuan, Xiaohu Du, Ruikun Luo, Jinze Wang, Mashud Rana +4 more
Abstract · 13 pages

Web assistants increasingly answer questions about a user's private documents by pairing a small on-device model with a cloud LLM behind a Web API. In collaborative decoding, the document stays on the client, and the cloud receives only a public view that the user approves, such as the question alone, but the two models exchange messages while the answer is generated. We show that the message transcript can disclose private content that neither the public view nor the answer contains. In a planted-secret audit, speculative decoding with private drafts did not include the secret code in its answers, yet sent the code to the cloud in rejected drafts. We present ADLER, a decoding protocol for streaming LLM APIs in which every message to the cloud depends only on the public view and the answer the user sees, regardless of how the client combines the two models or how the cloud behaves. A client-side leakage ledger then bounds how much the answer itself reveals relative to a public reference process built from the cloud's replies. In the audit, no ADLER configuration exposed the secret beyond the answer. On GSM8K math problems with numerical values masked in the cloud's input, ADLER with an unlimited leakage budget raises the exact-match accuracy of a 0.5B client from 0.41 to 0.83 and of a 1.5B client from 0.67 to 0.85 under lenient answer parsing, close to the 0.92 achieved by uploading the full problem. It also exceeds a zero-leakage program baseline (0.69) by 0.14 and 0.16, respectively. In this baseline, the cloud writes a program that the client runs offline using the private numbers. The ledger records the disclosure charge associated with this gain; budgets of 16 nats or less leave accuracy near the cloud-only baseline on GSM8K.

0.22

WebMaze: Defending Against LLM-Powered Web Scanners with Functionality-Coupled Deception

cs.CR ·2026-10-28 ·Pengfei Zeng, Changhua Luo, Penghui Li
Abstract · 10 pages, 4 figures

Large language models (LLMs) have enabled more effective vulnerability scanning of web applications. However, these capabilities also make it significantly easier for attackers to automate cyberattacks on the web. In this work, we investigate how to divert malicious LLM-based web scanning, with preservation of genuine functionality as a deployment requirement. We systematically study six LLM-based scanners and summarize how they gather information and iteratively select actions based on observations. Guided by the study, we then present WebMaze, a defense that preserves genuine application functionality while diverting LLM-based scanners to a generated decoy world. WebMaze embeds functionality-coupled traps in genuine pages and constructs view-tailored interactive decoy worlds from the application's existing views. The traps blend into genuine components and expose entry points to decoy worlds that sustain exploration through connected pages and interactions without changing genuine functionality. Across 12 web applications and three LLM-based scanners under the default no-disclosure setting, decoy pages account for an average of 93.0% of the distinct pages requested per run. This demonstrates that WebMaze effectively diverts scanners away from genuine application pages and into the generated decoy worlds. Even when explicitly informed about the decoy mechanism and instructed to avoid decoy pages, scanners continue to explore predominantly decoy pages.

0.21

Session-Bound Behavioral Watermarking for Autonomous LLM Agents

cs.CR ·2026-10-28 ·Md Facklasur Rahaman, Yong Xiang, Md Palash Uddin, Shang Gao, Guang Hua +1 more
Abstract · 12 pages, 3 figures, 5 tables

Large Language Model (LLM) agents are increasingly acting through Web interfaces and external tools. They produce trajectories. Those trajectories may pass through several intermediaries. They can store, modify, or dispute those trajectories. In Web-based agent deployments, trajectories may pass through APIs, marketplaces, and tool servers. As such, a copied or disputed log may not by itself reveal which agent owner produced it. Behavioral watermarking can embed ownership evidence in agent action choices. However, existing methods mainly address detection, removal, rewriting, or imitation. They do not directly test whether repeated observation of watermarked trajectories enables cross-session forgery by exploiting learned action preferences. We formulate this learned-forgery threat and propose AgentAnchor, a session-keyed behavioral watermark for owner attribution from trajectory logs. We derive each step key from a secret key, a fresh public session identifier, and semantic states of recent observations, while a guarded digest of the action-policy state diversifies linear coding rows. It embeds parity evidence through a KL-bounded exponential tilt, and a trusted verifier checks a disputed log without rerunning the agent, using wrong-key calibration and abstention.

0.21

Complexity of Differentially Private Selection via Federated APIs

cs.CR cs.DS ·2026-10-28 ·Hilal Asi, Vitaly Feldman, Jelani Nelson, Huy L. Nguyen, Kunal Talwar +1 more
published
Abstract · NeurIPS 2026

Differential privacy (DP) has become the standard for privacy-preserving data analysis, yet its practical implementation in federated or distributed settings often faces a significant utility gap compared to the central model. While existing frameworks, including local, shuffle, and secure aggregation models, offer vetted primitives, they can incur high error for important tasks such as selection. Our first contribution is to show that this is inherent in the aggregation model: any private selection algorithm relying on the standard noisy aggregation primitive requires $Ω(\sqrt{d}/\varepsilon)$ samples. To address this limitation, we propose adding a new primitive that implements a version of the Sparse Vector Technique (SVT) and argue that it has simple implementations under various trust models. We demonstrate its power by developing an $\varepsilon$-differentially private algorithm for the selection problem. For selection in dimension $d$, our algorithm achieves an additive error of $O\left(\frac{\log(dn)}{\varepsilon}\right)$, matching that of the central model when $n$ is polynomially bounded in $d$. Furthermore, we show that this can be done using a near-optimal expected number $O(d\log^*d)$ of SVT queries. Together with existing lower bounds for the shuffle model, this result establishes an exponential separation between the SVT model and shuffle or aggregation-based models. We demonstrate the model's broad applicability to diverse tasks such as $F_p$ loss minimization, heavy-hitter identification, and sparse mean estimation.

0.20

DecisionUnlearn-Bench: Unlearning Probabilistic Decisions under Changing Schemas

cs.CR cs.AI ·2026-10-28 ·Tiantong Wu, Wei Yang Bryan Lim
Abstract

Machine unlearning aims to reproduce the behavior of a model trained without designated data. For probabilistic decision models, this target also depends on the candidate options supplied at inference. We introduce DecisionUnlearn-Bench, an empirical framework that compares full, unlearned, and retain-only models across deletion granularities, label renderings, and candidate sets. Across 20 eligible model configurations on the BANKING77 dataset, full-to-oracle displacement increases consistently from example to association to class deletion. Replacing semantic option labels with opaque codes nearly eliminates the retrained oracle's recognition of deleted classes, even with successful acquisition of the full model, whereas changing the candidate set alters both the oracle and the measured unlearning progress. SimNPO satisfies the screening criterion in 19 of 20 configurations, but its gains vary substantially across seeds and deletion targets. Development tuning on the CLINC150 dataset does not establish a positive pooled effect across 48 intents. At approximately matched suppression, methods can differ markedly in oracle fidelity. Recovery experiments further distinguish initial forgetting from subsequent adaptation and show conclusions that depend on both architecture and metric. These results show that evaluating decision-model unlearning relative to a schema-specific retraining target, with separate reporting of effect size, utility, robustness, and recoverability.

0.20

The use of post-quantum cryptography in the automotive context

cs.CR cs.DC ·2026-10-28 ·Mariana do Carmo Barata Gonçalves, Valderi Reis Quietinho Leithardt
Abstract · 24 pages

This study comparatively evaluates post-quantum cryptography (PQC) in automotive communication scenarios, considering the threat posed by quantum algorithms such as Shor's and Grover's algorithms. The simulations use OMNeT++, Veins, INET, and SUMO on two hardware platforms and examine ML-KEM/CRYSTALS-Kyber, ML-DSA/CRYSTALS-Dilithium, and FALCON. ML-KEM and ML-DSA are finalized NIST standards, whereas FALCON is a NIST-selected candidate pending final standardization. The evaluation considers a rural V2V scenario and an urban V2X scenario supported by a roadside unit. Kyber exhibits low {key encapsulation mechanism (KEM)}\color{black} operation latency, Dilithium provides balanced signature performance, and FALCON has high key-generation and signing costs but efficient verification. The results are interpreted as a comparative assessment among the evaluated PQC schemes under the reported simulation conditions; {a conventional ECDSA/ECDH baseline measured in the same simulation setup is also reported for reference.}\color{black} The resulting dataset can support future studies of cryptographic selection in automotive systems.

0.19

Is it my preference? Role Grounding Attack against Agent's Memory via Social Interaction

cs.CR cs.AIcs.MA ·2026-10-28 ·Haoran Bu, Yiting Shen, Litian Zhang, Ruihan Hu, Xi Zhang
Abstract

LLM-based agents now share experiences on social platforms and remember new preferences to enhance their performance. However, existing memory systems contain a vulnerability in role-to-principal attribution that attackers can exploit. We identify Role Grounding Misattribution, in which an attacker-designed interaction causes the victim agent to incorrectly record a preference from the attacker as a preference of the victim's user. Specifically, we are the first to formalize Role Grounding Attack (RGA), an attack model that exploits this vulnerability through nested conversation structures. In the memory formation stage, an experience-sharing post quoting a conversation between the attacker agent and its user is published, which explicitly identifies its publisher. The inner conversation expresses preferences from the user of the attacker agent. When the victim agent remembers preferences from this nested structure, its memory loses the context binding and grounds the attacker's preference to the victim's user instead. In the memory retrieval stage, the misattributed preference persists and influences the victim agent's decisions when it encounters relevant scenarios. We instantiate RGA as tool selection manipulation and privacy leakage. Tool selection manipulation causes a tool-call preference to be misattributed to the victim's user, steering the victim toward an attacker-controlled tool. Privacy leakage extends this by misattributing both a tool-call preference and a task-requirement preference to the victim's user. The victim selects the attacker's tool, then includes sensitive information in tool arguments to satisfy the task requirement. We evaluate RGA across five mainstream memory systems and six language models. Tool manipulation and privacy leakage both achieve high attack success rates. These results call for stronger role grounding.

0.18

A Minimally Informed Black-Box Fake-Report Attack on DRL Traffic-Signal Control

cs.CR ·2026-10-28 ·Amro Saleh, Loukas Lazos, Mohammad J. Abdel-Rahman
Abstract · 7 pages, 3 figures, 5 tables. Submitted to IEEE VTC2027-Spring (under review)

Deep reinforcement learning (DRL) traffic-signal controllers may use vehicle-to-infrastructure (V2I) reports, exposing their perception layer to false-data injection. Existing demonstrations typically assume either white-box access to the controller's weights or full observability of the traffic state. We study the impact of a weaker and more realistic adversary who has no access to the RL model and whose traffic observations are limited to her own approach. This black-box, minimally informed attacker trains an imitation surrogate of the controller. She then crafts bounded fake reports, restricted to inflating vehicle counts on her own approach, that transfer to the real controller. We demonstrate that despite the loss of accuracy relative to an attacker who can observe the entire state and action space, this limited attacker can significantly increase the average waiting time for other approaches while barely changing the waiting time on her own approach, i.e. externalizing the added delay onto the other approaches. Because the only realizable move is to inflate one's own reported queue, we find that a zero-knowledge attacker that simply reports the maximum feasible queue is as damaging as the surrogate-based one, while an unstructured random inflation is not: the surrogate is not required to reproduce the degradation, and the exposed vulnerability is the V2I sensing channel itself. The surrogate's role is directional control, selecting a disruptive or a selfish objective at the same budget. The attack persists under +/-20% uniform demand variation. Finally, on a common per-lane budget we place the attacker on a knowledge ladder against white-box adversaries, and find the black-box surrogate matches a realizable white-box attacker that uses the victim's own gradients; only abandoning realizability raises the damage further.

0.16

CARI: Cross-Boundary Adversarial Runtime Injection in Coding-Agent Workflows

cs.CR cs.SE ·2026-10-28 ·Jie Fu, Shuang Wu, Ashish Kundu, Wendy Hui Wang, Yuan Hong
Abstract

Coding agents increasingly develop, test, and integrate software whose components are owned by different developers, teams, or services. This exposes a distinct attack surface: a malicious or compromised peer development role can manipulate a live component of the software under development (e.g., a backend service, test harness, CI job, or deployment service) and influence the runtime feedback that another role's coding agent must process to complete its task. Although such a component has legitimate diagnostic authority to report its own state, it has no action authority over the victim's local files, credentials, shell, or development environment. We introduce Cross-Boundary Adversarial Runtime Injection (CARI), a direct runtime-feedback injection setting that exploits this authority mismatch. Rather than explicitly compromising the victim's prompt, model, tools, or local environment, the attacker covertly embeds malicious instructions in task-relevant runtime feedback produced by a peer-controlled development component. The victim then interprets that feedback as part of normal debugging or recovery, and may translate the diagnostic guidance into unauthorized victim-side actions. We instantiate ten different attacks based on CARI, spanning resource exhaustion, information exposure, and development-state manipulation. Across four programming languages and five code-capable models, CARI succeeds in 2,845/3,000 (94.8%) trials. Furthermore, CARI transfers effectively to product-level coding agents, achieving 61.3% attack success on Codex and 58.7% on Claude Code. By contrast, prompt-level defenses exhibit a security-utility trade-off: generic warnings are ineffective, whereas explicit policies suppress attacks but substantially over-refuse legitimate actions.

0.15

Zero Trust-Enabled Adaptive Incident Response using Hierarchical Reinforcement Learning

cs.CR ·2026-10-28 ·Jisoo Lee, Bhagath Singh Bondili, Harshith Doppalapudi, Gwynith Godin, Nathaniel D. Bastian +1 more
Abstract

Modern enterprises increasingly operate through complex information systems (IS) enabled by cloud computing, remote and hybrid work practices, and digitally integrated supply chains. These structural shifts have expanded organizational attack surfaces and weakened traditional perimeter-based trust assumptions, giving rise to new and persistent forms of cyber threats. In response, adversaries have shifted toward identity-centric, trust-based intrusion techniques that exploit legitimate access within organizational systems. This evolution has created a pronounced speed asymmetry, whereby attackers can adapt and act at machine speed while defenders remain constrained by reactive or rule-based incident response (IR) processes. Drawing on design science research, we propose ZT-AIR, an adaptive IR artifact that uses zero trust architecture as its design theory and hierarchical reinforcement learning as its kernel theory. ZT-AIR learns context-aware trust decisions, separates strategic prioritization from tactical execution, encodes zero-trust controls as actions, and jointly optimizes security effectiveness and operational continuity. Experimental results across diverse adversarial strategies, network scales, and defense environments show that ZT-AIR achieves a strong balance between security and continuity through frequent, minimally disruptive interventions. Practitioner validation further confirms the design's practical viability. This study contributes to IS research by extending computational design science from threat identification to response execution, theorizing zero trust architecture for adaptive IR, and deriving reusable design principles for adaptive decision support artifacts.

0.14

A Defense-Aware Explainable AI Framework for Feature-Level Vulnerability Analysis in Optimal Power Flow

cs.CR ·2026-10-28 ·S. M. Monzurul Hoque Chowdhury, Qinghua Li, Jianhui Wang
Abstract

Optimal power flow (OPF) is a foundational analysis in power system operation, and inaccuracies in OPF decision-making can result in significant operational and security consequences. With the increasing adoption of machine-learning-based OPF surrogate models for real-time and large-scale applications, power system operation has become more susceptible to cyber threats, including data manipulation and false data injection attacks. Explainable artificial intelligence (XAI), particularly Shapley value-based approaches, has been employed to quantify the influence of input features on OPF outputs. Features exhibiting high influence are commonly interpreted as vulnerable. However, this interpretation neglects the role of attack detection mechanisms in the system, which typically monitor input-to-input relationships to detect anomalous behavior. Consequently, highly influential features may be readily detectable and may not represent the most vulnerable attack surfaces. To address this limitation, we propose a defense-aware explainable AI framework that explicitly incorporates attack detectability alongside operational impact. The framework integrates an OPF surrogate model representing input-to-output dependencies with a vanilla autoencoder representing input-to-input dependencies. Shapley-based explainability is applied to both models to construct a feature-level vulnerability metric. Empirical evaluation on IEEE test systems demonstrates that the proposed framework more accurately identifies security-relevant vulnerabilities than conventional OPF-surrogate-model-only explainability approaches.

0.11

Rosemary: Efficient and Adaptable In-Storage Acceleration for Fully Homomorphic Encryption

cs.CR cs.AR ·2026-10-28 ·Xuan Wang, Tianqi Zhang, Gabrielle De Micheli, Yujin Nam, Keming Fan +3 more
Abstract

Fully Homomorphic Encryption (FHE) enables computation on encrypted data and is promising for privacy-preserving computing. However, FHE deployment remains constrained due to massive computational, memory, and data-movement overheads. Existing acceleration strategies often overlook the storage I/O bottleneck caused by the enormous size of FHE data in large-scale applications, as most prior works focus on small benchmarks that fit in main memory. Our studies show that state-of-the-art FHE systems spend 71\% to 85\% of their execution time moving data between storage and compute units rather than performing useful computation. To address the I/O bottleneck, we present Rosemary, a heterogeneous FHE acceleration system that incorporates a novel in-storage processing (ISP) architecture with compiler-level and algorithmic optimizations. Our key innovations include 1) an adaptable hardware-aware FHE engine inside a NAND flash-based solid-state drive (SSD) for efficient and cost-effective in-storage FHE processing, 2) a Throughput-Data-Aware Program Allocator (TDAP) that partitions FHE workloads into heterogeneous programs, and 3) a Compiler-Level Ciphertext-Aware Optimizer (CLCAO) that exploits ISP-friendly FHE kernels for low-cost ISP execution. Across a diverse set of FHE workloads, including encrypted deep learning inference and cloud database queries, Rosemary achieves geometric-mean speedups of $185.7\times$ over state-of-the-art CPU/GPU implementations and $3.7\times$ over prior FHE accelerators, while improving energy efficiency by up to $3.9\times$ and Energy-Delay-Area-Product (EDAP) by up to $16.5\times$.

0.04

Systems & Networking 20

IMPACT: Constraint-Aware Search for Imperfectly Factorized Accelerator Mappings

cs.AR ·2026-10-28 ·Konstantinos Stavrakakis, Ilgaz Er, Theofilos Spyrou, Dimitrios Stavrakakis, Stephan Wong +2 more
top venue
Abstract · 7 pages, 6 figures. Accepted for publication in ASP-DAC 2027

The efficiency of Deep Neural Network (DNN) accelerators depends on how each tensor operation is mapped onto the hardware. A mapping specifies how the computation is tiled across the memory hierarchy, ordered, and parallelized across the processing elements. These choices form a combinatorially large search space, where efficient mappings are hard to find. Most violate hardware constraints, and most remaining mappings underutilize the hardware. Existing mappers either search only part of this space to keep the search tractable, excluding the most efficient mappings, or sample the full space randomly, so most mappings they evaluate are infeasible. We present IMPACT, a mapper that searches the full space. IMPACT pairs a genetic algorithm with a constraint-aware repair operator that converts each infeasible mapping into a feasible one instead of discarding it. On a flexible accelerator, IMPACT outperforms state-of-the-art mappers in EDP, latency, and energy on every DeepBench workload. On a compute-in-memory (CIM) accelerator, whose constraints leave few mappings feasible, it reduces EDP by 66%, latency by 47%, and energy by 18% over the strongest prior mapper across five networks, and rivals hand-crafted expert mappings. Across flexible and highly constrained accelerators, IMPACT finds efficient mappings automatically. IMPACT will be released as open source.

0.45

Hardwired Distributed Processing Units at the Sensor: Truly Parallel Vision Pre-Processing that Frees the Host CPU

cs.DC cs.AReess.IV ·2026-10-28 ·Raj Mali
Abstract · 7 pages, 6 figures, 7 tables. Companion to arXiv:2610.33107

Conventional vision pipelines route every pixel to a host processor, which must then wait for a complete frame before it can act and devote part of each frame period to routine pre-processing. We examine an alternative organisation in which this work is distributed across hardwired distributed processing units (HDPUs): small, fixed datapaths attached directly to the sensor, each committed to a single always-on task, operating truly in parallel on the same pixel stream and delivering compact results to the host. From one library of 3x3 operators sharing a common two-clock latency, we construct three HDPUs (Canny edges, Harris corners, and a line tracker that issues a steering decision) and run them side by side on a single camera stream in a low-cost Artix-7 FPGA, at a cost of 2,568 LUTs and an estimated 0.318 W. Every stage of every unit is bit-exact with its software model. In a full-system simulation with a datasheet camera model, each decision leaves its HDPU within 11.03 camera lines of the pixel that produced it, 22-67x sooner than the earliest moment at which a frame-based program on the host could respond; the HDPU latency varies by no more than 17 ns, whereas the host's own run times reached 12.5x their median. Offloading the three tasks returns 18-30% of a core to the host in every frame period at 640x480, and 63% of the frame period, with all threads in use, at 1080p. The cost of the units grows with frame width only through their line buffers, and in a closed-loop model of a line-following vehicle the HDPU raises the highest speed at which the line is held from 4.75 to 10.00 m/s.

0.43

MIRAGE: A GPU-Accelerated IQ-Level Channel Emulator for Real-Time Wireless Digital Twins

cs.NI eess.SP ·2026-10-28 ·Ali Saeizadeh, Angelo Feraudo, Avaneeth Anil, Leonardo Bonati, Pedram Johari +1 more
Abstract · 14 pages, 12 figures, 3 tables

Wireless digital twins enable network policies to be evaluated before deployment, but reproducing channel effects within a running protocol stack requires continuous waveform processing under strict timing constraints. We present MIRAGE, a GPU-accelerated channel emulator that applies propagation effects directly to in-phase and quadrature samples exchanged by software 5G New Radio stacks. MIRAGE combines batched multiuser, multiantenna filtering with timestamped shared-memory transport and radio-driver integrations for OpenAirInterface and OCUDU. External propagation models supply channel responses asynchronously, while versioned buffers ensure consistent coefficients within each processing interval. Experiments on one NVIDIA H100 GPU evaluate waveform accuracy, processing time, channel updates, and full-stack traffic delivery. At 61.44 million samples per second, eight single-antenna links, four two-by-two links, and one four-by-four link meet the tested 0.5 ms bidirectional processing criterion. Waveform errors remain within one integer quantization step per component, with no mixed coefficient versions across 16.7 million analyzed slots. Integration with the X5G testbed enables slice resource allocations to be evaluated before deployment. Both evaluated slices ultimately satisfy their throughput caps, requiring one physical reconfiguration under line-of-sight conditions and four under non-line-of-sight conditions. These results demonstrate full-stack policy evaluation and expose prediction errors affecting transfer to a physical network.

0.37

Exact Bistatic Detection With Dual-Use AF Relays in SAGINs

cs.NI ·2026-10-28 ·Changhao He, Xiaojuan Zhang
published
Abstract · This work has been accepted for publication in IEEE Wireless Communications Letters, Copyright transform to IEEE

We study bistatic UAV detection using dual use amplify and forward (AF) relays in satellite aerial ground integrated networks. The target reflects the pilot, payload, and relay noise, allowing the exact finite sample Neyman Pearson detection probability to be derived in noncentral chisquare form under Gaussian payload signaling. The AF cascade introduces a variance signature even without payload transmission and leads to a rate-constrained pilot allocation, distinct relay-power limits, and phase-dependent variance changes under residual direct-path interference. For a decorrelated or suppressed direct path, we further derive a phase-only generalized likelihood ratio test with exact single-integral error probabilities at an assumed target magnitude, the correlated case retains a known-coefficient detection law. A 15-GHz low Earth orbit link budget and Monte Carlo simulations validate the analysis.

0.36

Exploring and optimizing the Vivado Flow on large FPGA designs: A Practical Analysis of Strategies and Their Impact on Timing

cs.AR eess.SPhep-ex ·2026-10-28 ·Kachkachi Nor Eddine
Abstract · This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may…

Achieving timing closure in FPGA designs is crucial, especially for high-performance applications. The Vivado Design Suite offers several strategies for optimization, but finding the best configuration for a given project remains challenging. This paper explores the various Vivado's timing directives and strategies to improve performance across timing, area, and runtime. Through two test-case designs, an NQR spectrometer SoC-FPGA and a digital readout ASIC prototype for high energy physics particle detector's, different strategies and configurations are compared to identify the most efficient flow.

0.34

TableTune: Measured, Self-Correcting Draft-Length Tables for Dynamic Speculative Decoding

cs.ET cs.AI ·2026-10-28 ·Abhishek Malvankar, Paul Castro, Nick Mitchell
Abstract

Speculative decoding speeds up large language model (LLM) serving only while the GPU has spare compute, so the best draft length, the number of tokens a small drafter guesses per step, shrinks as the batch grows. vLLM lets operators set the draft length for each batch size with a hand-written table, but nothing measures or corrects that table. We present TableTune. Its tuner times the decoding step for every batch size and draft length on vLLM's own serving path and computes, in minutes, the table that maximizes throughput. Its controller, a vLLM plugin, tracks how many guesses are accepted in live traffic and rebuilds the table while serving. We measure TableTune only against other ways to speculate. On an H100, the tuned table delivers at least 94% of the throughput of the best fixed draft length at every load in four of five drafter settings (88% in the fifth) without the operator having to know which length that is. Against reimplementations of TurboSpec- and Nightjar-style control on the same engine, its worst-case loss is never higher than theirs, and with the EAGLE-3 drafter it is 3% against 9% and 12%. On a mixture-of-experts model, vLLM's documented table loses up to 18% and the tuned table at most 3%. Starting from mistuned tables, the controller raises throughput by up to 58% (up to 45% on vLLM's default runner), and after a workload switch with the DFlash drafter it recovers 96% of a stale table's loss. In llm-d's optimized-baseline deployment as shipped, two Qwen3-32B replicas on four GPUs, the tuned table is up to 32% faster than a fixed draft length of 3 and up to 25% faster than vLLM's documented table at high load, and the controller raises the documented table's throughput by up to 29%.

0.31

Slant: Shaping Traffic for Resource-Efficient Microservices

cs.DC ·2026-10-28 ·Congwei Zhang, Chieh-Jan Mike Liang, Mengbai Xiao, Yifei Zou, Shulun Wu +3 more
Abstract

Microservice architectures rely on load balancers to distribute requests across service replicas, yet existing designs largely optimize for balanced routing without considering how routing shapes autoscaling signals and resource efficiency. We present Slant, a resource-aware load balancing framework that intentionally creates safe traffic imbalance to improve resource utilization. Slant profiles service-specific critical points, computes conservative multi-dimensional capacity, derives imbalance-aware routing weights that maintain hot and cold pods, and falls back to safer behavior under anomalies. By exposing clearer utilization signals, Slant works with existing vertical pod autoscalers (VPAs) without modification. Experiments on three microservice benchmarks show that Slant reduces allocated CPU cores by up to 26\% while lowering P99 latency by up to 18\% compared to Round-Robin. Under dynamic workloads, Slant triggers VPA scale-up 45\% faster and completes scale-in 40\% faster. Under anomalies, it recovers 40\% faster from pod failures and limits latency amplification to 1.8$\times$ versus 2.6$\times$ during downstream slowdowns.

0.29

MCCL: Training Pytorch models across a fleet of Macs at Skywalker Sound

cs.DC cs.LG ·2026-10-28 ·Kevin Sasso, Chris Fiegel, Francis Aitken, Nicolas Tsingos, Nicholas Tong
code/data
Abstract

Skywalker Sound, a premier audio post-production facility, operates more than 200 M4 Max Mac Studios on a 10Gb Ethernet network. When not in use by artists, these machines sit idle, representing massive untapped unified memory and compute capacity. While these fleets are excellent candidates for distributed machine learning alongside existing NVIDIA GPU clusters, PyTorch lacks an accelerated distributed backend for Apple's Metal Performance Shaders (MPS). To bridge this gap, we introduce the Metal Communication Collective Library (MCCL), the first open-source MPS backend for multi-node PyTorch training over switched Ethernet. We evaluate its performance and scalability across 24 nodes, demonstrating that large all-reduce operations reach 98-99.7% of TCP goodput. Training a 113.5M-parameter Vision Transformer (ViT) scales from 198 to 4,568 samples/s, achieving 95.9% parallel efficiency. Benchmarked against an NVIDIA A100 GPU, one M4 Max Mac equals roughly 1.7x the throughput of FP32 matrix multiplication, and 5.5 Macs match an A100 using TF32. Harnessing 100 idle Macs for 12 hours nightly offsets approximately $200,000 annually in equivalent cloud compute costs. Our results demonstrate that existing Mac infrastructure can be seamlessly repurposed for high performance distributed training without abandoning the standard PyTorch codebase.

0.28

The Endgame of FSDP Communication: Layout-Aware Collectives on Copy Engines, RDMA, and NVSwitch

cs.DC cs.LGcs.PF ·2026-10-28 ·Zejian Wang, Yifan Zhao, Kaihui Ren, Xiaoling Zhou, Yubin Wang +9 more
Abstract · 18 pages, 13 figures, 9 tables

Fully Sharded Data Parallelism (FSDP2) maintains parameters and gradients as per-parameter tensors, whereas its collective path operates on flat, rank-major buffers. Converting between these layouts copies all-gather outputs into parameter storage and packs gradients into full-group fp32 buffers. Layout conversion alone accounts for 6.4% of step time in Llama-3.1 70B training at 8K tokens per GPU. Communication kernels impose a second cost: a default NCCL reduce-scatter can slow a concurrent GEMM that would otherwise complete in a single wave by 69%. We present TIDE, a set of zero-copy collectives built on public NCCL 2.31+ APIs. TIDE preserves FSDP2's per-parameter storage and compute kernels while folding layout placement into collective addressing. TIDE-AG delivers shards directly to parameter storage using NVSwitch multicast and paced recursive-doubling RDMA, with no SMs used for transport or placement. TIDE-RS consumes per-parameter bf16 gradients directly and reduces them into fp32 shards (a mixed-precision mode that NCCL does not provide) using copy engines and short-lived thread blocks (CTAs) for node-local accumulation and NVSwitch for cross-node reduction. A two-phase pipeline separates gradient reads from network progress; Slot-Reuse optionally stages gradients in vacated all-gather slots. On 64 GPUs, TIDE reduces step time by up to 27.1% relative to Stock FSDP2 and 6.0% relative to FSDP2+SymmMem across Llama-3.1 70B, Wan2.2, and Qwen3-30B-A3B. It reduces reduce-scatter SM time by factors of 12.5-14.5 and cuts interference with concurrent compute by up to an order of magnitude. Its reduce-scatter buffer remains nearly constant from 16 to 64 GPUs at a fixed per-GPU output size, and its peak memory is up to 4.7 GiB below FSDP2+SymmMem's and, with Slot-Reuse on Wan2.2, below Stock FSDP2's.

0.27

CoWalk: Cooperative Walks for Agentic Batch Queries on Disk-Resident Graphs

cs.DB cs.IR ·2026-10-28 ·Qiao Liang, Zhiguang Chen, Yue Yu, Nong Xiao
Abstract

Workloads such as query rewriting, deep research and group-relative policy optimization (GRPO) issue batches of semantically related vector queries. Existing disk-resident approximate nearest neighbor (ANN) systems process these queries independently. On three workloads, 31 to 67 percent of the I/Os of a batch fetch a page that another query of the batch has already fetched. A page cache shared by the batch removes these I/Os, yet it makes a single-thread search only $1.1$ to $1.2{\times}$ faster, because the search time goes to distance computation and the cache does not change the walk of any query. We present CoWalk, which runs the queries of a batch as cooperating walks on one thread. Every query keeps its own candidate list, a page is read once per batch, and the full-precision vector on every fetched page is ranked by every query of the batch. A query therefore reaches the target recall with a shorter candidate list, which computes fewer distances and issues fewer I/Os. At recall@10 of 0.95, CoWalk answers a batch on one thread $1.6$ to $1.9{\times}$ faster than the fastest per-query baseline and issues $2.8$ to $6.9{\times}$ fewer I/Os. On a server it serves $3.0$ to $6.9{\times}$ the throughput of that baseline at a similar batch latency.

0.26

Barren Plateaus in Quantum Linear Combinations: Interference and Lie-Algebraic Control

cs.ET quant-phcs.LG ·2026-10-28 ·Amir Koutahi, Pouya Kananian, Hans-Arno Jacobsen
theory
Abstract

A coherent sum of quantum circuits introduces interference that can alter the training signal. We distinguish a single linear combination of unitaries (LCU) from a circuit that repeatedly alternates branch operations and ancilla mixing. For a single-pass LCU, independent branches with zero matrix means cannot erase the differentiated branch's gradient variance: the total variance is at least the branch variance times the fourth power of its coefficient magnitude. This bound holds at any gate position and does not require the circuit unitaries to be Haar-random. Beyond this lower bound, we derive exact loss and gradient formulas that quantify the additional contributions from interference between branches. One of these terms depends on relative coefficient phases and vanishes when the branch ensemble is invariant under multiplication by $i \mathbb{I}$; a four-qubit matchgate example shows that it can be nonzero. For repeated interleaving with full special-unitary ancilla control, we determine the generated Lie algebra from the associative algebra of the branches, including the available central phase directions. This control result predicts concentration only with additional assumptions on circuit sampling. Matchgate examples exhibit polynomial single-pass LCU variances and exponential concentration under joint Haar-random sampling. We also derive coefficient-selection rules and bounds for postselection, measurement error, and approximate moments.

0.26

AnchorKV: Lossless Key-Value Cache Reconstruction from Layer-Sparse Residual Anchors

cs.AR ·2026-10-28 ·Hao Li, Jiashu Zhang, Kaleem Ullah Qasim, Heying Zhang
Abstract · 24 pages, 7 figures, 15 tables

Residual-stream caching can reconstruct transformer key-value (KV) caches losslessly, but its memory advantage depends on architecture. We present AnchorKV, a storage-replay spectrum connecting per-layer residual caching and full recomputation through configurable residual anchors. An architecture-dependent break-even rule identifies the first storage-saving spacing, while deployment selects an operating point from the spectrum and cascade replay reduces reconstruction from O(Ls) to O(L), with L the model depth and s the anchor spacing. For interleaved sliding-window architectures, exact dynamic programming places anchors under non-uniform reconstruction demand. Across CPU, CUDA, and Ascend NPU backends, six architecture families, and fp32/bf16 arithmetic, evaluations verify bit-exact KV reconstruction and unchanged continuation logits and tokens. At 4k context, cascade reconstruction is up to 16.3 times faster than naive replay; across eight applicable global-attention checkpoints at 32k context, deepest spacing yields 5.78-54.0 times full-KV reactivatable session capacity. Compared with seven external cache methods on a common model, AnchorKV preserves full-cache retrieval and question-answering scores, whereas lossy methods change output fidelity and task scores. The break-even spacing is computable from a model's configuration file alone.

0.25

GearUp: DAG-aware Acceleration for Serverless Inference Serving via Slack-guided Pre-loading

cs.DC cs.LGcs.OS ·2026-10-28 ·Qi Qi, Yifan Sui, Hanfei Yu, Hong Zhang, Hao Wang
Abstract

Serverless platforms increasingly host machine learning (ML) inference workflows, modeled as directed acyclic graphs (DAGs) of interdependent functions. Beyond container startup, each ML node must load libraries, model weights, and CUDA kernels before it can run. These costs accumulate along dependent DAG paths and place initialization on the end-to-end critical path. Existing systems optimize one half of this problem. DAG-agnostic pre-loading prepares nodes the request may never reach, wasting host and GPU memory. DAG-aware serving routes and scales whole functions, leaving the loading stages inside a node untouched. This paper presents GearUp, a serverless DAG inference serving system that removes ML-node loading from the end-to-end critical path at low resource cost. GearUp profiles the loading stages of each node and schedules them inside the slack time the DAG exposes, hiding initialization behind upstream execution. It reuses initialized instances within and across requests to avoid redundant cold starts. A probabilistic, traffic-adaptive keep-alive policy sets how long idle instances are retained, balancing memory cost against cold-start risk under bursty load. On the AMBER Alert and FINRA-ML workflows with NVIDIA L40S GPUs, GearUp reduces median end-to-end latency by 64.5% and 75.9% against SMIless, for median speedups of 3.1x and 10.0x over cold execution. Integrated GPU and host memory allocation fall by up to 83.9% and 94.6% against eager pre-loading in InstaInfer. GearUp achieves 4.2x strong scaling on four workers.

0.25

LASSO: An LLM-Agent Synthesized System for SQL Validation

cs.DB cs.SE ·2026-10-28 ·Shreyas Jadhav, S. Sudarshan
Abstract

The rise of Large Language Models (LLMs) has significantly advanced text-to-SQL capabilities, but generated queries may have errors. Test datasets are important for catching errors, but having only one or a few test datasets leaves many bugs undetected. Effective testing requires multiple test datasets that can expose subtle semantic bugs and edge cases. Recent work on test data generation and equivalence checking has focused on using constraint solvers. However, constraint-based approaches are slow and not scalable. In this paper, we introduce LASSO, a framework inspired by XData, but designed to minimize the use of constraint solvers, instead prioritizing use of data generation strategies based on fast heuristics. LASSO deploys a number of "attack strategies", each targeting a specific class of semantic errors. LASSO code was built using LLM-based agentic coding. We used carefully designed workflows and rules to support effective code generation and to avoid common code generation issues. A large number of internal test cases are included in LASSO to catch errors in generated code, and more are added when we extend LASSO, to avoid subsequent regressions. Our empirical evaluation across multiple benchmarks demonstrates that LASSO outperforms the state of the art, including XData, SpotIT, ParSEval, and Polygon.

0.24

The Code Behind the Layout: AI Agents for Analog Layout Generators

cs.AR cs.AI ·2026-10-28 ·Minxing Chu, Anhang Li, Martins Atilla, Ruichen Qi, Mehdi Saligane
Abstract

Analog layout remains largely manual and hard to automate. Generator frameworks such as gLayout express layouts as programs, producing layouts from parameterized, design-rule-aware generators; however, writing such generators still takes expert effort. Large language models (LLMs) could now write generator code, and frontier models can even draw layout geometry from design intent. This raises a practical question: if a model can draw the layout itself, what does a generator framework still contribute, and is applying these models to a generator framework better than having them draw a layout from scratch? Through controlled experiments on gLayout in SKY130, we find that the main obstacle to drawing a layout directly is its time and token cost, and that the main advantage of a model building on a generator framework is reusability. Its verified primitives keep each model edit small enough to check, and its parameters allow a single edit to be evaluated across an entire family of layouts. Accordingly, a model is most effective when it develops the layout from the generator source of the family that the design intent belongs to, and tests each edit on the family members most likely to fail, so that one verified change improves every layout the generator produces. The results of these family tests can then serve as supervision for a local model, which learns edits that hold across the family at a fraction of the inference cost.

0.22

DistOptimize: Automatic Optimization of Distributed Protocols

cs.DC cs.DB ·2026-10-28 ·David Chu McElroy, Shadaj Laddad, Lucky Katahanas, Mingwei Samuel, Octavio Lomeli +3 more
Abstract

Distributed consensus and consistency protocols are difficult to reason about yet foundational to the cloud; as such, they often undergo a rigorous proof and formal verification process before deployment. The same level of rigor is rarely applied to protocol modifications, which occur frequently in production and threaten to compromise safety. In this paper, we introduce DistOptimize, an automatic optimizer for distributed protocols that uses well-known optimization techniques to modify existing protocols without the risk of safety violations. DistOptimize adapts throughput optimizations from academic languages to frameworks used in production, verifying the correctness of the rewrites in Rocq and automating their application through a combination of calibration and ILP-based search space navigation. Applied to both fundamental distributed protocols and protocols used at AWS, DistOptimize improves throughput by up to 4x while increasing latency by at most 0.3ms.

0.21

Efficient Group-Penalty k-means for Text Clustering with Noisy LLM-Generated Constraints

cs.DB ·2026-10-28 ·Feiteng Nie, Yuan Sun, Shengkun Zhu, Sheng Wang, Zhiyong Peng
theory
Abstract

k-means provides a simple and efficient foundation for text clustering, but embedding alone can miss task-specific semantic boundaries. Must-link (ML) and cannot-link (CL) constraints can correct such errors, yet conventional constrained clustering assumes that these relations are already supplied. LLMs can generate them without task-specific labels, but exhaustive querying is costly and their judgments can be incorrect. We present GPKmeans, an efficient Group-Penalty k-means method for text clustering with noisy LLM-generated constraints. GPKmeans uses LLM as an offline local pairwise annotator, querying selected anchor-candidate pairs for sparse ML links and constructing set-valued CL evidence without all-pairs queries. It converts connected components of the ML-link graph and the generated CL sets into groups with normalized, bounded soft penalties, limiting each group's influence independently of its size. Cached coordinate descent is proposed to efficiently solve the constraints clustering problem. We prove finite-step convergence and derive deterministic and probabilistic oracle-referenced bounds that relate objective distortion to the penalty coefficients and to the group-level error probabilities of the generated constraints. Across nine datasets, GPKmeans uses fewer LLM calls and yields more accurate constraints. It achieves the best average ACC, NMI, and ARI while running approximately 30 times faster than state-of-the-art constrained clustering methods.

0.20

FlatCake: A Flat KVCache Storage Architecture for LLM Serving

cs.DC cs.OS ·2026-10-28 ·Xin Huang, Teng Ma, Zhuoran Li, Yibo Zhao, Zhaoshi Li +5 more
Abstract · 10 figures

KV cache reuse accelerates LLM serving, but existing systems manage host DRAM and NVMe SSD as a tiered hierarchy where DRAM acts as an intermediate cache. Production LLM workloads break both premises of this hierarchy. First, bimodal reuse defeats DRAM temporal filtering, since workloads are dominated either by one-shot blocks with long reuse distances or by multi-gigabyte agentic sessions that exceed DRAM capacity, putting SSD on the critical read path. Second, DRAM staging and eviction-ordered layout bottleneck SSD restoration. Holding host staging buffers for in-flight prefixes causes contention under load that aborts prefetches, while incremental eviction scatters a session's blocks across the SSD and prevents large sequential reads. We present FlatCake, a flat KV storage architecture that exposes host DRAM and NVMe SSD as peer media in a unified 64-bit address space. Exploiting the fact that KV blocks are write-once, immutable, and reused at session scope, FlatCake places each block directly onto its target medium in session-contiguous, layer-major order at creation time and never migrates it. FlatCake streams coalesced SSD reads directly into a bounded GPU window via GPUDirect Storage and decouples DRAM and SSD request queues to prevent head-of-line blocking. On a production Claude Code trace with DeepSeek-V4-Flash, FlatCake sustains a 94% cache hit rate as load increases (versus 64% for Mooncake), eliminates Mooncake's $6.5\times$--$6.7\times$ SSD read amplification, and reduces peak-load mean TTFT by $1.6\times$ while improving total token throughput by $1.6\times$. Across 8K--128K contexts on Llama-3.1-70B, FlatCake further reduces mean TTFT by up to $2.4\times$ over Mooncake and $1.8\times$ over Mooncake-GDS.

0.13

GI-RTL-Fuzz: Grammar-Informed RTL Generation with LLMs for End-to-End Differential Simulator Fuzzing

cs.AR cs.CR ·2026-10-28 ·Ziyue Zheng, Zhi Qu, Yangdi Lyu
Abstract

Register-Transfer Level (RTL) simulators are fundamental to modern IC design workflows, yet like any complex software, they frequently contain critical bugs that compromise simulation accuracy. To systematically stress-test these tools, we introduce GI-RTL-Fuzz, a grammar-informed fuzzing framework that leverages Large Language Models (LLMs) to generate high-quality RTL test cases. Our key innovation is guiding the LLM's generation process using Backus-Naur Form (BNF) grammar coverage information, enabling the exploration of complex and rarely-tested language features. By performing differential testing against different RTL simulators, our framework uncovered a significant number of previously unknown inconsistencies in both the linting process and simulation results of major Verilog simulators.

0.13

Tight Bounds for Anonymous Parity Certification

cs.DC ·2026-10-28 ·Pedro Montealegre, Jorge Valenzuela
theory
Abstract · 23 pages, 5 figures

Local certification is the task of verifying graph properties using certificates assigned to vertices and checked using only local information. A central question is which properties admit constant-size certificates, independently of network size. A fundamental test case is the parity of the network order. Bousquet, Feuilloley, Valenzuela, and Zeitoun (DISC 2026) showed that the complexity of certifying even order depends dramatically on the verification model. On arbitrary graphs, three bits per vertex suffice with unique identifiers and verification radius two. In the anonymous radius-one model, they established an $Ω(\log\log^* n)$ lower bound and an $O(\log n)$ upper bound. They also showed that constant-size certificates suffice in this model on bounded-expansion classes. In this paper, we settle the complexity of anonymous parity certification on general graphs and further investigate the role of graph sparsity. Our main contributions are: * We close the gap left by Bousquet et al., proving that the local complexity of certifying even order in the anonymous radius-one model is $Θ(\log\log n)$ bits. * We show that on graphs of degeneracy at most $d$, parity can be certified with $O(\log d)$ bits, while $Ω(\log\log d)$ bits are necessary. In particular, every class of bounded degeneracy admits constant-size certificates, strictly generalizing the result for bounded-expansion classes. * We show that constant-size certification cannot be guaranteed on nowhere-dense classes, a broader notion of sparsity that includes bounded expansion. We construct a fixed, subgraph-closed nowhere-dense class on which parity certification requires $Θ(\log\log n)$ bits. Together, our results provide a more complete picture of anonymous parity certification on sparse graph classes, showing that its complexity depends on the particular structural notion of sparsity.

0.13

Software & Programming 20

Template-Free Dynamic Invariant Detection

cs.SE ·2026-10-28 ·Erfan Arvan, Michael D. Ernst, Martin Kellogg
code/data
Abstract · 20 pages

A dynamic invariant detector proposes invariants, such as method pre- and postconditions, and rules out those that do not hold at run time. Traditional detectors propose invariants from a fixed library of templates, which limits their expressiveness. A large language model (LLM) can propose invariants instead of, or in addition to, template-derived ones. With the right prompt and context choices, we found that this idea can both improve the invariants' expressiveness and reduce the cost of inferring the invariants. Greater expressiveness brings complications: an LLM-proposed invariant can execute arbitrary code during invariant evaluation, causing side effects or infinite loops. Our novel instrumentation strategies mitigate these problems. We wrap each candidate and evaluate only one invariant per thread at a time, so nested invariant evaluations cannot trigger one another indefinitely. To recover from non-nested hangs, we monitor each run for stalls and retry it under a growing time budget. Because we cannot prove a candidate side-effect-free in advance, we use delta debugging over the invariants evaluated during a failing test to isolate and disable a side-effecting candidate without discarding the rest of the run. We compared our implementation, Oca, to the popular template-based Daikon invariant detector on real open-source Java projects. Oca produces more expressive invariants: on average, 28% of its held invariants call a project-specific method, which Daikon's templates cannot express. Its invariants are more often valid, with a false-positive rate of 18% in a manual inspection, vs. 76% for Daikon, and more useful: as properties for property-based testing, they exposed 9 of 25 real bugs, compared with 4 for Daikon. Oca also costs less, using 566 minutes and 27 GB of disk, compared with 2,817 minutes and 827 GB for Daikon.

0.40

Write Once, Instrument Anywhere: Portable, Sandboxed Tooling with WebAssembly Components

cs.SE cs.PL ·2026-10-28 ·Elizabeth Gilbert, Chris Fallin, Heather Miller, Ben L. Titzer
Abstract

Monitoring and debugging deployed software is crucial for scalability, performance, and robustness. Cross-cutting concerns are pervasive across software ecosystems, yet engineers must address them with a fragmented toolbox. While high-level frameworks exist to offer application-level types and semantics, they often lock a solution into specific languages, runtimes, or hardware architectures. Portability often forces engineers to choose lower-level instrumentation mechanisms. For example, in-language frameworks can trace a request across processes, but multi-language contexts require instrumentation at the boundary, for example using eBPF. While these frameworks provide powerful capabilities, they often result in 1) a tradeoff between application-level types or low-level bytes and packets, 2) non-composable domain-specific tools, 3) feature fragmentation across frameworks, and 4) tools that often run with unrestricted access to the program and its environment. We address these problems with Splicer, a WebAssembly component interposition framework that 1) provides semantically rich application-level types, 2) enables pluggable tooling with powerful capabilities, 3) reduces ecosystem fragmentation through porting across languages and hardware, and 4) confines each tool to a sandbox with explicitly granted permissions. We evaluate Splicer on a suite of benchmarks demonstrating the pluggability of non-trivial tooling written in multiple source languages and their performance.

0.40

DIFUZZI: A Differential Fuzzing Tool for Code Functional Similarity Assessment

cs.SE ·2026-10-28 ·Simantika Bhattacharjee Dristi, Matthew B. Dwyer
code/data
Abstract

A range of software engineering tasks, such as optimization and refactoring, seek to preserve the behavior of software components and require methods to assess functional similarity. Research has explored formal approaches to prove functional equivalence and used machine learning methods to approximate it, but the former does not scale and does not quantify the extent of similarity and the latter struggles when subtle semantic differences are present. Differential fuzzing offers a scalable approach for differencing programs semantically, but applying it often requires substantial manual effort to construct input generators that are well-suited to a target pair of programs. To address this challenge, we introduce DIFUZZI, which incorporates an LLM-based generator synthesis step that is iteratively validated and repaired. By automating this step, DIFUZZI paves the way for broader applicability of differential fuzzing. We release DIFUZZI at https://github.com/Dristi123/difuzzi_tool, with a demonstration video at https://youtu.be/FlHNOHuo9WE

0.39

When LLM Architectures Meet Smart Contract Vulnerability Detection: a Replication Study

cs.SE ·2026-10-28 ·Gerardo Iuliano, Gabriele De Vito, Dario Di Nucci
top venue
Abstract

Smart contracts are immutable programs deployed on blockchain platforms, where vulnerabilities can translate directly into financial loss. Among the techniques proposed for their detection, recent work has relied increasingly on Large Language Models (LLMs) guided by carefully crafted prompts. However, the evidence accumulated so far rests almost entirely on the dense Transformer architecture behind the GPT line and on model snapshots already scheduled for deprecation, leaving open the question of whether the reported behaviors reflect a general property of LLMs or an artifact of that specific design. This paper presents an extended replication of the empirical study by Chen et al. on LLM-based smart contract vulnerability detection. We replicate the original protocol on GPT-4 as a baseline, followed by a differentiated replication, executing the protocol on representative models based on Mixture-of-Experts and RNN-like designs, together with an attempted preliminary evaluation of State Space Models. The exact replication confirms the findings of Chen et al., including the high-recall, low-precision detection pattern they report on GPT-4. Nevertheless, the differentiated replication shows that this pattern does not transfer uniformly across the evaluated models. Those from the same architectural family may also exhibit different behaviors, indicating that architectural family alone is insufficient to predict the resulting detection profile. We observe the same qualitative dense-Transformer profile in both GPT-4 and Gemma-3, despite a gap of roughly two orders of magnitude in parameter counts, whereas two Mixture-of-Experts models from different lineages exhibit qualitatively divergent behaviors, indicating that architectural family alone is insufficient to predict the detection profile.

0.38

Precedent or Intent? What Coding Agents Follow When Evolving Existing Code

cs.SE cs.AI ·2026-10-28 ·Haichao Zhu, Geyang Xu, Zhaorui Yang, Jiyuan Wang, Qian Zhang
code/data
Abstract · 16 pages, 7 figures, 6 tables

Existing code reveals past implementation decisions but rarely explains the reasoning behind them. When coding agents implement new features or modify existing functionality, past decisions may conflict with new task instructions. This raises a fundamental question for agentic software evolution: when multiple implementations satisfy the same functional requirements, how do coding agents decide whether to follow existing code or new task instructions? We investigate this question through bidirectional, controlled experiments in purpose-built repositories, followed by validation in real-world open-source repositories. Our approach isolates three competing sources of guidance: (1) agents' default preferences, (2) repository precedents, and (3) explicit task instructions. We conduct 5,625 trials across four coding agents and six implementation pairs, with validation in four open-source repositories spanning three programming languages. We find that agents share the same default implementation choices, yet overwhelmingly follow repository precedent when it conflicts with those defaults (359/360 trials). Crucially, not all precedents carry equal influence: three supporting local files yield 96% adoption, compared with only 4% for sibling-resource files, and changing file placement alone can shift adoption from 0% to 100% at a fixed example count. When new task intent conflicts with existing code, written priority rules achieve substantially higher success than code examples alone (59/60 versus 38/60 joint successes for two agents on two implementation pairs). The effectiveness of explicit rules also extends to open-source repositories, where a harness-loaded rule raises joint success from 0/120 to 117/120. These findings motivate design-decision traceability that preserves not only implementation choices but also the rationale and priorities that should guide future changes.

0.36

Order Maintenance in Worst-Case Constant Time, Verified

cs.LO cs.DS ·2026-10-28 ·Denis Firsov
theory
Abstract

Order maintenance asks for a data structure that keeps a sequence of elements under insertions and deletions and tells which of two elements comes first. We give an algorithm, assembled from published components, that performs each operation in worst-case constant time, and verify it in Lean 4. The algorithm is a program for a random-access machine deeply embedded in Lean, whose interpreter counts every instruction. We prove that the program answers every query as a simple reference specification does, that each operation takes at most a fixed number of steps, independent of the number of elements and of the history, and that after $t$ operations no value it has computed has more than $O(\log t)$ bits, so it runs unchanged on a machine with words of that size. A verified translation that preserves step counts exactly connects the high-level proofs to the machine program, so the interpreter is the only cost model to trust. An unverified native implementation shows the algorithm's practical cost.

0.36

Mentorship, Interrupted: How GenAI Disrupts the GSoC Pipeline and Why Mentors Matter More

cs.SE ·2026-10-28 ·Ruizhe Xing, Zixuan Feng, Mary Radomile, Stephanie Taylor, Sophia Vargas +1 more
code/data
Abstract

Generative AI (GenAI) makes open source software (OSS) contributions cheaper to produce, and it makes them say less about the people who produced them. In OSS mentoring, that matters: mentors judge what newcomers understand from their work, and newcomers learn from mentors' feedback. We studied how GenAI is changing mentoring in Google Summer of Code (GSoC), a program in which today's mentees can become tomorrow's mentors. We screened every thread on GSoC's restricted mentors' mailing list from November 2022 through March 2026 and interviewed 15 mentors and organization administrators. GenAI helped with onboarding, coordination, and first-pass review, but proposals, entry contributions, and mentees' answers became weaker evidence of what newcomers could do. Mentors made up the difference by checking, and that checking drew on the same time and goodwill the program needs to renew itself; one organization had withdrawn from GSoC. Responses ranged from banning GenAI to accepting it, and the boundary an organization drew determined who did the checking. We propose AIM, a decision framework that links acceptable GenAI use, integration into the mentoring pipeline, and mentoring capacity, to help organizations choose responses they can sustain.

0.31

Wreath products of distributive quasi relation algebra frames and their application to representability

cs.LO ·2026-10-28 ·Wilmari Morton, Claudette Robinson
Abstract · 36 pages, 14 figures

We adapt a construction by Comer (1983) for polygroups to define the wreath product of the dual frames of two distributive quasi relation algebras satisfying certain conditions and show that it is itself a frame. Frames play a role analogous to that of atom structures for relation algebras. After introducing the notion of a frame network, we establish a correspondence between frame networks satisfying certain conditions and representability of finite distributive quasi relation algebras. Next we describe a frame network for the wreath product of two finite representable distributive quasi relation algebras that satisfies the aforementioned conditions, as well as additional conditions, and use this to conclude that the complex algebra of the wreath product is a representable distributive quasi relation algebra. Thus, our construction provides a method for obtaining representable DqRAs from certain finite DqRAs that are already known to be representable.

0.29

survivalGPU: Scalable Survival Analysis for R and Python

cs.MS ·2026-10-28 ·Antoine Poirot-Bourdain, Alexis Van Straaten, Emmanuel Bacry, Anne-Sophie Jannot, Jean Feydy
benchmark
Abstract

This paper introduces survivalGPU, a high-performance Python and R package designed to accelerate survival analysis computations using graphics processing units (GPUs). While the Cox proportional hazards model remains a cornerstone of survival analysis, it can impose substantial computational burdens when applied to large longitudinal data sets. To address this limitation, survivalGPU implements Cox proportional hazards models and weighted cumulative exposure (WCE) models in Python, using PyTorch and KeOps for efficient back-end computation, with interfaces in both Python and R. Numerical validation shows that survivalGPU reproduces the results of reference implementations within the tolerances expected for the selected floating-point precision, with relative errors below 10 -2 in float32 and below 10 -5 in float64. Benchmark experiments demonstrate substantial speedups. For Cox models, coxphGPU() was 5.6x faster than coxph() from the survival package on 1,000,000 patients, and 14.0x faster based on runtimes extrapolated to 1,000 bootstrap replicates on 100,000 patients. Gains were larger for WCE models: wceGPU() was 920.7x faster than WCE() from the WCE package on 100,000 patients, and 1,321.4x faster based on runtimes extrapolated to 1,000 bootstrap replicates. These results show that survivalGPU makes flexible time-dependent survival modeling and bootstrap-based estimation of confidence intervals feasible for large-scale pharmacoepidemiological applications.

0.26

Security Impact of Coding Agent's Memory Files

cs.SE cs.CR ·2026-10-28 ·Thiago Santos de Moura, Selma Güler, Yannic Noller
Abstract

Coding agents automatically load project memory files such as CLAUDE.md at the start of every session and treat them as trusted project guidance. Anyone who can change such a file, for example through a pull request, can therefore steer the agent without ever touching the developer's prompt. We study memory files as an attack surface for insecure code generation and ask (1) whether a malicious memory file makes agents produce code that passes its tests but is vulnerable, and (2) whether agents disclose to the developer that the file shaped their code. For each of 12 CWE classes, we crafted two malicious memory files, a qualified variant that presents removing defensive checks as engineering best practice and a forceful variant that demands it. We ran Claude Code (Sonnet 5.5) and OpenCode (Qwen 3.8 27B, DeepSeek V4.1 Flash) on 36 SecRepoBench tasks with and without these files. The malicious files changed the rate of secure completions by -8 to +6 percentage points, with no consistent pattern across agents or variants. Agents differ sharply in what they tell the developer about the file. Claude Code names it in about two thirds of its final messages, whereas OpenCode mentions it almost exclusively in its reasoning trace, not in the final message the developer reads. Flagging the file as an attack does not reliably protect the code, as such agents still generated vulnerable code. We outline the larger studies needed to quantify this risk across agents and to design defenses.

0.25

Connecting Base-Extension Semantics, Beth Semantics, and Ecumenical Logic

cs.LO math.LO ·2026-10-28 ·Victor Barroso-Nascimento, Joaquim T. Waddington, Elaine Pimentel
Abstract

We investigate the relationship between base-extension semantics, Beth semantics, and Prawitz's ecumenical logic. We first present a base-extension semantics for the propositional fragment of Prawitz's system and prove that it is sound and complete. The semantics interprets intuitionistic and classical operators directly within a common proof-theoretic framework, without reducing the classical operators to their intuitionistic counterparts by means of a translation. We then analyse the effect of incorporating Beth-style clauses into base-extension semantics. When arbitrary bases are admitted and $\bot$ is governed by atomic explosion, these clauses trivialise. By contrast, when $\bot$ is treated as a basic sentence and attention is restricted to consistent bases, the Beth-style clauses for atoms and disjunction characterise, respectively, classical atoms and classical disjunction in Prawitz's system. We also obtain equivalent, path-independent formulations of the classical clauses for atoms, disjunction, and implication. These results reveal that the consistency constraint is not merely a technical restriction: within base-extension semantics, it transforms clauses originally associated with intuitionistic semantics into semantic characterisations of classical operators.

0.24

cuPDQPx: A GPU-Accelerated Restarted Halpern PDHG for Large-Scale Convex Quadratic Programming

cs.MS math.OC ·2026-10-28 ·Haihao Lu, Zedong Peng, Jinwen Yang
Abstract

We present cuPDQPx, a GPU-accelerated, factorization-free first-order solver for large-scale convex quadratic programming (QP). Building on the restarted reflected Halpern primal-dual framework of cuPDLPx, we introduce an auxiliary-variable splitting that yields explicit updates without linearizing the quadratic objective or solving inner quadratic subproblems. A preconditioned proximal point interpretation yields both sublinear and linear fixed-point residual convergence guarantees. For practical efficiency, cuPDQPx integrates diagonal preconditioning, step-size selection, adaptive restart, and adaptive primal and quadratic weight updates. cuPDQPx is implemented in C and CUDA with optimized sparse matrix-vector products, kernel fusion, and CUDA Graphs. Numerical experiments on the Maros-Mészáros collection, a convex subset of QPLIB, and synthetic large-scale QPs demonstrate that cuPDQPx has strong numerical performance at both $10^{-4}$ and $10^{-8}$ tolerances.

0.21

Detecting Incorrect LLM Explanations for Automated Program Repair

cs.SE ·2026-10-28 ·Lam Nguyen Tung, Chenxi Gu, Xiaoning Du, Haoye Tian, Aldeida Aleti +1 more
Abstract

LLM-generated explanations help developers understand automated program repair (APR) rationale and assess patch quality. Incorrect explanations can mislead developers about a repair's reliability, making their detection important. However, manually inspecting the growing volume of generated repairs is impractical, motivating automated detection. Existing methods mainly target incorrect answers from general question-answering LLMs, often by sampling multiple responses and detecting contradictions. Applied to APR, they face two limitations: (1) they may depend on external verifiers that are unavailable or lack APR-specific reasoning; and (2) sampling strategies such as paraphrasing or re-prompting the same input often yield superficial, redundant responses that repeat the original error rather than expose it. We propose MetaReJ, the first method designed to detect incorrect APR explanations using metamorphic testing. Inspired by debate and viva examinations, MetaReJ challenges explanations through an argumentative strategy without requiring external verifiers. It applies novel metamorphic relations (MRs) to recursively question the LLM under equivalent and non-equivalent repair scenarios and detect contradictions. To obtain diverse samples, MetaReJ transforms the buggy program into variants that preserve or alter the underlying bug, then asks the LLM how to repair each variant. Contradictions among responses are encoded as MR violations, which indicate that the original explanation is incorrect. We evaluate MetaReJ on a manually labelled dataset of 1,756 explanations from multiple LLM-based APR systems. MetaReJ achieves an MCC of up to 0.43, outperforming prior methods by up to 333.3% in MCC.

0.19

Extending Higher-Order Critical Pairs

cs.LO ·2026-10-28 ·Johannes Niederhauser, Aart Middeldorp
Abstract

Confluence is a fundamental property of computation which implies that the nondeterminism of a system does not affect its outcomes. A key concept of confluence research in rewriting theory is the notion of critical pairs which captures the potential local sources of non-confluence. In this paper, we lift the definition of higher-order critical pairs to a more expressive class of higher-order rewrite systems. Thereby, we facilitate confluence analysis of rewrite systems which can express complex one-step computations as well as program transformations from functional programming.

0.19

Beyond Passing Tests: A Systematic Review and Controlled Re-evaluation of Techniques for Improving the Quality of LLM-Generated Code

cs.SE cs.CLcs.LG ·2026-10-28 ·Prashank Kadam
surveybenchmark
Abstract · 55 pages, 5 figures, plus 120 pages of supplementary material

Large language models pass most functional-correctness benchmarks, but passing tests says little about whether their code is readable, efficient or secure. Many techniques claim to improve these qualities, yet their evidence is scattered, usually covers one quality dimension and is rarely compared at an equal compute budget. We contribute a systematic review and a controlled re-evaluation. A systematic literature review of 628 primary studies (July 2021 to October 2026), with disclosed, author-reviewed LLM assistance, organizes the techniques by stage, signal source and quality target. Most studies measure only the dimension they target, gains often come with unchanged or unreported correctness, single automated quality signals are gamed, and compute, variance and contamination are rarely reported. A controlled re-evaluation runs ten techniques on four models at matched budgets on correctness, maintainability, efficiency and security benchmarks, extends it to thirteen current models and repository-level tasks, and compares 66 highlighted techniques head-to-head on Gemma 4 31B. At one call, quality prompting raises analyzer issue counts and leaves complexity and judge scores almost unchanged. At matched budgets, best-of-n sampling with an analyzer-weighted ranker matches or beats prompting, planning, self-refinement, a reviewer agent and retrieval on measures the ranker does not optimize, while an analyzer feedback loop and fine-tuning beat it only on analyzer issue counts. Large correctness losses on HumanEval+, a likely contaminated benchmark near its ceiling, do not appear on contests held after the models' training cutoffs. In the head-to-head, 1 of 66 techniques beats best-of-n on its primary target beyond the 95% interval, a rate consistent with chance. Gains should be reported against a budget-matched sampling baseline and on measures independent of the technique's own signal.

0.19

Skeleton Coding: Program Structure as a Specification for Agents

cs.SE cs.AIcs.LG ·2026-10-28 ·Aditya Golatkar, Alessandro Achille, Renos Zabounidis, Vlad Sobal, Yangruibo Ding +7 more
position
Abstract

Coding agents are typically provided end-to-end behavioral specifications that the code they produce should satisfy, such as tests and input--output examples. How many such specifications must a developer provide, and which ones, for the agent to produce the developer's intended program? We argue that this cannot be determined a priori, for it depends on the structure of the implementation, which is left to the agent to decide. An agent first commits to a structure---a decomposition into data types, modules, and functions---which fixes a class of candidate programs; behavioral specifications then select a program within that class. A set of tests that suffices for one structure may leave another badly underdetermined, requiring different edge cases or exponentially more examples. Since the structure is chosen by the agent, developers cannot ensure their specification is complete without reverse-engineering it from thousands of lines of generated code. To resolve this, we introduce Skeleton Coding, in which developers jointly specify the structure---in high-level, code-like format---alongside behavioral specifications. Theoretically, we model AI-assisted coding as a control problem, and show that skeletons can reduce exponentially the number of tests developers must supply to steer an agent toward an acceptable program. Empirically, across four models and three benchmarks, the best-performing skeleton specifications reduce mean test error by 48.4% relative to the best end-to-end behavioral specifications. Even when the agent can query a complete reference implementation, adding a skeleton reduces mean test error relatively by up to 27.0%. Skeletons also narrow the average gap between small and large models by about 52.4% while lowering implementation cost by 15--26x, and reduce code erosion by 24.7% on average.

0.18

Repairs for Whom? Evaluator-Executor Mismatch in Selecting Repairs for Failed LLM Agents

cs.SE cs.CL ·2026-10-28 ·Yijun Liao, Fanwei Liang
Abstract · 17 pages, 4 figures, 11 tables

When an LLM agent fails, repair methods choose which action to replace and what to put in its place. They often score candidate repairs with greedy replay, a stronger model, an environment oracle, or a judge that reads the failed log. We ask whether these evaluators pick repairs that help the agent that will actually execute them. Holding failed trajectories and candidate single-step repairs fixed, we label every candidate with independent rollouts of the deployed policy and measure how much deployment value each evaluator's choice loses. On failures not screened for repair value, pre-specified tests show that greedy replay loses 0.08-0.13 in expected net success gain in ALFWorld and 0.06-0.13 in net reward in WebShop, relative to selection with the deployed agent's own rollouts. At an equal rollout budget, selecting repairs for Qwen2.5-7B with Qwen2.5-14B rollouts loses 0.25 in ALFWorld and 0.07 in WebShop, twenty 14B rollouts per candidate are worth no more than two of the 7B agent's own, and fresh rollouts confirm that the two models prefer opposite repairs on selected disagreement cases. Between similar-sized models from different families, the losses appear in both directions in both environments (0.07-0.17). The scoring executor also changes repair-location labels and method rankings. Repairs should be verified with the executor that will run them, and repair labels should name their executor.

0.18

Beyond atol and rtol: Sound Test Oracles for Low-Precision GPU Kernels

cs.MS cs.PLcs.SE ·2026-10-28 ·Xinyi Shi
Abstract · 37 pages, 5 figures

Low-precision GPU kernels are commonly tested against a reference using hand-picked absolute and relative tolerances. Legitimate numerical error varies with the input, format and precision plan, so a fixed tolerance trades false alarms for missed faults. We present numoracle, which derives per-element acceptance regions from precision-annotated reference programs. Its calculus tracks shared rounding errors, provenance-ordered products, block-scaled quantizers and calibrated math-library errors. A rounding-interval check detects faulty stores that symmetric bounds can admit. Soundness is conditional on explicit rounding, hardware, domain and implementation assumptions; a fixed-input mode removes the zero-mean rounding assumption. We evaluate 13 Triton operators from Liger-Kernel, Unsloth and vLLM on two GPU generations. The main campaigns produce no false alarm on 33,960 correct-kernel and behavior-preserving-variant runs per GPU. Under input shift, the GPU campaigns record no false alarm while tuned tolerances flag up to 30% of correct runs; final-version reference checks identify inputs without complete finite-range certificates. On the RTX PRO 6000, numoracle kills 116 of 124 effective mutants, versus 59-83 for four fixed baselines. Among 400 tuned tolerance configurations, requiring at most 1% false alarms in every evaluated Liger input group permits at most 31 of 53 main-mutant kills, versus numoracle's 50. Historical replays expose accuracy faults in eight change pairs without rejecting their matched corrected controls; a ninth change concerns rounding parity.

0.14

When Relevant Code Is Wrong: Runtime-Aware Context for AI Coding Agents

cs.SE ·2026-10-28 ·Wesley K. G. Assunção, Ramesh Periyathambi
Abstract

AI coding agents increasingly rely on retrieved code and development history to guide software changes. However, highly relevant artifacts may describe implementations that are inactive or incompatible with the target deployment. Neither semantic similarity nor recency establishes whether retrieved information remains applicable under feature flags, architectural migrations, and deployment-specific configurations. We challenge relevance-oriented retrieval by introducing operational context validity, a task-, deployment-, and time-dependent criterion for determining whether software evidence can legitimately guide implementation decisions. We instantiate this new retrieval principle in CASE, an agent-agnostic framework that connects repository artifacts with runtime configurations and architectural evolution, classifies evidence as valid, invalid, or unknown, and generates context packages distinguishing actionable guidance from historical rationale and unresolved constraints. We investigate its feasibility through a reproducible preliminary evaluation involving six development tasks, 15 artifacts, and three deployments. Conventional retrieval exposes 16 incompatible artifact occurrences as actionable, whereas CASE excludes them while retaining all applicable artifacts under the prototype's reference rules. Recency filtering reduces valid recall to 0.43, while removing feature-flag evidence from CASE reduces it to 0.13. Although these results demonstrate feasibility rather than improvements in coding-agent correctness, they reveal limitations of relevance-based retrieval and motivate a new research direction: enabling coding agents to reason not only about which information is relevant, but also whether it remains operationally valid.

0.09

The Reality Gap in Automated Driving Systems Testing: Landscape, Challenges, and Research Directions

cs.SE cs.RO ·2026-10-28 ·Stefano Carlo Lambertenghi, Oliver Weissl, Johannes Betz, Andrea Stocco
Abstract

Simulation is essential for testing automated driving systems (ADS), but discrepancies between simulation and reality can undermine testing conclusions. It remains unclear which discrepancies the literature addresses and what evidence supports correspondence claims. We present a systematic mapping of 235 primary studies that measure or mitigate the reality gap in ADS testing, selected from 4,195 publications. We characterise objectives, targets, approaches, experimental setups, and evaluation evidence. Our classification distinguishes simulated and real-world execution across six ADS test components, separately for each approach and its evaluation reference. This distinguishes real-world information used to construct an approach from evidence used to evaluate its claims. Of the included studies, 76% address mitigation, primarily targeting scenario and actor realism and perception, while 94% report validation or measurement procedures. Only 61 studies report metrics from multiple families covering sensor/perception correspondence, ADS behaviour, vehicle dynamics, or safety outcomes; this coverage does not establish causal links between discrepancies and test verdicts. The synthesis highlights that real-world content and physical execution must be interpreted against the claim evaluated. The findings motivate claim-specific evaluation and component-level reporting to establish which discrepancies affect testing conclusions and when simulation evidence supports conclusions about real-world ADS behaviour.

0.08

Theory & Algorithms 45

Local Important Separators with Applications to Unbreakable Decomposition and Multicut

cs.DS ·2026-10-28 ·Aditya Anand, Euiwoong Lee, Yaowei Long, Thatchaphol Saranurak
top venue
Abstract · To appear in SODA 2027

We present the first local FPT algorithm for finding important separators: Given an undirected graph $G$ with $n$ vertices and $m$ edges, vertices $s,t$ and parameters $k$ and $ν$, we can enumerate all $s$-$t$ important separators of size at most $k$ and volume at most $ν$ in time $\tilde{O}(ν2^{O(k)})$. Known approaches spend either $Ω(m)$ time (not local) or $ν^{Ω(k)}$ time (not FPT). Using this local algorithm, we show the following applications. (1) The first is an improved construction of unbreakable decomposition, a powerful generalization of tree decompositions. For any $k$, we can compute a $(5k,k)$-unbreakable decomposition with adhesion $4k$ in $\tilde{O}(k^{O(k)}m)$ time. This strictly subsumes the result of Anand, Lee, Li, Long, and Saranurak (SODA 2025) and is faster than the recent $O(k^{O(k^2)}m)$-time algorithm by Korhonen (STOC 2025) when $k = ω\Bigl(\sqrt{\frac{\log \log n}{\log \log \log n}}\Bigr)$. Our algorithm is also simpler than that of Korhonen (STOC 2025). (2) We give a near-linear time algorithm for single-sink important separators. For any sink $t$, we can list all $v$-$t$ important separators of size at most $k$ across all $v \in V(G)$ in time $\tilde{O}(2^{O(k)}m)$. Known approaches trivially compute $v$-$t$ important separators for each $v$, taking $Ω(4^k k mn)$ time. (3) Using the above result on single-sink important separators, we speed up undirected shadow removal and give the first almost-linear time FPT algorithm for Edge Multicut.

0.71

Dynamic $(Δ+1)$ Vertex Coloring with Adaptive Adversaries: Entropy Compression to the Rescue

cs.DS ·2026-10-28 ·Sepehr Assadi, Helia Yazdanyar
top venue
Abstract · 57 pages, 6 figures, to appear in SODA 2027

We present a randomized algorithm for maintaining a $(Δ+1)$ vertex coloring of a fully dynamic graph against an adaptive adversary with $\tilde{O}(\sqrt{n})$ amortized update time. Our algorithm improves upon the recent randomized algorithms of Behnezhad, Rajaraman, and Wasim (SODA 2025) with $\tilde{O}(n^{8/9})$ and Flin and Halldórsson (ICALP 2025) with $\tilde{O}(n^{2/3})$ amortized update time. Our result (nearly) matches a fundamental limit of $O(\sqrt{n})$ update time identified by Flin and Halldórsson and settles their main open question. The previous algorithms' main improvement came from efficient coloring of the dense parts of the graph. Our approach complements these prior work by handling the sparse parts more efficiently, using ideas inspired by the recently developed dynamic entropy compression method. A key new aspect of our work is to tailor this method specifically to dynamic algorithms---as opposed to original local search algorithms---and achieve update times that can be sublinear in the number of involved constraints per update.

0.60

Output-Sensitive Enumeration of Maximal Closed Substrings in Linear Space

cs.DS ·2026-10-28 ·Johannes Fischer, Dominik Köppl
Abstract

We revisit the output-sensitive enumeration of maximal closed substrings (MCSs). Given a string of length $n$ over an integer alphabet, let $k$ denote its number of MCS occurrences. We combine a right-to-left sweep on the string with range-successor queries to obtain all MCSs with two deterministic word-RAM bounds: for any fixed $\varepsilon>0$, we can report all MCSs in $O(n\sqrt{\log n}+k\log^\varepsilon n)$ time using $O(n)$ words of space, or in $O(n\sqrt{\log n}+k\log\log n)$ time using $O(n\log\log n)$ words. Both bounds include preprocessing and report occurrences by their endpoints. In particular, these algorithms improve the known $O(n\log n)$ worst-case time bound for sufficiently small output.

0.53

Estimating the number of nodes and paths in a DAG by following a random path

cs.DS math.CO ·2026-10-28 ·Günter Rote
Abstract · 28 pages, 5 figures, 10 tables, 4 appendices

We investigate a simple method to compute an unbiased estimate of the number of nodes of a directed acyclic graph (DAG) by following a random directed path. This method is due to Leonard Pitt from 1987, and it generalizes a method of D. E. Knuth from the 1960s for estimating the size of a rooted tree. A slight variation makes it possible to estimate the number of source-sink paths. We extend the algorithm to use many "agents" that walk cooperatively through the graph. This extension unifies and generalizes several methods that have been proposed for approximately counting the nodes of a tree.

0.53

Maxwell's equations are universal for sparse linear algebra

cs.CC physics.optics ·2026-10-28 ·Yannick Augenstein
theorycode/dataunusual
Abstract · 109 pages (18 pages main text, 91 pages Supplementary Information), 8 figures, 4 tables. v2: adds numerical illustrations (Fig. 3, SI Part V…

Optical and wave-based analog computing has re-emerged as a route to accelerating linear algebra, with metamaterials and photonic circuits shown to invert matrices and solve integral equations. However, it has remained unclear which problems such passive structures can represent, and at what cost, once encoding, precision, conditioning and readout are accounted for. Here, we show that every polynomially conditioned sparse linear system can be compiled into a three-dimensional dielectric -- a cavity filled with lossless, isotropic material of bounded permittivity -- whose time-harmonic response encodes its solution. For $m$ nonzero coefficients, the volume of the structure and the cost of writing it down and reading out the answer scale as $\tilde O(m^{3/2})$, coefficients require only $O(\log m)$ bits, and every field that approximately satisfies Maxwell's equations decodes to an approximate solution. We prove the construction for the continuum equations and give a computer-assisted version for the Yee discretization underlying finite-difference solvers. Read in reverse, it implies that electromagnetic simulation is, up to this overhead, as hard as general sparse linear algebra: a sufficiently fast general-purpose Maxwell solver would yield a faster sparse linear solver.

0.50

Exact Second-Order Asymptotics for the Gaussian Gray--Wyner Problem

cs.IT ·2026-10-28 ·Lin Zhou, Lei Yu
theory
Abstract

We revisit the Gray-Wyner problem of lossy compression of bivariate Gaussian sources under the quadratic distortion measure and derive the exact second-order asymptotics. Specifically, we first identify the boundary rate tuples using the explicit first-order asymptotics by Yu (TIT 2023) and then characterize the tradeoff among three second-order coding rates subject to a constraint on the joint excess-distortion probability. Our main contribution is the achievability proof, where we derive a covariance shell covering lemma that is analogous to the type covering lemma for the discrete case. In particular, our covering lemma classifies the source sequences into power shell types via their empirical covariance matrix. For each such power shell type, we derive sufficient conditions on the compression rates so that covering within distortion levels are achievable. Finally, asymptotical analysis via the Berry-Esseen theorem yields the desired result.

0.48

Simultaneous Gradient Learning in First-Price Auctions

cs.GT econ.TH ·2026-10-28 ·Janik Bürgermeister, Julius Durmann, Martin Bichler, Mete Şeref Ahunbay, Bary Pradelski +1 more
top venue
Abstract · accepted for publication at NeurIPS 2026

We study equilibrium learning in discretized Bayesian games, focusing on first-price auctions as a testbed for gradient-based learning dynamics under private information. While the symmetric independent private values model admits a unique Bayesian correlated equilibrium (BCE) in the continuous limit, standard convergence guarantees for gradient-based learners do not apply: classical no-regret theory predicts convergence only to Bayesian coarse correlated equilibria (BCCE), which can remain large and far from the competitive equilibrium even under fine discretizations. To bridge this gap, we introduce the notion of Bayesian semicoarse correlated equilibrium (BSCCE), an equilibrium concept that refines BCCE by restricting deviations to probability-preserving transformations consistent with projected gradient dynamics. We provide a linear programming characterization of BSCCE with polynomially many variables and constraints and establish the inclusion hierarchy $BCE \subseteq BSCCE \subseteq BCCE$. Our main finding is that the BSCCE set contracts in many cases as the discretization of the Bayesian game is refined. Even with priors for which BCCE fails to concentrate, the diameter of the BSCCE polytope shrinks toward zero, indicating convergence to the unique continuous-game Bayes-Nash equilibrium. These results provide a principled explanation for the observed robustness of gradient-based learning in discretized Bayesian games, showing that these dynamics select a smaller equilibrium set than standard no-regret theory predicts.

0.43

Fast Sparse $α$-Navigable Graph Construction with Improved Resource Augmentation

cs.DS ·2026-10-28 ·Alex Conway, Laxman Dhulipala, Martin Farach-Colton, Ben Landrum, Christopher Musco +3 more
top venue
Abstract · 26 pages. To appear in SODA 2027

Graph-based nearest neighbor search methods, which run greedy search on a directed graph over the data points, have surged in popularity in recent years. A common goal is to construct a search graph with the $α$-navigability property, which ensures that greedy search returns provably good approximate nearest neighbors and converges in few iterations. Increasing $α\geq 1$ improves accuracy and convergence rate, but requires more edges, which increases both space and search time. This tradeoff has motivated the study of algorithms for constructing a sparsest $α$-navigable graph for a given set of points. For any point set and any distance function, we show how to compute an $α$-navigable graph with at most $O(\log n)$ times as many edges as the sparsest $(1+\varepsilon)α$-navigable graph, in $O(n^ω\log(Δ)/\varepsilon)$ time, where $ω< 2.372$ is the matrix multiplication exponent and $Δ$ is the aspect ratio of the input. This improves on Khanna, Padaki, and Waingarten (SODA 2026), who achieve the same runtime but are only competitive with the sparsest $2(1+\varepsilon)α$-navigable graph. Since sparsity often increases rapidly with $α$ (typically $α\leq 1.1$), their bound can be significantly weaker. Previously, the best algorithm competing with the sparsest $(1+\varepsilon)α$-navigable graph required $\widetilde{O}(n^{2.5})$ time. Our techniques also extend to constructing sparse $τ$-monotonic graphs, another popular target in nearest neighbor search. Finally, we give a fine-grained complexity analysis suggesting that improving the $O(n^ω)$ running time is unlikely, or would require a very different approach. It is based on hardness results for problems related to verifying whether a graph is navigable, a topic of independent interest.

0.37

Ancilla-free randomized Toffoli gates and the T-count of Boolean oracles

cs.CC quant-ph ·2026-10-28 ·Arul Rhik Mazumder
code/data
Abstract · 27 pages, 4 figures, 1 table. Verification scripts: https://github.com/arulrhikm/randomized-oracle-checks

In fault-tolerant Clifford+T circuits, a Boolean oracle costs T gates and extra qubits, and on four or more qubits every exact Toffoli gate needs an extra qubit. Gosset, Kothari and Zhang showed that sampling a fresh random circuit for each call makes the T-count of the Toffoli gate independent of its size, but a direct implementation of their sampler uses about one extra qubit per bit of precision. We show that computing the sampled parities in place, on the gate's own controls, removes every extra qubit once the gate is large enough: each random circuit then uses about 8 T gates per bit of precision and borrows only the gate's own idle inputs. This is one T gate more per circuit than the best measurement-free variant that keeps one clean ancilla. The same holds for exclusive-ors of indicators of affine subspaces, such as equality tests and checks of linear equations. For randomized circuits that do not measure, this is within a factor of eight of optimal up to an additive constant, even with extra qubits, and when each random circuit computes a Boolean function exactly, closing the factor asymptotically is equivalent to determining the leading constant in the T-count of exact multi-controlled Toffoli gates. Finally, the oracle of the indicator of a random set of fixed density needs exponentially many T gates at constant error, while that of an affine subspace of the same density needs a constant number, and extra qubits reduce the exponential cost only quadratically.

0.35

Vendster: An Evaluation Arena to Investigate How Forecasting Tool Quality Affects LLM Agent Behaviors in Buyer-Vendor Negotiation

cs.GT cs.AI ·2026-10-28 ·Charin Polpanumas
code/data
Abstract

Time-series forecasters are ranked almost entirely by accuracy metrics, but a forecast's decision value, its effect on the downstream decision, need not track accuracy and is rarely measured. This problem compounds as large language model (LLM) agents increasingly consume forecasters as tools, mediating them through stochastic, opaque policies. We present Vendster, an evaluation arena measuring how a forecaster's bias and prediction-interval width change the outcome of a buyer--vendor discount negotiation between two LLM agents. In the reference instance Vendster-M5, sampling base demand patterns from the M5 dataset, we evaluate two frontier-closed LLM families (Claude Opus 4.6 and GPT-5.6 Sol) and two open ones (DeepSeek-V3.2 and Qwen3-235B-A22B). Model-generated forecasts raise sales target attainment by up to +17 percentage points (p<0.001), yet standard forecast evaluation metrics such as relWQL and MASE do not predict negotiation outcomes at a statistically significant level. To isolate the causal structure, we substitute in synthetic forecasts with controlled bias and interval width. Under-prediction bias significantly improves outcomes for agents that react more to the forecasts---dose-response slopes -0.387 and -0.416 (p<0.001). Interval width has no statistically significant effect at zero bias. These findings indicate that the direction of forecast bias and agent forecast-reactivity are more strongly associated with downstream outcomes than forecast accuracy. Vendster is fully customizable and released under CC BY-NC 4.0 at https://github.com/amazon-science/Vendster.

0.34

Fast Perfect $L_2$ Sampling

cs.DS ·2026-10-28 ·David P. Woodruff, Samson Zhou
published
Abstract · PODS 2027

Perfect $L_2$ sampling is a powerful primitive for turnstile data streams, allowing algorithms to extract an index with probability proportional to its squared frequency, with applications to dynamic join size estimation, query optimization, heavy-hitters, sampling from self-joins, and real-time monitoring of evolving datasets. The setting $p=2$ is also the largest $p$ for which $L_p$ sampling can be achieved in polylogarithmic space. However, existing space-optimal perfect samplers rely on explicit polynomial duplication, requiring inefficient polynomial update times per stream arrival. Resolving a open question left by Swartworth, Woodruff, and Zhou [FOCS 2025], we present a perfect $L_2$ sampler that achieves an expected $\text{polylog}(n)$ update time while maintaining an optimal $\tilde{\mathcal{O}}(\log^3 n)$ bits of space.

0.34

Beyond Reachability: When Edit Order Preserves Couplings in Graph Flow Matching

cs.IT cs.LG ·2026-10-28 ·Zilin Li, Zixing Song
Abstract · 14 pages. Theoretical preprint

Shortest-path optimality does not guarantee that a memoryless graph generator preserves a prescribed source-target correspondence. We study whether reordering legal shortest edits can retain this correspondence without changing the pairing, mandatory edits, or sampling budget. Under a common Poisson clock, we establish an exact criterion for endpoint coupling recoverability: the source-blind Markov projection attains the unavoidable clock-error floor if and only if routes to distinct targets do not share graph states. Routes to the same target may merge without sacrificing endpoint fidelity. This characterization leads to an exact finite-state routing formulation and a legality-preserving scheduling procedure with recovery certificates. A marker-free construction demonstrates how shared edge structure can retain source distinctions, while a capacity-constrained family exhibits a sharp separation in paired error between edit orders with identical marginal error. Finally, a target-conditioned aliasing bound connects structural route design to the terminal fidelity of a learned graph-only generator. These results characterize when legal edit ordering can preserve a prescribed coupling under fixed operational costs.

0.33

Determination of weight enumerators of binary extremal self-dual codes of length 60

cs.IT math.CO ·2026-10-28 ·Masaaki Harada
Abstract · 9 pages

We construct three binary extremal self-dual codes of length $60$ with weight enumerators for which no binary extremal self-dual codes of length $60$ were previously known to exist. This completes the determination of the weight enumerators that occur for binary extremal self-dual codes of length $60$.

0.33

All-even sets in $\mathbb{R}^2$ and $\mathbb{R}^3$

cs.CG math.CO ·2026-10-28 ·Daniel Kalmanovich, Gabriel Nivasch
theory
Abstract · 32 pages, 10 figures

Call a finite nonempty point set $S\subset\mathbb{R}^d$ all-even if for every direction $\ell \in \mathbb{P}(\mathbb{R}^d)$, the number of hyperplanes perpendicular to $\ell$ that intersect $S$ is even. In 2026, Francisco Santos presented a question of Juan Benigno Seoane-Sepúlveda on the existence of all-even sets in dimensions $3$ and higher, together with examples of such sets in dimension $2$. Call a set $S\in\mathbb{R}^d$ line-all-even if for every direction $\ell \in \mathbb{P}(\mathbb{R}^d)$ the number of lines parallel to $\ell$ that intersect $S$ is even. As Santos pointed out, every all-even set in $\mathbb{R}^3$ is line-all-even. We conjecture that there exist no line-all-even sets in $\mathbb{R}^3$ that affinely span $\mathbb{R}^3$ We show that our conjecture is equivalent to the statement that every planar all-even set satisfies a certain property, which we call height-rigidity. This property has a natural interpretation in linear algebra terms. Furthermore, we present a hierarchy of properties of planar point sets that imply height-rigidity. We construct several all-even sets in the plane, including some infinite families, and show that all the sets we constructed are height-rigid. In addition, we prove that every all-even set in the plane that can be covered by six or fewer parallel lines is height-rigid.

0.30

Robustness of Magic and the Directed Geometry of State Conversion

cs.IT quant-phmath.QA ·2026-10-28 ·Satoshi Kanno, Yoshi-aki Shimada
Abstract · 25 pages

In this work, we introduce a state-space distance calibrated by stabilizer states based on the dual structure of the robustness of magic (RoM), and show that RoM is exactly represented as the distance to the free-state set. We then extend this geometry to state conversion by defining the directed conversion deficiency $C_{\mathcal F}$ as the distance between a target state and the set of states freely reachable from a given input state. The quantity $C_{\mathcal F}$ characterizes free convertibility and is related to the minimum stabilizer quasiprobability weight required to represent the conversion residual through the relation $1+2C_{\mathcal F}$. It also contains the conventional magic state resource lower bounds based on RoM and can detect stronger conversion obstructions. For single-qubit systems, we analytically construct a continuous family of mixed states with identical RoM but asymmetric convertibility.

0.30

The 120-Cell Dissected by the Hyperplanes of Its Final Stellation: the golden ring Z[phi] as foundation, from Kramer's seven cells to 2,455,201 cells in 338 orbits

cs.CG math.MGmath.NT ·2026-10-28 ·Miguel E. Pérez Ibarra, Jaziel Flores Rodríguez
Abstract · 37 pages, 14 figures; catalogue of the 338 orbits and the output of the exact checks in the appendices

The diagonals of a regular pentagon dissect it into eleven pieces and draw a pentagram; the dodecahedron dissected by the twelve face planes of its inner core gives Kramer's seven "elementary cells" of 1982. We define the stellation dissection of a golden regular polytope in any dimension as the dissection of the convex hull of its final regular stellation by the facet hyperplanes of the core, show that the core has phi^(-d) of the inradius of the hull for d = 2, 3, 4, and carry the dissection out for the first time on the 120-cell. The foundation is the arithmetic of the golden ring Z[phi], the ring of integers of Q(sqrt 5): vertices and hyperplanes live in Z[phi]^d, every orientation test is the exact sign of an element of Z[phi], decided with integers, and the pieces are named by sign vectors. We develop this foundation from scratch (conjugation, norm, units, the primes sqrt 5, 2, 3, the exact sign test, the lattice x -> (x, conj x), the vertex modules of the dodecahedron and the 120-cell, the icosian ring) and let it carry the geometry. The definition bifurcates in dimension four: the star cut at depth phi^(-4), a unit, has 2,455,201 cells in 338 orbits of H4; the hereditary cut at depth 1 - phi^(-4) = sqrt 5 / phi^2, a unit times the ramified prime, induces the three-dimensional dissection on every facet and has 8,161 cells in 8 orbits. New here: all vertices of the dodecahedral and of the hereditary dissection lie in the vertex module of the polytope being cut, whereas the star cut needs the denominators 2, 3 and 5; the seven kinds of pieces of the dodecahedron are exactly the I_h-orbits of sign vectors, with stellation layer equal to the number of plus signs. We catalogue the four-dimensional pieces and treat the circumscribed disc, ball and 4-ball. Every number quoted is recomputed exactly by a script whose output is reproduced in an appendix.

0.30

Moments of Magic Squares: Line-Sum Conditions, a Permutation Threshold, and Random Energies

cs.DM math.COmath.NTmath.PR ·2026-10-28 ·Kyle Elliott Mathewson
theory
Abstract · 29 pages, 3 figures, 4 tables; ancillary verification scripts, C enumerators, and exact lattice certificates included

Every semi-magic square of order $n$ satisfies the axial moment identities $\sum_{i,j}s_{ij}j^m=\sum_{i,j}s_{ij}i^m=M(n)\sum_jj^m$ for all $m\ge1$, where $M(n)=n(n^2+1)/2$. For real matrices the converse requires $1\le m\le n-1$, whereas for arrangements of $\{1,\dots,n^2\}$ fewer moments suffice once $n\ge14$. We define the permutation threshold $k^*(n)$ as the least $k$ such that every arrangement satisfying these identities for $1\le m\le k$ is semi-magic. In terms of the column-sum deviations $γ_j$, the degree-$k$ conditions state that $\sum_jγ_jt^j$ is divisible by $(t-1)^{k+1}$, linking $k^*(n)$ to the Prouhet--Tarry--Escott problem. While Schur's bound on real zeros gives $k^*(n)\le\sqrt{2(n-1)\log(n^4/4)}$, we prove the converse realisability bound $k^*(n)\ge(\sqrt2-o(1))\sqrt n$ (and $k^*(n)\ge(\sqrt6-o(1))\sqrt n$ when $4\mid n$) by perturbing semi-magic squares with complementary row pairs via horizontal swaps. We also show that $k^*(n)=n-1$ for $3\le n\le13$, $k^*(n)=n-2$ for $14\le n\le17$, and $k^*(n)\le n-3$ for $n\ge18$, and determine $k^*(n)$ for $18\le n\le40$ with computer assistance. Finally, at the $k=1$ level of the hierarchy, we specialise the Mantel--Hubert quadratic assignment variance formula to obtain exact means, variances, and limit laws under a uniform random arrangement for the full line-sum energy $E_{\mathrm{full}}$, for the four-term energy $E_4$ combining the first moments and the two diagonals, and for Loly's moment of inertia, and find that $|L_4|=14,639,440$ arrangements of order $4$ satisfy $E_4=0$.

0.29

What Can Firms Do in the Face of Model Distillation?

cs.GT cs.AI ·2026-10-28 ·Diptangshu Sen, Jibang Wu, Juba Ziani
Abstract · 32 pages

Model distillation has become a major concern in the rapidly growing generative AI discourse. In its simplest form, distillation is the phenomenon where smaller firms can use data distilled from transcripts of conversations with powerful proprietary closed-weight models to train their own models, thus achieving reasonably high model quality but at much lower costs. Besides the legitimate safety and security concerns associated with distillation, another major issue is how it can also distort marketplace outcomes. This motivates us to initiate an exploration of the problem from an economic theory lens. We consider a setting with two players: an incumbent and a distilling competitor and characterize how the incumbent's model characteristics drive the incentives around how much the competitor should distill. We also consider the impact of distillation on key market outcomes like user-side surplus and the incumbent's utility. Our key contribution, however, is to study interventions for preventing or limiting distillation attacks. In this context, we study two settings: i) when it is possible, with a certain probability (which grows in the extent of distillation conducted), to detect an attack on the incumbent's model; and ii) when a distillation attack cannot be reliably detected. In the first setting, we show that penalty-based regulations can be very effective in completely deterring distillation, although it can have potentially adverse effects on market competition and user surplus. In the latter setting, we show that the incumbent can use self-distillation as a mitigation strategy. Importantly, the effectiveness of self-distillation depends on how cost-efficiently the incumbent can deploy a new self-distilled model on the market. On the positive side, self-distillation can preserve market competition thus protecting user surplus to some extent.

0.28

Capacity of the Hopfield-Hebbian dynamics with correlated patterns

cs.IT stat.MLcond-mat.dis-nncs.LG+1 ·2026-10-28 ·Mihailo Stojnic
Abstract

Allowing for small fraction of errors in retrieved stored binary patterns, Hopfield-Hebbian neural network (NN) associative memory dynamics [14] achieves linearly scaled capacity. The famous proportionality constant $α_c\approx 0.14$ is obtained in a large dimensional scenario as the ratio of the number of reliably stored patterns $m$ and the patterns' length $n$ under the assumption that uniformly randomly chosen binary patterns are \emph{independent} among themselves. While such an idealized scenario is now well understood, more practical scenarios with correlated/noisy datasets remaining challenging even for classical statistical mechanics replica approaches. We consider precisely such practical settings where each \emph{archetype} pattern is replaced with $M$ correlated \emph{examples} with parameter $r$ governing the quality of correlation. We develop a powerful Random Duality Theory (RDT) based framework to analyze correlated data and already on the first level of RDT lifting obtain capacity characterizations that remarkably well match behavior exhibited through simulated dynamics. Moreover, characterizations explicitly feature correlation parameters $M$ and $r$, which allows to precisely determine their effects on phase-transitioning behavior of the underlying dynamics. Our results are also a well-fitting complement to the existing literature. Namely, to circumvent challenging statistical mechanics analyses, an alternative combination of spectral and replica methods is considered in [4]. We confirm that [4]'s elegant spectral characterizations are excellent approximations in the low correlated regime. At the same time, we also close the gap in the regimes with higher correlation levels where spectral methods deviate from the practical dynamics performance.

0.28

Separations between $\#\mathrm{BIS}$ and $\#\mathrm{IS}$

cs.CC cs.DS ·2026-10-28 ·Ijay Narang
Abstract

We give complexity-theoretic separations between $\#\mathrm{BIS}$ and $\#\mathrm{IS}$. Given a graph $G$, and a fugacity $λ>0$, the hard-core model is given by $Z_G(λ)=\sum_{I\in\mathcal I(G)}λ^{|I|}$, where $\mathcal I(G)$ denotes the independent sets of $G$. For $F\ge1$, an $F$-factor approximation to $Z_G(λ)$ is an estimate $\widehat Z$ satisfying $F^{-1}Z_G(λ)\le\widehat Z\le FZ_G(λ)$. Let $\mathcal G_Δ$ denote the class of $Δ$-regular graphs and $\mathcal G_{Δ,Δ}$ the class of $Δ$-regular bipartite graphs. We give two separation results. (1) There exist an integer $Δ\ge3$, a sufficiently large fugacity $λ$, and a constant $c>0$ so that the following hold: (a) For every $δ\in(0,1)$, deterministic $(1+δ)$-factor approximation on $\mathcal G_{Δ,Δ}$ is possible in time $\exp(cn)\operatorname{poly}(n,1/δ)$. (b) Under the Gap-Exponential Time Hypothesis, there exists no deterministic algorithm running in time $\exp(2cn)\operatorname{poly}(n)$ on $\mathcal G_Δ$ that gives an $\frac{11}{10}$-factor approximation. (2) For every fixed $Δ\ge3$ and $\varepsilon\in(0,1)$, for all sufficiently large fugacities $λ$, the following hold: (a) There exists an algorithm which outputs a deterministic $(1+\varepsilon)^n$-factor approximation to $Z_G(λ)$ on $\mathcal G_{Δ,Δ}$ in polynomial time. (b) Under the assumption $\mathrm P\ne\mathrm{NP}$, there exists no deterministic polynomial-time algorithm on $\mathcal G_Δ$ that gives a $(1+\varepsilon)^n$-factor approximation. Both our algorithmic results follow from a new spanning-tree encoding of independent sets that enables an efficient representation of large independent sets.

0.27

Is Dual Polarization Always Optimal? Unlocking New Spatial Degrees of Freedom

cs.IT eess.SP ·2026-10-28 ·Adrian Agustin, Xavier Mestre, Carlos Suárez García, Alberto Pellon Sancho
Abstract · Submitted to IEEE Communications Magazine

Current wireless communication systems are designed based on channel models obtained under the assumption of far-field operation, where electromagnetic waves can be approximated by a planar wave. In those circumstances, dualpolarized antennas seem to be the optimal solution for maximizing the spectral efficiency. However, the recent research on XLMIMO systems represents a paradigm shift, since communication may take place in the near-field and an additional degree of freedom can be achieved. This work presents a modular, lowcost tripolarized antenna at 2.4 GHz band, that is used to experimentally validate theoretical findings about tripolarized antennas, which are commonly based on orthogonal infinitesimal dipoles. The evaluation is carried out in two indoor scenarios and one outdoor scenario, with transmitter comprising multiple tripolarized antennas and a receiver equipped with a single tripolarized antenna. We show that this type of antenna can increase the number of spatial streams up to three or extend the region where terminals are able to achieve two spatial streams

0.27

MacKay-Neal Codes for DNA Channels under Block-MAP Decoding

cs.IT ·2026-10-28 ·Kenta Kasai
theory
Abstract · 14 pages

We prove achievability below the symmetric information rate (SIR) under block maximum a posteriori (MAP) decoding for a fixed-degree MacKay-Neal (MN) socket ensemble on the quaternary insertion-deletion-substitution channel used to model DNA reads. For every integer $\ell\ge4$, the $(\ell,3,3)$ socket ensemble, with one transmitted bit fixed and an independent uniform known offset, has limiting rate $r_\star=6/\ell$ bits per stored nucleotide. If the joint SIR $J_t$ of a fixed number $t$ of reads under independent, uniformly distributed input nucleotides exceeds $r_\star$, the ensemble-average full extended-word block error is $O(1/n)$, and deterministic graph-offset sequences attain the same limiting rate and message-average error bound. This result concerns the discrete rate family $6/\ell$, not arbitrary rates approaching a given SIR. The channel permits arbitrarily many insertions and unbounded drift, with no alignment supplied to the receiver. The proof combines established MN parity and distance estimates with a strict arbitrary-candidate-list bound and a list construction that accounts for hidden output boundaries.

0.26

A Witnessing Duality for ReLU Network Depth

cs.CC cs.LGcs.LO ·2026-10-28 ·Dimitrios Myrisiotis
theory
Abstract · 132 pages, 1 figure, 3 tables

A depth lower bound for ReLU networks asserts that every sufficiently shallow network fails on some input. What does a proof of such a bound compute? We study this question for exact computation of the maximum of $n$ real inputs, with depth measured by the number of hidden layers. For every separately fixed prime $q$ and depth bound $K$, we give a uniform $\mathsf{TC}^0$ refuter for rational networks with at most $K$ hidden layers and more than $q^K$ inputs, provided that the reduced denominators of their nonbias weights are not divisible by $q$. Biases may be arbitrary rationals. The refuter finds a rational input on which the network fails, and its correctness is provable in $\mathbf{VTC}^0$, the bounded-arithmetic theory of uniform constant-depth threshold computation. The underlying numerical depth lower bound follows from Averkov, Hojny and Merkert (ICLR 2025), extending Haase, Hertrich and Loho (ICLR 2023); our contribution is an internal proof that makes its geometric argument into a feasible refuter. On the construction side, $\mathbf{VTC}^0$ verifies locally checkable certificates for the balanced integer-weight maximum network of depth $\lceil\log_2 n\rceil$. At $q=2$, the construction and refuter meet: for each fixed $K$, the theory verifies that integer-nonbias-weight networks with arbitrary rational biases can compute the maximum with at most $K$ hidden layers exactly when $n\le 2^K$. We also give certified rational constructions of polynomial encoding length and depth $O(\log n/\log\log n)$. These results instantiate a witnessing duality: bounded proofs of certified existence and of refutation correspond, respectively, to threshold-computable constructors and refuters whose guarantees the theory proves.

0.26

A Characterization of Functions on the Integer Lattice That Are L-Extendable through Local Convex Extensions

cs.DM math.CO ·2026-10-28 ·Ken Yokoyama, Yuni Iwamasa, Kei Kimura
Abstract · 17 pages, 3 figures

L-extendable functions on the integer lattice are functions that admit a $\frac12$UJ-convex relaxation; that is, values can be assigned to the additional half-integral points, while keeping the original values at integer points, so that the resulting function is $\frac12$UJ-convex. Various examples of L-extendable functions and their applications are known. A general characterization of L-extendability for functions on the integer lattice, however, remains open. In this paper, we study the L-extendability of functions on the integer lattice. We first obtain a necessary condition: L-extendable functions are integrally convex. We then consider a special subclass of L-extendable functions, namely, those whose local convex extensions give L-convex relaxations. This subclass has a useful optimization property: the optimal value of the original function coincides with that of the $\frac12$UJ-convex relaxation given by its local convex extension, a property not guaranteed for general $\frac12$UJ-convex relaxations. We characterize this subclass by showing that a function is L-extendable through its local convex extension if and only if it is integrally convex and its restriction to every unit hypercube is D-convex, where D-convexity is a discrete convexity associated with reflected Freudenthal simplicial divisions. This subclass strictly contains the class of D-convex functions. We also give a combinatorial characterization of D-convex subsets of the vertices of a unit hypercube: a set $X\subseteq\{0,1\}^n$ is D-convex if and only if it is majority-closed and contains a relative opposite pair.

0.26

Random Non-Commutative Error-Correcting Codes

cs.IT math.GR ·2026-10-28 ·Michael Best, Alexander Lubotzky
Abstract · 18 pages, 1 table

Chapman, Dinur, and Lubotzky recently introduced the theory of non-commutative error-correcting codes, generalizing classical coding theory to arbitrary groups via $δ$-robust subsets used to detect proper subgroups. Building on this foundation, this paper systematically studies an analogue of random codes for arbitrary groups, exploring the extent to which the classical Gilbert-Varshamov (GV) bound holds in this non-commutative setting. While random universal codes of linear size in free groups are shown to be asymptotically bad, we establish a complete characterization for general families of finite groups. Specifically, a family of finite groups admits good random codes (forming a GV family) if and only if it exhibits uniform polynomial maximal-subgroup growth. Furthermore, we provide explicit quantitative bounds on the rate and optimal distance range based on intrinsic parameters, proving that families of finite solvable groups always form GV families and determining the exact probabilistic rate for direct powers of a fixed finite simple group.

0.25

Improved LCL Equivalences and Curve Decoding Alphabet Sizes

cs.IT cs.CR ·2026-10-28 ·Rohan Goyal, Venkatesan Guruswami, Yihang Sun, Mary Wootters
theory
Abstract · 60 pages

Local Coordinate-wise Linear (LCL) properties, introduced by Levi, Mosheiff, and Shagrithaya (2025), capture coding-theoretic properties such as list decoding and list recovery. They showed that every LCL property has the same rate threshold for random linear codes (RLC) and random Reed--Solomon codes (RRS). Brakensiek, Chen, Dhar, and Zhang (2025) extended this to random additive codes. Thus any property of all subspace design codes holds for RLCs and RRS at the same rate. In this work, we improve the probabilistic guarantees in the RRS local equivalence theorem and extend the equivalence to all known subspace design constructions individually. Each such construction, including folded Reed--Solomon codes, has the same LCL threshold as RLC and RRS codes. We apply these results to alphabet-size bounds for curve decoding. Curve decoding, introduced by Goyal and Guruswami (2025) and Carmon et al. (2026), is a row-span-constrained LCL property implying proximity gaps (PG), correlated agreement (CA), and mutual correlated agreement (MCA), relevant to IOPs and SNARKs. Informally, a rate-$R$ code $C \subset \mathbb{F}^n$ is $(\ell,δ,a,b)$-curve decodable if any degree-$\ell$ curve $u:\mathbb{F}\to\mathbb{F}^n$ that is $δ$-close to $C$ on at least $a$ points has at least $b\leq a$ nearby codewords lying on a degree-$\ell$ curve in $C$. Goyal and Guruswami (2025) and Goyal, Guruswami, Sun, and Wootters (2026) established curve decodability for RLC and RRS codes at radius $δ$ approaching $1-R$, but with large alphabets in $\ell$ and $η:=1-R-δ$. Using our improved LCL equivalence and rank-sensitive counting inspired by Goyal and Guruswami (2026), we exponentially improve alphabet requirements while matching the best known values of $(a,b)$. For $\ell=O(1)$, our alphabet-size bounds match the best known for list decoding, and improves results for PG, CA, and MCA.

0.25

Rydberg Atomic Quantum Receiver Empowered Hybrid Non-orthogonal Multiple Access

cs.IT cs.NI ·2026-10-28 ·Xinya Shen, Zheng Chu, David Chieng, Chiew Foong Kwong, Qihao Peng +3 more
Abstract · 13 pages, 13 figures

In this paper, we propose a novel uplink Hybrid Non-Orthogonal Multiple Access (NOMA) architecture assisted by Rydberg Atomic Quantum Receivers (RAQRs) to improve energy efficiency by addressing the physical constraints of atomic sensing. We formulate a joint optimization problem encompassing phase measurements and multi-user power allocation to minimize the total transmit power. This problem is particularly challenging due to the non-linear amplitude-only observations of atomic vapor cells. To decomposes the resulting non-convexity, we develop a hierarchical three-layer decomposition optimization framework that systematically decouples the global problem. The inner layer employs a customized Expectation Maximization-Gerchberg-Saxton (EM-GS) algorithm to resolve phase ambiguity and estimate equivalent channel gains. The middle layer derives closed-form optimal power allocation solutions via Karush-Kuhn-Tucker (KKT) conditions. The outer layer utilizes linear Minimum Mean Square Error (MMSE) detection for high-reliability signal recovery. Simulation results demonstrate the superiority of the proposed framework over conventional RF-based Hybrid NOMA. Notably, the atomic architecture achieves approximately a 93% reduction in total power consumption compared to Orthogonal Multiple Access (OMA) baselines while maintaining strict backward compatibility, thereby laying a solid foundation for practical deployment in next-generation green communication networks.

0.25

Generative Rate-Distortion-Perception Theory on Riemannian Manifolds

cs.IT ·2026-10-28 ·Tong Wu, Zhiyong Chen, Meixia Tao, Dazhi He, Feng Yang +1 more
theory
Abstract

Generative communication uses transmitted information to control stochastic generation at the receiver. A fundamental question is how much source-specific information must be communicated so that outputs generated at the receiver satisfy prescribed fidelity and reliability requirements while preserving the conditional source distribution. We address this question for sources on compact Riemannian manifolds by developing a generative rate-distortion-perception (GRDP) framework without common randomness. We formulate an asymptotic operational GRDP function under a geodesic distortion radius and a failure-probability constraint, and establish its single-letter characterization through an equivalent informational formulation. Furthermore, for the informational formulation with the high-reliability limit, we derive a geometric lower bound determined by the largest source probability mass of a region with bounded geodesic diameter, together with an upper bound based on probabilistic coverings of the source manifold. Small-radius analysis of the lower bound reveals the roles of intrinsic dimension and scalar curvature in the required communication rate. For uniformly distributed sources on compact homogeneous Riemannian manifolds, we prove that the bounds coincide, yielding an exact characterization of the limiting GRDP function. We further specialize this result to unit spheres, obtaining an explicit expression that relates the minimum communication rate to the geodesic distortion radius and intrinsic dimension, revealing that for the $d$-dimensional unit sphere with $d\ge2$, increasing the communication rate by $d$ bits per source symbol can reduce the distortion radius by more than half.

0.25

Dimension-free exponential entropy quotients over $\mathbb{F}_2$

cs.IT math.COmath.PR ·2026-10-28 ·Josiah Park
theory
Abstract · 14 pages, no figures

Let $X$ be a random vector in $\mathbb{F}_2^m$, and let $X'$ be an independent copy. We prove that, for every $\varepsilon>0$, there is a subspace $G$ with $\dim G\le 2^{320(1+\varepsilon)/\varepsilon}H(X)$ such that $H(q_GX)\le(1+\varepsilon)(H(q_GX+q_GX')-H(q_GX))$. In particular, the quotient entropy is at most $(1+\varepsilon)(H(X+X')-H(X))$. The coefficient is independent of the ambient dimension, and its exponential order in $1/\varepsilon$ is necessary. The structural result is a logarithmic rigidity estimate for finite weighted families of laws with averaged subspace entropy sparsity. A common variational quotient controls both a family and its convolution family without decreasing the original family's normalized mean entropy defect. Posterior families preserve averaged sparsity, and a fresh-index argument combines a local-to-global estimate with the entropic Marton theorem to increase the defect ratio at a fixed multiplicative sparsity cost. We also obtain two-input almost-additivity and concentration of moderate-doubling sets on subspace cosets at the same exponential scale. The constants are not optimized.

0.25

RARE: Response-Aware Rational Equilibrium in Budget-Constrained Pricing Games

cs.GT ·2026-10-28 ·Bhaskar Ray Chaudhury, Jugal Garg, Xin Huang, Eklavya Sharma
theory
Abstract · Changes in v2: improved title, exposition, and bibliography

We study a pricing game in markets with budget-constrained buyers, where each seller offers a single good and sets its price. We consider both non-rivalrous goods, such as datasets, and rivalrous goods, such as physical goods. For non-rivalrous goods, each buyer chooses a utility-maximizing bundle within her budget. For rivalrous goods, an allocation rule selects a Pareto-optimal allocation subject to the buyers' budgets and the goods' supply constraints. Pure Nash equilibria often fail to exist in this game. A key reason is that Nash equilibrium assumes that the other sellers keep their prices fixed when one seller deviates. In practice, however, sellers harmed by a price change may respond, for example, by matching the new price. We propose a new equilibrium notion called Response-Aware Rational Equilibrium (RARE). At a RARE, every deviation by a seller can be deterred by a response from the other sellers, in which no responder earns less than she did right after the deviation. Every Nash equilibrium is a RARE. Assuming that the allocation rule is continuous in prices, we show that a RARE always exists for rivalrous goods, and for non-rivalrous goods when buyers have equal budgets. Moreover, the RARE that we construct extracts the buyers' entire budget. For non-rivalrous goods with unequal budgets, we show that an approximate RARE exists, where a deviating seller can increase her revenue by a factor equal to the ratio of the average budget to the minimum budget. We also study a natural dynamic in which sellers repeatedly raise their prices to earn the revenue that they could guarantee themselves by deviating. We show that, in the settings where we prove exact existence, this dynamic reaches a RARE with additive error $δ$ within $\lfloor B/δ\rfloor$ rounds, where $B$ is the buyers' total budget. This paper's main results have been formally verified in Lean 4.

0.24

Polynomial-Time Constructions of Asymptotically Good Random Access Codes from Linear Codes

cs.IT ·2026-10-28 ·Ruho Kondo, Yota Maeda
Abstract · 13 pages, 5 figures,

Random access codes (RACs) encode a $q$-ary input string into fewer $q$-ary symbols while allowing any requested input symbol to be recovered with probability above the random-guessing value $1/q$. However, it remains unclear what asymptotic guarantees on compression and recovery probability should be required of RACs. We address this question by introducing the notion of asymptotically good RACs and give two polynomial-time constructions. First, an $[n,k,d]_q$ linear code with $d\ge2$, together with an encoder based on its orthogonal-array property, yields an $(n,n-k)_q$-RAC with worst-case decoding success probability \[ p_{\rm wc}=\frac{1}{q}+\left(1-\frac{1}{q}\right)\frac{d-2}{n}. \] These OA-RACs are asymptotically good whenever the underlying linear codes are, and Garcia--Stichtenoth algebraic-geometry codes give a concrete polynomial-time realization. Second, a direct-sum strategy using OA-RACs derived from $q$-ary Hamming codes and the trivial $(1,1)_q$-RAC yields asymptotically good RAC families at every prescribed RAC rate $R_{\rm RAC}=(n-k)/n\in(0,1)$. Numerical results demonstrate favorable finite-length performance of both constructions.

0.23

How Hard Is It to Compute Cover Time?

cs.CC ·2026-10-28 ·Ryan Aghazadeh
Abstract · 24 pages including title page, 3 figures. Substantially reorganized and shortened from 29 pages, with simplified proofs, expanded explanatio…

It is $\#\mathrm{P}$-hard. Even for connected simple undirected unweighted graphs, deciding whether an expected random-walk cover time is at most a given rational threshold is $\#\mathrm{P}$-hard under polynomial-time Turing reductions, both from a specified starting vertex and when maximized over starting vertices. Hardness holds even for graphs of diameter two with a universal vertex, which we call the "hub". Deleting the hub leaves a planar bipartite graph, and every non-hub vertex has degree at most five. A shared construction encodes an Euler-tour count in rare covering walks. For specified starts, a number of fractional bits linear in the queried graph's vertex count suffices to recover the Euler-tour count. For the maximum-start problem, a fixed cube attachment makes the maximizing starts known, and one finite-difference calculation on expected cover times recovers the count. For general graphs, both comparison problems can be solved in polynomial time with access to an $\mathrm{NP}^{\mathrm{PP}}$ oracle.

0.23

Larger Crossing Families

cs.CG math.CO ·2026-10-28 ·Joachim Orthaber, Patrick Schnider
Abstract

We give the first explicit constant for crossing families of linear size by showing that every set of $n$ points in general position in the plane contains a crossing family of size approximately $\frac{n}{79}$. We further identify some conditions where there are even larger crossing families. The first such condition forbids the existence of a large different pattern, namely non-convex 4-clusters: We show that every point set contains either a large non-convex 4-cluster or a large crossing family. The second condition is the existence of many crossings: We give a bound for the size of a crossing family in terms of the number of crossings of the complete geometric graph drawn on the point set.

0.23

Principal-agent hypothesis testing under adaptive experimentation

cs.GT stat.MEmath.ST ·2026-10-28 ·Ben Chugg, Aritra Ghosh, Aaditya Ramdas
Abstract · 26 pages; 3 figures

Principal-agent hypothesis testing studies statistical contracts in which a principal rewards an agent according to data the agent generates. Existing work focuses on the single-shot setting, where the agent produces a single dataset. We extend the framework to sequential settings, where evidence accumulates over time, rewards may be deferred to the end or disbursed incrementally, and the agent may stop the contract at a strategically chosen stopping time. We call a contract null-deterring if no ''null agent'' (those whose data-generating distribution lies in the null hypothesis) can profit in expectation under any stopping rule, and our main conceptual result shows that such contracts are characterized by e-processes. We then turn to the agent's best response and identify the optimal stopping rule through the Snell envelope of the reward process. We develop simpler, more interpretable approximations that often require the agent to have only partial knowledge of its distribution.

0.22

A $π/ 4$ lower bound for random access in symmetric linear ensembles under optimal decoding

cs.IT ·2026-10-28 ·Daniella Bar-Lev, Ohad Elishco, Ryan Gabrys
theory
Abstract

We study the expected number of random linear observations required to recover a prescribed information symbol. For ensembles with uniformly random supports, independently sampled nonzero coefficients, and uniformly bounded column weights, we prove that the asymptotic normalized expected recovery time is at least $π/4$ under optimal decoding. The result holds over finite or infinite fields and extends the lower bound previously established for binary fully symmetric codes under peeling decoding. For singleton - pair observations, we determine the exact limiting expectation and show that optimal and peeling decoding have the same limit, independently of the field and coefficient distribution. The minimum within this family is approximately $0.793933445$, which explicit constructions attain asymptotically over every fixed finite field.

0.22

Fully Dynamic All-Pairs Min Cuts on Planar Graphs via Planar Pattern Emulators

cs.DS ·2026-10-28 ·Arnold Filtser, Gramoz Goranci, Ali Momeni, Neel Patel, Maximilian Probst Gutenberg
Abstract

We study the fully-dynamic all-pairs minimum cut problem on planar graphs. Given an $n$-vertex undirected planar graph $G=(V,E)$ undergoing edge insertions and deletions, the goal is to efficiently process these updates while supporting $s$-$t$ minimum cut queries for any pair of vertices $s$ and $t$. We give a $(1+\varepsilon)$-approximate dynamic algorithm that supports edge updates and minimum cut queries in $n^{o(1)}$ time, for any $1/\text{poly}(\log n) < \varepsilon < 1$. For general graphs, previous results either are limited to partially dynamic settings or achieve only logarithmic-factor approximations. The main conceptual and technical contribution is to show that a suitable adaptation of the planar emulator of Chang, Krauthgamer, and Tan~[STOC'22] satisfies the stronger pattern emulator introduced by Chen and Tan~[ICALP'25]. This leads almost immediately to a nearly linear-time construction of pattern emulators, which, when combined with the recent fully dynamic $(1+\varepsilon)$-approximate all-pairs shortest paths algorithm of Filtser, Goranci, Patel, and Probst Gutenberg~[FOCS'24], yields our result.

0.21

On Worst-Case Guarantees for Graph-Based Nearest-Neighbor Search

cs.DS cs.LG ·2026-10-28 ·Majid Daliri, Mohammadreza Daneshvaramoli, Purna Dutta, Edo Liberty, Cameron Musco +2 more
theory
Abstract

Graph-based nearest neighbor search algorithms like HNSW, DiskANN, NSG, and others have surged in popularity in recent years due to their excellent empirical performance in comparison to space-partitioning methods like IVF data structures and Locality Sensitive Hashing (LSH). However, in contrast to LSH and related methods, obtaining strong worst-case theoretical guarantees for graph-based search has proven challenging. In this paper, we prove a strong lower bound that helps explain the lack of progress towards such guarantees. We show that no standard graph-based method (any that runs a variant of greedy search on a flat or hierarchical graph) can achieve sublinear query time for approximate nearest neighbor search in Euclidean space, even for an arbitrarily large approximation factor. On the other hand, we prove that sublinear query time is possible if we allow for Steiner points in the search graph. We obtain bounds comparable to LSH, showing e.g., that graph-based search with Steiner points can solve $c$-approximate nearest neighbor search in Euclidean space with query time $\tilde{O}(n^{1/(c-1)})$. We complement this result with initial experiments exploring how Steiner points can be incorporated into practical graph-based methods.

0.21

Destroying Rigidity in the Plane

cs.DM ·2026-10-28 ·Yu Cong, Chao Xu, Yajie Zhao
theory
Abstract

Rigidity and global rigidity are fundamental graph properties in discrete geometry, with applications to structural engineering and network localization. In two dimensions, both admit combinatorial characterizations in terms of the rigidity matroid and graph connectivity. We study the computational complexity of finding minimum-cost sets of edges or vertices whose deletion destroys either property in a finite simple graph. With unit costs, we give deterministic polynomial-time algorithms for edge deletion destroying global rigidity and vertex deletion destroying either property. We prove that edge deletion destroying rigidity is NP-hard and that, with positive integer costs, all four problems are NP-hard. These results resolve, in dimension two, the open questions on recognizing robustness under edge and vertex deletions posed by Jordán and Whiteley (2017), as well as the minimum edge-deletion problem for rigidity recorded in EGRES Open.

0.19

Optimal Encodings for Next/Previous Larger/Smaller Value Queries: A Unified Approach

cs.DS ·2026-10-28 ·Seungbum Jo, Geunho Kim, Kunihiko Sadakane, Srinivasa Rao Satti
Abstract

Given a 1D array over a total order and a query position in the array, a next larger value query returns the nearest position to the right whose value is larger than the value at the query position. Next smaller value queries and previous larger/smaller value queries are defined analogously. These queries are closely related to the $k$-th range minimum/maximum problem, which asks for the position of the $k$-th leftmost minimum/maximum value when the query range contains multiple minimum/maximum values. In this paper, we consider the encoding problem for answering any arbitrary subfamily of these four queries, where the queries must be answered without accessing the input array. We give a unified approach for obtaining space lower bounds for such encodings. We also obtain data structures for the query subfamilies that match the lower bounds up to lower-order additive terms, while supporting the queries in constant time. Our results generalize and improve several previous results on this problem [Gawrychowski and Nicholson, ICALP-2015; Jo and Satti, TCS-2016; Tsur, IPL-2019; Jo and Kim, Algorithmica-2025]. In addition, these results give tight bounds on the encoding complexity of all subfamilies of these queries for (1) permutations, (2) arrays without consecutive equal elements, and (3) general arrays.

0.19

Dimension Reduction and Coresets for Min-Sum $k$-Clustering

cs.DS cs.CG ·2026-10-28 ·Asaf Dallal, Robert Krauthgamer
Abstract

The min-sum $k$-clustering problem asks to partition an input of $n$ points into $k$ clusters so as to minimize the total pairwise distance within clusters. It is a natural alternative to center-based objectives when the data is not well captured by a small set of representative points. We study for this problem two standard sketching methods, dimension reduction and coresets, which represent the input more compactly while preserving the min-sum cost of every $k$-partition up to $1\pm\varepsilon$ factor. Our first contribution addresses dimension reduction in Euclidean space. The Johnson-Lindenstrauss (JL) Lemma readily implies that a random linear map to target dimension $t=O(\varepsilon^{-2}\log n)$ preserves, with high probability, the min-sum cost of all $k$-partitions. We show that target dimension $t=O(\varepsilon^{-2}(ddim\cdot\log(1/\varepsilon)+\log k+\log\log n))$ suffices for Euclidean inputs of doubling dimension at most $ddim$, and moreover this bound is nearly optimal. We also show that target dimension $t=O(\varepsilon^{-2}(\log(1/\varepsilon)+\log k+\logΔ))$ suffices for Euclidean inputs of aspect ratio at most $Δ$. Our second contribution is to design the first $\varepsilon$-coreset for metric min-sum $k$-clustering. It is actually a movement-based coreset, has size $(1/\varepsilon)^{O(ddim)} k\log n$ for inputs of doubling dimension at most $ddim$, and can be constructed in randomized near-linear-time. We further use these coresets to design streaming algorithms for the problem.

0.18

Local Quantum Fault Tolerance Without Expansion

cs.CC quant-ph ·2026-10-28 ·David Ponarovsky
theory
Abstract · [v2] - Fixed cross-referencing issues in the arXiv version. No changes to the results or proofs

We study fault-tolerant quantum computation using a fixed, local error-correction rule and a geometrically simple code: the cubical $(d,d')$-Toric code with $d',d-d'\geq 2$, including the 4D Toric code. Under sufficiently weak local-stochastic Pauli noise, our construction uses noiseless, bounded-depth classical processing per physical time step. Its correction cycle acts only on the current syndrome, has constant depth, and is repeated unchanged during idle logical computation; it requires neither multiscale decoder states nor scale-dependent communication delays. Our proof treats errors and syndromes as chains in a cubical complex, combining geometric shrinking under the sweep rule with a space--time explanation-tree argument inspired by Toom's theorem. It does not rely on expansion. We establish three results: 1.Explicit higher-dimensional sweep rule. We give a local algorithm for choosing sweep corrections on cubical $(d,d')$-Toric complexes, resolving the local-existence question of Kubica and Preskill. 2. Fault-tolerant computation. We prove a circuit-level threshold theorem for noisy sweep cycles interleaved with logical gates. In particular, the 4D Toric code admits a fixed, constant-depth syndrome-based correction cycle without a multiscale decoder or expansion-based encoding. We do not claim constant total-depth overhead. 3. Ideal decoding under local-stochastic noise. For an initial local-stochastic Pauli error of rate $p$ below threshold, $n^{d+1}$ noiseless sweep cycles recover the encoded state on a side-length-$n$ $(d,d')$-Toric code with probability at least $1-O(\operatorname{poly}(n))p^{Ω(n)}$, extending earlier guarantees for independent errors.

0.16

On the Complexity of Computing with p-adic Lattices

cs.SC math.NT ·2026-10-28 ·Pierre Fromonot, Mercedes Haiech, Tristan Vaccon
theory
Abstract

We study the complexity of two fundamental problems attached to p-adic lattices: computing an N-orthogonal basis, and deciding whether a given basis is N-orthogonal. When the norm is supplied through an explicit diagonal representation, both problems are polynomial. However, in the black-box model, where the ultrametric norm N is fixed and only accessible through an evaluation oracle, we prove that both problems require $Ω$(p^{v-1} ) queries in the worst case, where v $\le$ n is the largest multiplicity of a commensurability class of N and n is the dimension of the ambient vector space. We then analyse the optimality of these bounds by providing explicit algorithms. Our results answer formally an open question of Zhang, Deng and Wang and refine the recent results of Duan and Wu by providing pointwise query bounds depending on the norm's structure, in addition to uniform bounds.

0.14

Adaptive Local Measurements For Quantum Learning Can Exponentially Outperform Nonadaptive Global Measurements

cs.IT quant-ph ·2026-10-28 ·Alireza Goldar, Zhen Qin, Zhihui Zhu, Zhe-Xuan Gong, Michael B. Wakin
Abstract · 32 pages, 0 figures

Can measurements chosen adaptively, in response to earlier outcomes, be exponentially more informative than measurements fixed in advance? In many quantum settings, adaptivity helps at most polynomially; known exponential advantages require coherent measurements across copies or restricted measurements. We show an exponential advantage measuring one copy at a time, with no quantum memory kept between state copies, even when the nonadaptive learner may perform arbitrary global measurements on each copy. We construct explicit families of multi-qubit states in which a scalar parameter is encoded in one of exponentially many candidate observables, with an auxiliary marker indicating which one. An adaptive protocol identifies the relevant observable and then measures it using only single-qubit measurements and one round of adaptation. At fixed accuracy, it needs logarithmically many copies as the qubit count grows, whereas every nonadaptive protocol needs exponentially many. The same mechanism yields an exponential gap in local asymptotic minimax mean squared error constants, tight up to the exponential base. The states are separable and classical in label-dependent product bases; no entanglement or other nonclassical correlations are required. The gap comes from measurement incompatibility: the candidate observables cannot be measured jointly, so a fixed measurement must divide its information budget among them, while adaptivity selects the informative one. Accordingly, the separation disappears when the candidate observables commute, and cannot occur with fully observed classical samples. Measurement incompatibility thus emerges as a resource for quantum learning, distinct from entanglement and quantum memory, that classical feedback alone can exploit.

0.13

Upper Bounds for CSP Non-redundancy via Locally Decodable Codes

cs.DM cs.CCcs.ITmath.CO ·2026-10-28 ·Joshua Brakensiek, David Conlon, Venkatesan Guruswami, Xiaoyu He, Jiaxi Nie +2 more
Abstract · 24 pages

In the study of constraint satisfaction problems (CSPs), the non-redundancy of a relation measures the maximum number of clauses in a CSP instance which can be logically independent. In the last few years, CSP non-redundancy has found a number of applications including in query complexity, sparsification, kernelization and streaming. In this paper, we present a novel technique for upper-bounding the non-redundancy of some relations using locally decodable codes (LDCs) and locally correctable codes (LCCs). More precisely, known limitations on the structure of 2-query LDCs and LCCs can be translated into limitations on the size of certain non-redundant CSP instances. This technique builds on a technique used by Brakensiek and Guruswami [STOC 2025] to bound non-redundancy via techniques from matching vector codes. Using LDCs and LCCs, we give improved upper bounds on non-redundancy of the BCK predicate pioneered by Bessiere, Carbonnel, and Katsirelos [AAAI 2020] as well as its well-studied Boolean variant. We also give the first example of a Boolean predicate whose non-redundancy is provably not of the form $n^{k\pm o(1)}$ for some positive integer $k$.

0.12

Fully Dynamic Unconstrained Submodular Maximization

cs.DS ·2026-10-28 ·Mohammad Hajiaghayi
theory
Abstract

We study unconstrained maximization of a fixed nonnegative submodular function when the ground set changes by insertions and deletions. A uniformly random subset gives a $1/4$-approximation with constantly many value-oracle queries per update. We obtain improvements over this bound for arbitrary oblivious update sequences, without knowing future deletions. For value estimation, we give a $(\sqrt2-1-\varepsilon)$-approximation using $O_{\varepsilon,δ}(\log^2(N+1))$ expected amortized oracle queries, with failure probability at most $δ$ at each fixed time. Here $N$ bounds the number of active elements. The algorithm maintains Poisson samples of the positive mass of random chain lower bounds. We prove that the factor $\sqrt2-1$ is tight for this certificate, even when polynomially many independent orders are used. For every fixed integer $K\ge2$, we maintain a feasible $(1/4+Ω(K^{-2}))$-approximate set using $\widetilde O_K(N^{1/K})$ amortized queries. Thus every fixed exponent $γ>0$ permits a fixed improvement over $1/4$ with $\widetilde O_γ(N^γ)$ update queries. In the square-root regime, a stronger algorithm attains $ρ_{\rm sq}-\varepsilon$, where $ρ_{\rm sq}>0.3177$. The maintained sets are feasible after every update, and their approximation guarantees hold in expectation. All bounds count oracle queries; the value estimator may query previously seen inactive elements, whereas the set algorithms use only currently active elements.

0.07

Robotics, Control & Signals 67

OmniQuad: a Wheeled-Legged Hybrid Robot with Omnidirectional Wheels

cs.RO ·2026-10-28 ·Francesco Iotti, Alok Ranjan, Franco Angelini, Manolo Garabini
publishedhighly cited
Abstract

Combining wheeled and legged locomotion within a single robotic platform offers the potential to harness the advantages of both systems. The development of such hybrid systems remains an active area of research. While legged robots exhibit remarkable adaptability for traversing uneven and rugged terrains, their relatively low speed limits their practicality in indoor environments. Conversely, wheeled robots demonstrate higher efficiency on flat surfaces but often encounter difficulties when navigating obstacles such as steps. This paper introduces OmniQuad, a novel hybrid robot that integrates mecanum wheels with legged locomotion to exploit the benefits of both modalities. Mecanum wheels provide omnidirectional mobility, enhancing the robot's manoeuvrability by enabling agile movement in confined spaces and the ability to maintain a consistent orientation during task execution. In this work, we first present the design of a custom wheeled-foot featuring mecanum wheels. Subsequently, we describe the complete robot assembly, comprising these wheeled feet and EM-Act actuators. Finally, we detail the control scheme developed to govern the OmniQuad and evaluate its performance through experimental trials conducted in both indoor and outdoor environments.

0.64

From Local to Global: Sheaf-Theoretic Control Barrier Functions on Manifolds

eess.SY ·2026-10-28 ·Keith A. Currier, Wilmer Leal, Brandon C. Fallin, James Fairbanks, Warren E. Dixon
Abstract · 8 pages, 2 figures

Control barrier functions (CBFs) provide formal safety guarantees for dynamical systems by rendering a safe set forward-invariant via pointwise-optimal feedback control. While the CBF framework is well established for dynamical systems in Euclidean space, there exist gaps for systems evolving on non-Euclidean manifolds. For systems modeled as constrained control differential inclusions, we introduce the Sheaf-CBF: a sheaf-theoretic framework for CBFs on smooth manifolds and vector bundles. The differential inclusions are formulated intrinsically on the control bundle, and analogous regularity properties in the Euclidean case are established fiberwise under local trivializations. Regularity conditions are given under which a pointwise-optimal feedback law is single-valued and continuous over local charts. Framed as a local-to-global optimization problem, sheaf-theoretic restriction and gluing compatibility conditions on overlapping charts are shown to uniquely determine a continuous global section of the control bundle, thereby guaranteeing forward invariance of the safe set. Numerical simulations on a nontrivial M{ö}bius strip demonstrate that failure of sheaf compatibility results in violations of safety, whereas the Sheaf-CBF maintains forward invariance.

0.56

SKiD-SLAM: Resource-Aware Distributed Multi-Robot LiDAR SLAM for Planetary Missions with Field Deployments at Analog Sites

cs.RO ·2026-10-28 ·Hogyun Kim, Juwon Kim, Dongjin Cho, Jiwon Choi, Geonmo Yang +5 more
top venuesurveycode/data
Abstract · 31 pages. Published in IEEE Transactions on Field Robotics. Code and field data: https://sparolab.github.io/research/skid_slam/

Future planetary exploration missions require teams of robots to collaboratively map unknown environments under severe communication limits, constrained onboard computation, and no central infrastructure. In this field report, we present SKiD-SLAM, a distributed multi-robot LiDAR simultaneous localization and mapping (SLAM) system for such environments, and report its deployment across two mission concepts: cave exploration and scientific surface exploration. We survey the communication and computation budgets of actual planetary missions and select modules that fit these budgets for inter-robot place recognition, relative pose estimation, and distributed trajectory fusion. Before field deployment, we verify the assembled system on public datasets resembling planetary environments and in a simulator built on NASA JPL's Open Source Rover. We then validate SKiD-SLAM through field deployments in South Korea: a team of wheeled rovers explores the Majang lava-tube cave on Jeju Island, and quadruped robots carry out a crater-and-rock science scenario on a planetary analog terrain at the Korea Institute of Civil Engineering and Building Technology (KICT) in Yeoncheon. Across both deployments, SKiD-SLAM reliably performed distributed multi-robot LiDAR SLAM within a 250 kbps inter-robot communication budget, demonstrating its feasibility under challenging field conditions. Supplementary materials, including code and field data, are available at https://sparolab.github.io/research/skid_slam/.

0.54

Minimum Sufficient Communication for Independent Behaviour Tree Robot Teams

cs.RO ·2026-10-28 ·Sherry Jiang, Oliver Obst
top venuetheory
Abstract · 10 pages; accepted for presentation at the Australasian Conference on Robotics and Automation (ACRA 2026)

We present an experimental framework for determining sufficient communication in robot teams running fixed, independent Behaviour Trees with private blackboards. Within a declared policy family, we vary message content and admission frequency and measure traffic in recipient-weighted bits per robot decision. A dense sweep identifies candidate policies for comparison with full-state communication. We retest selected candidates and the reference on fresh, paired scenarios, using one-sided simultaneous confidence bounds to assess whether success remains within five percentage points of the full-state reference. The bounds also assess whether tested cheaper policies can be ruled out as insufficient. Depending on the evidence, the framework reports an exact minimum, a positive bracket, a sufficient upper bound, or an unresolved boundary. Experiments with eight content policies across four warehouse conditions identify zero communication as the exact minimum under low contention. In the scaled condition, the no-release policy at gate $q=0.992222$ is confirmed sufficient, achieving 96.33% success at an estimated rate of 738.55 bits per robot decision, or 0.544% of full-state traffic. The positive minimum remains unresolved. These claims apply only to the declared policy family, sampled gate values and operating conditions.

0.52

CuraEval: Evaluating Demonstration Curation on Mixed-Skill Robot Data

cs.RO cs.AI ·2026-10-28 ·Jiuyi Xu, Meida Chen, Yang Ye, Yangming Shi
code/data
Abstract · Accepted by the CoRL 2026 Workshop "Oops, I Erred". 10 pages, 2 figures, 2 tables

Robot imitation learning uses demonstrations from operators of varying skill, but selecting training data remains challenging. Existing methods score demonstrations using smoothness, diversity, or estimated influence on policy performance, but differences in selection and evaluation protocols make it difficult to determine whether they outperform simple baselines. We introduce \textbf{CuraEval}, a framework combining mathematical analysis of demonstration scores with controlled comparisons against fixed size-matched random subsets and full-data training on mixed-skill robomimic data using Diffusion Policy and BC-RNN. We show that the evaluated PSD score depends only on episode length and squared position magnitudes, and produces nearly the same ranking as episode length on \textit{square} ($ρ=0.999$). With Diffusion Policy, selecting the shortest 50\% of demonstrations improves \textit{square} success from 70.7\% under random selection to 76.5\%. Compared with random selection, our adapted geometric-entropy LOO selector reduces success at 25\% retention from 94.7\% to 74.0\% on \textit{can} and achieves 41.7\% on \textit{square}, below the mean success rates of all three independently sampled random subsets (59.8--62.5\%). On the same 3 \textit{square} Diffusion Policy training seeds, length-based selection at 75\% retention improves success over full-data training from 76.7\% to 86.7\% under the original protocol, but lowers it from 80.5\% to 79.5\% after checkpoint reporting, rollout count, and evaluation seed change jointly. No evaluated curated condition significantly outperforms full-data training after Holm correction, although confidence intervals still permit useful gains. Code is available at https://github.com/jiuyixu25/CuraEval.

0.46

xCHEM: SDR-Based Multi-Link Channel Emulator for Real-Time Hardware-in-the-Loop Digital-Twin

eess.SP ·2026-10-28 ·Ali Saeizadeh, Gabriele Gemmi, Tamerlan Aghayev, Michele Polese, Pedram Johari +1 more
Abstract · 13 pages, 16 figures, 4 tables

This paper presents xCHEM, a radio-frequency channel emulator that inserts programmable channels between physical radios using a single NI USRP X4xx software-defined radio. External channel models stream impulse responses to its field-programmable gate array without firmware changes or interruption of signal processing. A streaming datapath implements six independently configurable directional channels across four radio interfaces. Each channel uses a sparse finite impulse response filter whose memory-backed delays decouple delay span from multiplier resources. The real and complex variants support 32 and 16 nonzero taps, respectively, with a maximum delay of 16666.62 ns and 4.069 ns resolution. Coefficient scaling and arithmetic right shifts control attenuation. Hardware measurements show root-mean-square errors of 9.5 ps in delay over the full span and 0.11 dB in attenuation. Multipath delays are recovered within one sample. An FPGA coefficient engine produces tested Doppler shifts from +/-1 Hz to +/-1 MHz with mean frequency errors below 0.032 Hz relative to quantized commands after carrier-offset calibration; at 200 Hz, it reduces the measured near-carrier spur by 53.3 dB compared with 1 kHz host updates.

0.39

AtomTree: A Hierarchical Framework for State-Aware Embodied Instruction Following with LLMs

cs.RO cs.AIcs.CL ·2026-10-28 ·Haotian Zhai, Tianming Sha, Junnan Li
publishedbenchmarkcode/data
Abstract · Accepted at IROS 2026

Embodied instruction following in interactive environments requires agents to interpret natural language commands and complete long-horizon tasks. Existing planners either rely on large labeled trajectories, or use few-shot and zero-shot LLM prompting that produces a whole plan in a single serial reasoning process. These approaches depend on linguistic common sense, commit to a plan from the initial instruction, and only update after failures, which weakens robustness and state awareness. We present AtomTree, a hierarchical framework for state-aware task planning with LLMs. AtomTree incrementally builds a tree of subgoals and actions, where the LLM decides at each step, from the current symbolic state, whether to further decompose or to output a grounded low-level action. A lightweight memory module summarizes past outcomes for reuse in later decisions. When a low-level action stalls, AtomTree triggers hierarchical backtracking replanning to return to the parent node and regenerate the subtree plan from the latest state, preventing continued drift along an incorrect trajectory. In zero-shot evaluation on the ALFRED benchmark, AtomTree outperforms prior LLM-based planners. Project page: https://zhaihaotian.github.io/atomtree-website/.

0.38

LLM-PowerTrace: A Dataset of Time-Resolved GPU Power Traces for AI Compute

eess.SY cs.AIcs.OS ·2026-10-28 ·Jiansheng Wang, Jaskeerat Singh, Haochen Xu, Xinyi Li, Mingyuan Yan +4 more
datasetcode/data
Abstract · In Submission; Code and dataset are available at https://systems-power-review.github.io/training-power-trace-explorer/

AI compute is becoming large and fast-changing electrical loads. Its impacts on data center infrastructure and power grids depend not only on total energy consumption, but also on how their power draw moves over sub-second to minute timescales, including peaks, ramps, and synchronized fluctuations across accelerators. Existing energy benchmarks for large language models (LLMs) report aggregate quantities such as joules per token and rarely release the underlying time-series measurements, leaving these power dynamics poorly characterized and difficult to model. We introduce LLM-PowerTrace, an open dataset of time-resolved, per-GPU power traces for LLM fine-tuning and inference, sampled every 100\,ms on varying NVIDIA platforms paired with complete workload and system metadata. For fine-tuning, LLM-PowerTrace profiles open model families spanning a wide range of model sizes and fine-tuning methods, together with controlled sweeps over batch size, sequence length, precision, parallelism, and checkpointing. For inference, it covers variations in concurrency, GPU frequency, KV-cache and weight quantization, prompt and output lengths, and request-arrival patterns. LLM-PowerTrace profiles a set of open model families across a wide range of model sizes, and controlled sweeps over batch size, sequence length, precision, parallelism, and checkpointing for fine-tuning, and it sweeps concurrency, GPU frequency, KV cache and weight quantization, prompt and output lengths, and request arrival patterns for inference. Built upon power traces, we define power-grid-facing metrics as well as AI-compute power\&energy scaling law for extrapolating beyond the measured settings. Together, LLM-PowerTrace provides a dataset foundation for modeling, forecasting, and managing the fast-changing electrical behavior of AI compute.

0.37

Semantic Anchor Communication System for Point Cloud Transmission in Internet of Vehicles

eess.SY ·2026-10-28 ·Ruixing Ren, Junhui Zhao, Wei Zhao, Gaoxin Zhang, Xiaoming Wang
Abstract · 10 pages, 3 figures

Joint transmission of point cloud geometry and semantics under bandwidth constraints is required for cooperative perception in the Internet of Vehicles (IoV). Per-point semantics cannot be explicitly preserved by existing geometric compression methods, while interpretable three-dimensional point sets cannot be directly reconstructed via task-feature communication. The paper proposes a spatial semantic anchor communication system. Distinct from global codewords and task features, anchors with explicit coordinates are employed as transmission units. Local shape descriptors and category labels are bound to a unified spatial reference, preserving the correspondence between local position, shape and semantics. At the transmitter, anchor representations are extracted and then converted into digital frames through fixed-point quantization and lossless compression.Atthereceiver,labeledpoint clouds are reconstructed through anchor-constrained local expansion, and pose-aligned multi-view observations are fused. On the 500-frame SemanticPOSS test set, the results demonstrate that spatial semantic anchors enable interpretable joint reconstruction and fusion of geometry and semantics with low transmission overhead.

0.37

A Multi-Layer Navigation System for Smart Wheelchairs with a Limited Input Interface

cs.RO ·2026-10-28 ·Piero Simonetto, Anna Polato, Francesco Pasti, Muhammad Rameez ur Rahman, Sebastiano Vascon +4 more
published
Abstract · 8 pages, 6 figures, 3 tables. Presented at IROS 2026 (IEEE/RSJ International Conference on Intelligent Robots and Systems); to appear in the…

Smart wheelchairs are robotic devices that gather information from the surroundings and allow people to safely drive through various environments. They usually rely on the interpretation of high-level user's intentions, typically decoded through human-robot interface (HRI) or brain-machine interface (BMI). However, navigating in an unknown environment and accurately understanding user commands remain significant challenges. Misinterpretation of commands or the environment can lead to user frustration, distrust of the system, or even unsafe situations. In this paper, we present a novel Multi-Layer Navigation System (MLNS) for a smart wheelchair that enables navigation in unknown environments while interpreting and safely contextualizing user's intentions even in the case of limited and faulty commands. The stack relies on a Virtual Sensors-Driven Potential Fields (ViSPF) method for obstacle detection that accounts for the wheelchair's actual size, a Model Predictive Control (MPC) module for trajectory generation, and an open-vocabulary computer vision algorithm for target identification. Finally, a system based on Priority Queue (PQ) was implemented for target selection. We validated our system on a real smart wheelchair with experiments involving 10 participants without disabilities, including scenarios with faulty input interfaces. The results demonstrate the robustness and effectiveness of our system in real-world environments, proving its potential for practical applications in assistive technologies.

0.36

Belief-Driven Search in Connectivity-Constrained Mobile Sensor Networks with Convergence Guarantees

eess.SY ·2026-10-28 ·George J. Ziavras, Ziqiao Zhang, Shreyas Sundaram, Christopher G. Brinton
theory
Abstract · 8 pages, 6 figures, 2 tables. Extended version with full proofs

Multi-agent search and coverage is critical to autonomous exploration, search-and-rescue, and defense applications. This paper investigates settings where a mobile sensor network is tasked with coordinating to discover the position of a stationary target within a bounded region. We develop a Bayesian update scheme where agents iteratively combine local uncertainties derived from their sensor measurements through inverse variance weighting to get a better global understanding of where the target may be located given some prior belief. Network agents' trajectories are governed by a primal-dual gradient update rule that simultaneously minimizes the expected sensing uncertainty and enforces network connectivity among the agents. We prove that the belief converges to the target under mild planner conditions, with an explicit exponential rate of false-hypothesis rejection and a bound on expected search time. Simulation results show that we can successfully find the target and collapse belief to the ground truth target location while ensuring connectivity of our sensor network. They also empirically confirm our theoretical bounds and planner conditions. Finally, we compare against a naive lawnmower sweeping strategy, showing that our method locates the target more quickly when the prior is informative.

0.36

A Sub-Aperture-Based Ray-Tracing Framework for Near-Field U-MIMO THz Communications

eess.SP ·2026-10-28 ·Federico Ruggeri, Sotiris Droulias, Jiming Chen, Angeliki Alexiou, Jie Zhang
Abstract · 5 pages, 5 figures, submitted to EUCaP27

Terahertz (THz) communications are characterized by antennas with large apertures and extended near-field (NF) region, offering the opportunity to develop adaptable beam manipulation strategies, particularly related to ultra-massive MIMO (U-MIMO). While Ray-Tracing (RT) offers an attractive alternative to full-wave methods, conventional ray-based approaches are generally formulated under far-field (FF) assumptions. We propose a technique that extends RT to the NF regime, enabling ray-based NF characterization of transmitting arrays. We model the transmitting array as an antenna aperture and develop a RT framework that partitions the transmitting aperture into sub-apertures emitting FF contributions which can be propagated via geometric optics (GO). We develop a Ray-Launching algorithm based on the proposed framework and assess its beam-reconstruction accuracy against full-wave propagation. The results demonstrate that, through tunable aperture partitioning, the proposed approach can reproduce beam manipulation effects characteristic of NF U-MIMO THz communications using GO ray propagation.

0.35

Proprioceptive Finger Force Estimation for Reliable Box Manipulation

cs.RO ·2026-10-28 ·Junha Cha, Chanwoo Kim, Kisang Park, Yeoncheol Jang, Doyoung Kim +3 more
code/data
Abstract · Project page: https://finger-force-estimation.pages.dev/

Reliable box manipulation requires successful placement without damage, which depends on maintaining sufficient contact force without excessive loading. We hypothesize that per-finger force feedback improves manipulation reliability relative to open-loop finger commands. To this end, we present Finger External-Force Estimation from Proprioceptive SignaLs (FEEL), an uncertainty-aware residual method that estimates per-finger external force magnitude from proprioceptive histories. Our method combines a calibrated virtual-spring estimate with a residual distribution predicted by a Mixture Density Network (MDN) and requires no dedicated contact sensors during deployment. The estimated force is used for continuous per-finger adaptation, while predictive uncertainty is evaluated as an empirical signal of ambiguity in proprioceptive force estimation. Across ten fingers and contact forces spanning 0-20 N, our method achieves a mean RMSE of 0.251 N, reducing the error by 92.8% relative to the virtual spring and by 9.4% relative to an end-to-end MDN. In box pick-and-place, force adaptation maintains task success at 10/10 and increases both damage-free outcomes and reliability from 7/10 to 10/10. The project page is available at https://finger-force-estimation.pages.dev/

0.35

HERON: Real-Time Obstacle-Aware Reachability Maps for Mobile Manipulators

cs.RO ·2026-10-28 ·G. Dupoiron, F. Michaud, J. P. Roberge
Abstract · 19 pages, 5 figures, preprint

Mobile manipulator robots rarely reposition their base during manipulation, because doing so requires real-time knowledge of how well the arm can reach its surroundings among obstacles. Reachability maps (RMs) encode this knowledge, but existing methods are slow, coarse (typically 5 cm), or ignore real obstacles. This article asks whether a neural network can predict an obstacle-aware RM at 2 cm resolution from a live depth camera, on embedded hardware, fast enough for online use. We present HERON, a 3D U-Net trained on RMs computed with the GPU-accelerated inverse kinematics of cuRobo, first on synthetic scenes and then on real depth data, and deployed on a Jetson AGX Orin with TensorRT and INT8 quantization. HERON predicts the RM of a 2.56 m cube around the arm in 115 ms on average, for a measured system rate of 8.57 Hz, close to the 10 Hz update rate of the obstacle map, whereas computing the same map with cuRobo takes about 90 min. Against the cuRobo ground truth, 95% of the voxels lie within 1 class out of 51 with the FP32 model and within 3 classes with the INT8 model, with a Pearson correlation of 0.99 and a quadratic weighted kappa above 0.9. Accuracy degrades mainly at the workspace boundaries.

0.35

Selective Sliding Mode Control Enhances Robustness of Aerial Robots Under Disturbances

cs.RO eess.SY ·2026-10-28 ·Anoop Kiran, Thales C. Silva, Nora Ayanian
hot direction
Abstract · 8 pages, 5 tables, 8 figures

Quadrotor control under disturbances such as wind gusts and payload variations remains challenging. First-order sliding mode control (SMC) and matched/unmatched uncertainty distinction are classical approaches; however, neither tells a practitioner which loops in a cascaded controller should be made robust to sustain disturbances, how to tune those loops, nor what the additional costs are on an embedded processor. Applying robust control uniformly across all axes addresses disturbances indiscriminately, adding tuning effort on axes where no disturbance acts. On inherently underactuated vehicles, disturbances along some axes cannot be opposed directly at all. We propose a principled methodology for deciding where robust control belongs in a cascaded architecture, with all control running onboard the vehicle. We recover the acting disturbance from PID hover flights under disturbance by inverting the equations of motion, use the recovered signal to identify and quantify the excited axes, and deploy robust control where it counts. Experiments on a Crazyflie 2.1 quadrotor under slung-payload and oblique-wind disturbances show tracking improvements concentrated on the axes the rule selects, and gain sweeps confirm that the switching gain reduces error on matched axes but has no effect on an excited unmatched axis, as the analysis predicts.

0.35

Waveform Bandwidth and Frequency Diversity in FDA-MIMO Range-Angle Estimation: A CRLB Perspective

eess.SP ·2026-10-28 ·Guoqing Xia, Qu Luo, Yu Liu, Pei Xiao, Cicek Cavdar +1 more
theory
Abstract · 14 pages, 10 figures

Frequency-diverse array multiple-input multiple-output (FDA-MIMO) radar introduces range-dependent phase diversity through element-dependent carrier-frequency offsets, in addition to the delay information carried by transmitted waveforms. This paper develops a bandwidth-aware Cramer-Rao lower bound (CRLB) framework to characterize how these two mechanisms jointly affect range-angle estimation. Starting from a Nyquist-rate sampled observation model that retains waveform delays and FDA frequency offsets, we derive an equivalent Fisher information matrix (EFIM) after eliminating an unknown quasi-static complex scattering coefficient as a nuisance parameter. We derive closed-form CRLB approximations that reveal the interplay between waveform root-mean-square (RMS) bandwidth and FDA frequency diversity. A three-factor decomposition of the squared normalized cross-information further characterizes range-angle coupling. Results show that when increased independently, both waveform RMS bandwidth and FDA frequency diversity reduce the range CRLB, but FDA frequency diversity increases the angle CRLB at fixed nonzero waveform bandwidth, whereas waveform RMS bandwidth reduces it. Additionally, under a fixed aggregate spectral budget, the range CRLB exhibits an interior maximum, whereas the angle CRLB increases monotonically with the FDA spectral-allocation fraction. Numerical results corroborate the closed-form approximations and the predicted trends.

0.34

CSO: Refining Robotic Policies via Skill Distribution Alignment and Skill-Grained Optimization

cs.RO cs.LG ·2026-10-28 ·Zhiyuan Xiang, Xiang Deng, Wanjie Tao, Xu Chen, Zhao Jielun +1 more
top venue
Abstract · Accepted at NeurIPS 2026. 31 pages, including appendices and checklist

Discretizing continuous actions into discrete skills using methods like VQ-VAE has emerged as a powerful paradigm for robotic manipulation. However, quantization errors yield a suboptimal distribution for the skill prior, degrading its performance. While reinforcement learning offers a path for refinement, its direct application faces unstable encoder updates and a granularity dilemma in importance sampling. To address these challenges, we introduce Cascaded Skills Optimization (CSO), a two-stage post-training framework. First, Rejection-Sampling Supervised Fine-tuning aligns the model's observation-to-skill mapping with the distribution of successful online trajectories via supervised fine-tuning. Second, with the observation encoder frozen, Skills Policy Optimization computes an independent, clipped importance ratio for each complete skill, enabling stable and efficient updates. With a lightweight QueST-style policy, CSO raises overall LIBERO success from 80.1% to 97.5%, improves performance on RoboTwin2.0, and achieves 61.0% success across five physical single-arm tasks. Additional comparisons with FAST, discrete BEAST, and fixed-buffer IQL further validate the effectiveness of our two-stage post-training strategy.

0.34

Phase-Compensated Frequency-Scan Measurement of Harmonic Transfer Function Impedance

eess.SY ·2026-10-28 ·Yiyang Liao, Yitong Li, Jinjun Liu
Abstract

Impedance models obtained through frequency scanning provides terminal characterization for small-signal stability analysis without requiring detailed knowledge of internal converter parameters. When unbalanced steady-state components introduce cross-frequency coupling, conventional $2\times2$ impedance scan methods must be extended to identify harmonic transfer function (HTF) impedance. However, the measured phases of frequency-coupling terms depend on the starting time of the fast Fourier transform (FFT) window, potentially introducing phase errors when different time references are used. This paper presents a $dq$-domain measurement method of HTF impedance that explicitly accounts for this time dependence. Phase compensation aligns the measured Fourier coefficients to a common electrical phase reference, correcting the resulting phase discrepancies and enabling consistent impedance reconstruction. Comparisons with analytical models under balanced and unbalanced conditions verify the measurement method, recovering the conventional $2\times2$ impedance and the dominant coupling blocks of a $6\times6$ HTF representation, respectively.

0.34

Passive Identification of Unequal Phase Grid Impedances by Phase-Domain Source Cancellation

eess.SY ·2026-10-28 ·Songyang Zhang, Xun Gong, Bo Cao, Reza Iravani
Abstract

Passive grid impedance estimators that return a single resistance--inductance pair cannot represent a grid whose phase impedances are unequal. This paper identifies a separate pair for each phase from recorded point-of-common-coupling (PCC) voltage and current without injected perturbations. In each phase, the unknown grid-source voltage is cancelled by a periodic difference (PD) or by a second-order-generalized-integrator notch difference (SOGID), and moving-window integration (MWI) replaces the current derivative by an integral and an endpoint feature. The resulting two-parameter regression requires a period or notch-frequency estimate but no phase-locked-loop angle. A voltage-balance analysis gives the source residual left by frequency mismatch and bounds the parameter error by the ratio of this residual to the conditioning of the two features. In the unequal-phase interval (8.3--9.8~s) of a 10-s EMTP grid-forming record, SOGID-MWI attains phase-averaged R/L errors of 5.09/0.33\%; a shared value has a minimum error of 41.67\% for the phase ratio 1:0.5:2. On a 50-s constant-impedance grid-following record, the errors are 0.56/0.06\% over the 20--30-s synchronization disturbance. In a controlled 40-dB noise study, a noise-referenced update rule reduces the white-noise GFL error from 55.9/65.0\% to 12.3/11.5\%.

0.34

Controlling Continuum Robot Shape in Enclosed Space with Embedded Sensing using IMU and Actuator Feedback

cs.RO ·2026-10-28 ·Jiahe Wang, Sultan Haidar Ali, Eric Weissman, Jiefeng Sun
Abstract · 8 pages, 7 figures

Controlling continuum robots' shape to operate in confined, enclosed environments is challenging, since it cannot rely on external observation. Existing embedded shape sensing methods, although they can provide shape reconstruction, are insufficient to enable real-time shape control due to limitations in reconstruction accuracy, computational speed, and compatibility with shape-state representation. In this work, we present an embedded shape-sensing framework that uses readily available IMU and actuator feedback to reconstruct the continuum-robot shape with a neural network, providing accurate shape feedback at $400~\mathrm{Hz}$ for physical closed-loop control. We systematically investigate four key reconstruction design choices: shape-state representation, network architecture, training objective, and sensing inputs, through more than $5{,}000$ trained models on one- and three-segment continuum robots. The selected framework achieves global shape errors of $0.9\%L$ and $1.7\%L$ (of the robot length) for the one- and three-segment systems, respectively. In closed-loop control, one-segment tip tracking achieves a $6.1\%L$ error and three-segment shape control achieves a $2.3\%L$ global shape error, whereas IMU--PCC reconstruction using the same uncompensated measurements fails. Enclosed-space experiments further demonstrate operation without external observation. These results provide a practical framework for translating readily available embedded sensing into high-rate feedback for continuum robot closed-loop control.

0.34

Weak and Strong Data Informativity with Structural Priors for System Analysis

eess.SY ·2026-10-28 ·Yuan Zhang, Wenxuan Xu, Yun Feng, Keke Huang, Keyou You
Abstract

When measurement data are insufficient for unique system identification, data informativity evaluates whether a dataset determines a property across all systems consistent with the data. This paper introduces a unified framework distinguishing between weak and strong data informativity under prior structural knowledge described by affine constraints: data are weakly informative if at least one system consistent with the data and prior knowledge possesses the target property, and strongly informative if every consistent system satisfies it. The latter is consistent with the notion of data informativity in the existing literature. By explicitly parameterizing all consistent systems as affine functions of free parameters via the Moore-Penrose pseudoinverse, we characterize these notions for fundamental system properties-including controllability, stabilizability, and stability. This formulation establishes that verifying weak and strong informativity, when tractable, reduces to rank tests of matrix pencils constructed from the data and prior knowledge. Our findings show that structural priors can lower data-excitation requirements and enable reliable analysis even when data fail to be strongly informative. Finally, the framework is extended to a class of noisy data.

0.32

Computationally Constrained First-Order Prediction-Correction Approach for Time-Varying Convex Optimization

eess.SY ·2026-10-28 ·Negin Musavi, Oscar Jed R. Chuy, Melkior Ornik, Matthew T. Hale
Abstract · 8 pages, 2 figures, Submitted to ECC

Time-varying optimization methods often predict the locations of time-varying optima, then sample the objective function being minimized, and then compute corrections to those predictions. However, existing methods typically do not account for settings in which the number of computations is limited. Therefore, in this work we formulate a prediction-correction algorithm that is designed to obey a budget constraint on the total number of iterations it can execute over a given time horizon. We first identify a threshold on the average number of computations between samples of the objective, above which a bound from prior work certifies contraction of iterates toward a minimizer. Below this threshold, we introduce a finite-horizon contraction certificate, which we use to give budget conditions under which the tracking error admits an explicit bound that decays at a prescribed rate. We also present an algorithm to determine how many prediction and correction steps should be executed for each sample of the objective, while obeying a given budget on the allowable number of computations. Empirically, we show in a simulation that our approach reduces cumulative tracking error by 45.5 % relative to a comparable baseline approach.

0.32

Semantic-Assisted Joint Channel Estimation and Symbol Detection for MU-MIMO Systems

eess.SP ·2026-10-28 ·Hao Chang, Chentao Yue, Haiyao Yu, Yonghui Li, Branka Vucetic +1 more
Abstract

This paper presents a semantic-assisted joint channel estimation and symbol detection (Sem-JCED) receiver framework for text transmission in multi-user multiple-input multiple-output (MU-MIMO) systems. Sem-JCED integrates a linear minimum mean-square error (LMMSE) channel estimator, an expectation propagation (EP) symbol detector, and a semantic corrector based on ByT5, a byte-level language model. At each outer iteration, LMMSE estimates the channel from known pilots and data symbol feedback, and EP detects the data symbols. ByT5 then maps the data symbol probabilities to byte probabilities, which provide soft symbol statistics and semantic priors for feedback to LMMSE and EP, respectively, in the next iteration. Simulation results show that Sem-JCED requires about 7~dB lower signal-to-noise ratio (SNR) than conventional JCED to reach a bit error rate (BER) of $1\times10^{-2}$ with four users and four receive antennas, and outperforms maximum-likelihood detection with perfect channel knowledge over the simulated SNR range.

0.32

Data Center Load Reductions: Attribution, Measurement Limits, and Estimation

eess.SY ·2026-10-28 ·Abdallah Alalem Albustami, Ahmad F. Taha, Mohammad Alwahdanee, Zuhaib Shakeel Muhammad
Abstract

Hundreds of gigawatts of data center loads are seeking interconnection in the United States given the massive growth of artificial intelligence and cloud computing. In recent years, several incidents of data centers reducing their loads by hundreds to thousands of megawatts have been documented, raising grid reliability concerns. U.S. regulators have since ordered reliability standards for the integration of computational loads, and the North American Electric Reliability Corporation (NERC) now calls on transmission owners to install disturbance recorders at these facilities to support root cause analysis. Such recorders usually observe only electrical quantities at the point of interconnection (POI). A sudden load reduction can be caused by a change in workload or a switch to backup supply. This work shows when POI measurements can distinguish the two mechanisms. It is established that they can produce identical POI voltage and current in a quasi-static constant power model, and during the transient their separation depends on the reduction relative to measurement noise, the clearing time, the dynamics through which workload reaches the POI, and the recorder bandwidth. When POI data are insufficient, an estimator that uses workload data is proposed to determine how much of the facility stayed connected. The results of this work identify the measurements and grid information utilities need to determine why data center load dropped.

0.31

Self-Supervised Doppler-Guided RF Odometry

cs.RO cs.LG ·2026-10-28 ·Kunzhe Song, Jialuo Du, Jingkai Lin, Huacheng Zeng
top venue
Abstract · Accepted at NeurIPS 2026

Radio-frequency (RF) sensing enables robot ego-motion estimation in darkness, smoke, and fog, yet RF odometry remains challenging because multipath and moving reflectors make cross-frame matching highly ambiguous. Existing methods often rely on auxiliary inertial sensors or ground-truth pose supervision. We present RFWalk, a self-supervised, RF-only odometry framework that models dense correspondences across range-azimuth frames as random walks on a space-time graph and learns them from unlabeled sequences through temporal cycle consistency. To rule out correspondences that are cycle-consistent but physically implausible, we introduce a Doppler consistency loss that constrains correspondence-induced motion using radial velocity measured by the RF sensor itself. Ego-motion is then recovered through robust $\mathrm{SE}(2)$ registration that suppresses uncertain correspondences and those arising from independently moving objects. Without pose labels or auxiliary sensors, RFWalk remains competitive with supervised baselines in static scenes and reduces absolute trajectory error by 58% in dynamic scenes with moving pedestrians.

0.30

Necessary and Sufficient Structural Conditions for the Validity of Lyapunov-Type Exponential Stability Criteria

eess.SY ·2026-10-28 ·Yixiang Guo, Lin Tie, Quan Quan
theoryunusual
Abstract

Whether the exponential decay of a Lyapunov function along trajectories implies exponential stability of the state---an inference we call migration---is long assumed in Lyapunov's second method but never characterized at the criterion level. We formalize stability criteria over trajectory families, distinguish admissibility of a criterion (its structural condition suffices for its conclusion) from tightness (its necessity), and establish the necessary and sufficient structural conditions for validity. In logarithmic coordinates the theory reduces to a single two-sided displacement function of the comparison pair: in discrete time, migration holds if and only if a regularity package (excluding failure by infinitely many steps of finite displacement) and a joint lower bound (excluding failure by a single step of unbounded expansion) hold simultaneously. Continuous time strengthens the latter; the two criteria are separated exactly by the minimum step size. Written without its inverse-function formulation, the Panteley--Lor'ia criterion coincides with our continuous-time conditions on every nondecreasing pair, and loses soundness beyond: it was already complete on the classical regularity class. An envelope construction shows that regularization costs no discriminative power, and radial unboundedness, a standing assumption since Hahn, is dispensable.

0.30

SimRetract: Zero-Shot Sim-to-Real Robotic Assistance for Retractor Take-Over in Open Surgery

cs.RO ·2026-10-28 ·Xucheng Wang, Zhizhou Yang, Sung Eun Kim, Romain Hardy, Michael P. Sampson +5 more
code/data
Abstract

We demonstrate the first zero-shot sim-to-real transfer for a surgeon-robot collaboration task in open surgery: retractor take-over. A robot arm takes a retractor from the surgeon's hand and preserves retraction at handoff, using a policy trained only on simulated demonstrations, without human demonstrations or real-world fine-tuning. Imitation learning from real demonstrations is limited by collection and labeling costs, and by performance degradation when the environment changes. Our framework, SimRetract, generates simulated demonstrations designed for transfer: the policy observes masked instrument images and grasp geometry, learns to recover from off-nominal states, and acts with margin for execution drift. Deployed zero-shot, it grasps the retractor in 90% of 50 surgeon-defined phantom presentations and preserves retraction at handoff in 78%. Results on unseen retractors suggest potential for generalization to instruments with similar grasp geometry. Across ten policy configurations and 1,200 simulated and real trials, we localize the sim-to-real gap to components: in simulation, grasp rates across ablations span 80-96%, while on the real robot removing the instrument mask alone drops performance to 22% grasped and 12% held. Transfer-oriented design thus makes zero-shot sim-to-real transfer feasible for a surgical assistance task without human demonstrations. Code is available here: https://anonymous.4open.science/r/simretract-58D2.

0.29

Safety of layered control systems via control barrier morphisms

eess.SY math.CTmath.OC ·2026-10-28 ·Aaron D. Ames, Joe Moeller, Massimiliano de Sa, Paulo Tabuada
theory
Abstract

Layered control architectures provide a powerful paradigm for synthesizing controllers for complex systems. Yet, translating guarantees at each layer to system-level guarantees is challenging due to the interactions between layers. To address this problem, this paper presents a categorical perspective on layered control architectures. Individual systems are represented as control coalgebras, their interactions are specified as morphisms to mediators, and system-level behaviors are obtained via composition defined categorically as a pullback. To analyze the safety of these systems, we formulate control barrier morphisms (CBMs), the categorical generalization of control barrier functions. We prove that feedback controllers satisfying a CBM condition render the associated safe subobject forward invariant. It is through CBMs that we can establish the safety of layered systems: given a CBM at each layer subject to compatibility conditions, the universal property of the pullback implies the existence of a CBM for the composed system. Safety of a full-order model being safety filtered by a reduced-order model is recovered as a specialization of the categorical framework.

0.29

Physics-Guided Inferencer-in-the-Loop Digital-Twin for Real-Time Adaptive Emulation of Nonlinear PMSMs

eess.SY ·2026-10-28 ·Songyang Zhang, Yuzhong Zhang, Venkata Dinavahi
Abstract

A real-time digital-twin (RTDT) of a permanent magnet synchronous machine (PMSM) allows drive controllers to be tested in hardware-in-the-loop (HIL) before they meet the real machine. Its machine model rests on identified parameters or flux look-up tables (LUTs) from finite-element analysis (FEA) or characterization tests; the maps are regenerated offline when the machine changes, and online estimators track a few parameters such as resistance and PM flux. This article makes the executed model learnable from reference data. A physics-guided machine learning (PGML) network advances the currents and predicts torque in one step, trained with analytical one-step labels. In the proposed Inferencer-in-the-Loop (IIL) architecture, a GPU retrains the network while a field-programmable gate array (FPGA) executes the last accepted weights. For damper circuits, shifted parameters, and an FEA-calibrated saturated machine, one kernel form reaches average normalized mean absolute errors (NMAEs) of 0.19%-0.67% at measured FPGA kernel latencies of 0.30-0.34 $μ$s, and retraining for a shifted machine nears its final loss within 94 s. On a magnetically linear 200-W drive, a residual form corrects the analytical step and learns its physical parameters from operating data; each step takes under 2.5 $μ$s, 20 times faster than real time and ahead of the physical drive's measurements even at 20-kHz sampling. On six held-out runs (1193 s), its $i_q$ error is 5.9%-9.4% of measured RMS, against 10.4%-13.2% for an independently fitted grey-box model, and over 30 min of loaded heating its online-updated heads track a 4.1% winding-resistance rise with an RMS error of 0.44 m$Ω$.

0.29

Beyond Fresnel Approximation: Exact Airy Beamforming for Near-Field THz Communications

eess.SP ·2026-10-28 ·Fangjie Shen, Weidong Mei, Aoran Liu, Peilan Wang, Zhi Chen
Abstract

Line-of-sight (LoS) blockage can severely degrade the performance of conventional focused beams in terahertz (THz) communications. Recently, bending beams (e.g., Airy beams) have attracted increasing attention due to their ability to bypass obstacles by propagating along curved trajectories under near-field propagation conditions. However, existing Airy-beam designs in wireless communications often rely on the Fresnel approximation for the focal distance, which becomes inaccurate at large steering angles or short propagation distances. To address this limitation, this letter proposes a more accurate Airy beamforming design method based on the exact focal-distance expression and the corresponding caustic geometry. Specifically, we jointly optimize the bending curvature, virtual focal distance, and steering angle to maximize the received signal power subject to receiver-alignment and obstacle-avoidance constraints. To tackle the resulting non-convex problem, we first use geometric preprocessing to narrow the feasible range of the bending curvature. Then, a two-layer optimization framework is developed with the outer and inner layers optimizing the bending curvature and the other two parameters, respectively. Numerical results based on the cascaded geometrical and wave channel model demonstrate that the proposed method achieves higher received signal power than conventional Airy-beam designs and offers complementary advantages over conventional focused beams under different blockage ratios.

0.29

pCoRev: Physics-Informed Contrastive Learning for Unsupervised Reverberation Artifact Suppression in Ultrafast Plane-Wave Imaging

eess.SP ·2026-10-28 ·Hyunwoo Cho, Yeeun Kwon, Yangmo Yoo, Jinbum Kang
Abstract

Reverberation artifact suppression in ultrafast plane-wave (PW) imaging remains challenging because artifact-free reference data are generally unavailable and multipath echoes often resemble tissue signals. Coherent plane-wave compounding (CPWC) attenuates reverberation only partially and sacrifices the frame rate that motivates PW imaging. We propose pCoRev, an unsupervised physics-informed contrastive learning framework that leverages the distinct angular coherence characteristics of tissue echoes and reverberation artifacts. A single-angle I/Q frame is the network input, and the remaining angles of the same CPWC acquisition provide angle-consistent positive pairs, residual-derived negative pairs, and coherence-guided self-supervision, so no paired artifact-free data are required. On a fabricated reverberation phantom with 31 validation acquisitions, pCoRev achieved the best contrast ratio (CR), contrast-to-noise ratio (CNR), and generalized CNR (gCNR) of 38.76 +/- 5.40 dB, 11.75 +/- 1.81 dB, and 0.932 +/- 0.051 against delay-and-sum (DAS), delay-multiply-and-sum, a supervised 3D convolutional neural network, and unsupervised deep coherence learning. Because the network is trained on single-angle inputs, it also operates on a single PW transmission: with one angle, pCoRev reached a CR of 32.10 +/- 7.79 dB and a CNR of 10.39 +/- 3.02 dB, exceeding 75-angle compounded DAS (20.45 +/- 3.84 dB and 8.38 +/- 2.00 dB) with 75-fold fewer transmissions. On four clinical acquisitions from other scanners, pCoRev improved CR over DAS by 9 to 21 dB without fine-tuning. These results indicate that physics-informed unsupervised contrastive learning suppresses reverberation effectively while preserving the frame-rate advantage of ultrafast PW imaging.

0.28

Adaptive Orbit-Aided Vector Tracking for Orbit Determination of LEO Satellites

eess.SP eess.SY ·2026-10-28 ·Zihong Zhou, Bing Xu
Abstract

Reliable orbit determination for CubeSats is essential for emerging low-Earth-orbit positioning, navigation, and timing (LEO-PNT) applications. However, onboard orbit determination remains constrained by single-frequency GNSS code-based receivers, intermittent signal reception, and highly dynamic orbital conditions. This paper proposes an adaptive orbit-aided vector tracking (OAVT) algorithm for LEO-PNT CubeSat orbit determination under these conditions. The method employs a GNSS vector tracking loop (VTL) as the primary navigation estimator and loosely couples it with a low-fidelity orbital dynamics model (ODM), which provides auxiliary position and velocity pseudo-measurements. The contribution of the ODM is regulated through adaptive covariance assignment based on GNSS stress and the residual consistency between the ODM and GNSS solutions. Under nominal conditions, OAVT preserves the accuracy advantage of VTL over the conventional scalar tracking loop. The 3D position root-mean-square error (RMSE) is reduced from 2.631\,m for STL to 1.081\,m, while the 3D velocity RMSE decreases from 0.3985\,m/s to 0.1023\,m/s. Under degraded conditions, OAVT provides substantially greater robustness than both STL and standalone VTL. The 3D position RMSE remains bounded at 8.27\,m during a complete signal outage, 10.17\,m during gradual signal loss and reacquisition, and 26.65\,m under variable visibility, whereas STL fails to produce valid solutions and standalone VTL loses lock. These results demonstrate that OAVT substantially improves solution robustness and state continuity with only a moderate increase in computational burden, making it suitable for CubeSat orbit determination and future LEO-PNT missions.

0.28

Finite-State Uncertainty-Aware Widely-Linear Equalization for Aeronautical GMSK Links

eess.SP ·2026-10-28 ·Ahmad Tanha, Ali Khalesi
Abstract

Aeronautical and unmanned aerial vehicle (UAV) receivers must equalize doubly selective channels under synchronization uncertainty and limited onboard computation and memory. We propose a finite-state, uncertainty-aware widely-linear (WL) equalization algorithm for Gaussian minimum-shift keying (GMSK) links. The algorithm uses synchronization estimates and their reliability descriptors to select one of a small number of precomputed filters, avoiding repeated online coefficient synthesis. Using a finite-memory Laurent model, we design the stored affine WL filters from conditional physical-state moments. We derive an exact characterization of excess mean-square error (MSE) relative to adaptation to continuous side information, Lloyd-type conditions for finite-state design, local timing--frequency mismatch certificates, and an orthogonal decomposition of design and information losses. On a reference channel at $E_b/N_0=8$~dB under high synchronization-estimation uncertainty, a $16$-state bank reduces pseudo-symbol MSE by $51.9\%$ relative to point-estimate WL time-varying MMSE equalization and by $6.6\%$ relative to posterior-mean (PM) substitution. With identical partitioning, routing, and storage, conditional-moment filters reduce MSE by $10.1\%$ and full-waveform average bit error rate (ABER) by $15.6\%$ relative to representative-state filters. Across $18$ synthetic three-path channels with individually fitted banks, MSE and ABER decrease by $5.3\%$ and $12.9\%$, respectively, relative to PM substitution. The $16$-state receiver stores $2048$ real coefficients. Posterior integration, robustness sweeps, and software timing measurements quantify the tradeoff between equalization accuracy, online computational cost, and bank refresh.

0.27

ELEVATE: Agentic Controller Acquisition from Task Failure for Quadrupedal Manipulation

cs.RO ·2026-10-28 ·Shiqi Liu, Radu Corcodel, Ding Zhao
code/data
Abstract · 23 pages, 14 figures, including appendix. Project page and videos: https://agentic-quad-elevate.github.io/public/

Quadrupedal manipulators can extend their workspace through whole-body interaction, but task-planning agents' capabilities remain limited by their available controllers. We present ELEVATE, a framework that uses persistent failures during task-program refinement to identify and acquire missing whole-body control capabilities. When these failures suggest a missing physical capability, a coding agent formulates and iteratively refines a reinforcement-learning problem, specifying behavior-specific rewards, curricula, and termination conditions. Validated controllers are added to the capability library and composed with existing controllers to retry the task. Starting from a single pretrained loco-manipulation controller, ELEVATE autonomously acquires controllers for wall-supported elevation, supported manipulation, and wall descent across a sequence of seven simulation tasks. These controllers are reused in subsequent reaching and object-manipulation tasks. On the five tasks beyond the initial controller's workspace, ELEVATE achieves mean success rates of 98.4%-100% across five independent runs, while coding-agent baselines with a fixed controller library fail to complete these tasks. The synthesized task programs, together with the acquired controllers, also transfer to a physical quadrupedal manipulator without real-robot training or fine-tuning, achieving 18/20 successes in rack retrieval and 16/20 in floor-to-rack placement. Videos and demonstrations are available at https://agentic-quad-elevate.github.io/public/.

0.27

Stable Dynamical Mass Transportation via port-Hamiltonian Structure

eess.SY math.OC ·2026-10-28 ·Dongjun Wu, Yankai Lin, Gang Feng
Abstract

Dynamical optimal mass transportation over nonlinear systems is often hindered by the curse of dimensionality and open-loop implementation. We propose an alternative framework that sacrifices exact optimality in favor of stability, robustness, and scalability. We use a stochastic port-Hamiltonian system as a reference model. Under suitable structural assumptions, we show that its density converges asymptotically to a prescribed steady-state distribution. We then shape a deterministic nonlinear system into a compatible port-Hamiltonian form and add a local density-feedback term so that its closed-loop continuity equation matches the Fokker-Planck-Kolmogorov equation of the reference model. The resulting feedback drives a broad class of initial distributions to the target distribution. Because the density term is local at the particle level, the controller admits decentralized implementation and avoids solving a global finite-horizon transport problem online. Nonlinear examples and simulations illustrate the method. For linear systems with Gaussian targets, we derive necessary and sufficient conditions for stabilizing density feedback and give explicit controller constructions.

0.27

Metropolized backward sampling for conditional sequential Monte Carlo

eess.SP stat.COmath.PRstat.ME+1 ·2026-10-28 ·Adrien Corenflos
theory
Abstract · 12 pages + appendix. A Lean formalization was AI-generated after the draft of the paper was written as an experiment. I have subsequently ed…

Particle Gibbs samplers with backward sampling are widely used for sampling the smoothing distribution of state-space models, but, with few particles, the new trajectory often coincides with the current one at many time steps. A known but rarely used remedy replaces the random selection of the final particle by a Metropolis--Hastings move which avoids the current particle. We show that, with backward sampling, this remedy only helps at the last time steps, because its effect on the update rates decays geometrically with the distance to the final time. We propose instead Metropolized backward sampling, which makes such a move at every step of the backward pass, forcibly switching trajectories whenever possible. Additionally, we introduce a coupled version which prevents proposing the current trajectory, maximizing the chances of an actual switch versus the current trajectory and performing perfect sampling when the state-space model has no observations, a feature not shared by backward sampling. We prove that, at every time step, both methods dominate backward sampling as they update the state more often and have smaller lag-one autocorrelations than it does. Our methods are drop-in replacements for backward sampling, and can be used to improve any particle Gibbs sampler.

0.27

Machine learning-guided real-time AC optimal transmission switching

eess.SY ·2026-10-28 ·Weimin Huang, Weiwei Yang, Bistra Dilkina, Konstantina Mellou, Ishai Menache +3 more
benchmark
Abstract

The operation of power grids has substantial economic consequences, affecting generation costs and, ultimately, electricity prices. The Optimal Transmission Switching (OTS) problem seeks to reduce these costs by selectively switching off transmission lines. However, OTS gives rise to a challenging mixed-integer nonlinear optimization problem that is difficult to solve at operationally relevant timescales, particularly as rapidly changing loads create a need for real-time switching decisions. To address this challenge, we develop a machine learning-guided method that identifies high quality line switching decisions in real time. Specifically, we construct a grid-specific switching-value model that rapidly estimates the generation-cost reduction associated with opening transmission lines by fine-tuning a pretrained model for AC optimal power flow (AC-OPF). The switching-value model provides signals to guide promising switching decisions, and topologies proposed by the model are evaluated using exact AC-OPF solvers to verify feasibility and obtain the actual generation cost. We develop multiple strategies that use the ML surrogate to efficiently explore the combinatorial space of transmission topologies, including Fixed-Rank Search, Adaptive-Rank Search, and Swap-Enhanced Search. Across multiple benchmark networks, our proposed algorithms identify line switching decisions that significantly reduce generation cost within a 50s computational budget. In particular, under a realistic data center buildout scenario on the 2,000-bus synthetic Texas grid, our approach reduces generation costs by 31.97% and gross nodal load payments by 94.34%. More broadly, our results demonstrate that a fine-tuned surrogate can effectively guide search even without guaranteeing feasibility, enabling classical optimization methods to tackle large nonlinear combinatorial problems on operationally relevant timescales.

0.26

Quantize by Doing: Task-Guided Mixed-Precision Quantization of Vision-Language-Action Models

cs.RO ·2026-10-28 ·Jiuyi Xu, Xiao Hu, Meida Chen, Song Wang, Yang Ye +1 more
code/data
Abstract · 13 pages, 3 figures, 8 tables

Post-training quantization (PTQ) reduces the memory footprint of vision-language-action (VLA) models, but lowering bit widths while maintaining task performance remains challenging. Existing work has used sensitivity estimates from calibration data to assign bit widths to different model components. However, using task success directly to guide mixed-precision allocation is underexplored because closed-loop evaluations are costly and the effect of quantizing one component depends on the precision of other components. To this end, we propose \textbf{QbD-VLA}, a task-guided mixed-precision quantization method that uses task success from short closed-loop evaluations to assign weight and activation bit widths while accounting for interactions between model components. Experiments covering four VLA models and all four LIBERO suites show strong task performance at average precisions of approximately 4 bits for weights and fewer than 6 bits for activations. On SIMPLER, the gain over uniform quantization is small. On a Franka Research 3 robot arm, simulation-selected allocations show no detected success-rate loss on X-VLA but fail on $π_{0.5}$. These findings support using task success to guide mixed-precision VLA quantization, while highlighting the need for real-world evaluation before deployment. Code is available at \href{https://github.com/jiuyixu25/QbD-VLA}{\texttt{github.com/jiuyixu25/QbD-VLA}}.

0.26

Movable Beyond-Diagonal RIS for ISAC: Optimal Placement and Pareto Boundary

eess.SP ·2026-10-28 ·Chandan Kumar Sheemar, Sourabh Solanki, Symeon Chatzinotas
Abstract

This letter investigates integrated sensing and communication using a movable beyond-diagonal reconfigurable intelligent surface (BD-RIS). Under a rank-one line-of-sight BS--RIS channel, we derive the exact communication--sensing Pareto boundary of a fully-connected BD-RIS in terms of user--target channel correlation. For line-of-sight RIS links, maximizing this correlation yields the optimal placement. We characterize the minimum track aperture that eliminates the illumination trade-off and show that a movable diagonal RIS then attains the same optimum. Below this threshold, a dynamic program offers certified placement accuracy. Numerical results examine hardware connectivity, angular mismatch, and Rician fading.

0.26

A Parallel Two-Stage Prony Framework for Robust Delay-Doppler Estimation

eess.SP ·2026-10-28 ·Liangchen Sun, Yutaka Jitsumatsu
Abstract

This paper addresses off-grid physical delay-Doppler parameter estimation based on an orthogonal time frequency space (OTFS) pilot waveform over doubly selective channels. Based on a known single-pilot waveform with a frequency-shifted sinc pulse structure, we propose a parallel dual-order estimation framework, in which Prony-based Doppler-first and delay-first procedures are executed concurrently and their candidate paths are jointly fused. The two estimation orders provide complementary resolution characteristics, thereby mitigating ambiguities caused by closely spaced multipath components and fractional delay--Doppler mismatch. Numerical results show that the proposed framework improves detection reliability and estimation stability compared with a single-order Prony estimator, especially in off-grid scenarios and low-to-moderate signal-to-noise ratio regimes.

0.26

Minimum-Time Trajectory for Escort-Assisted Evasion from a Sensing-Limited Pursuer

eess.SY ·2026-10-28 ·Al Faysal, Alexander Von Moll, Artur Wolek, Satyanarayana G. Manyam, Thomas Chapman +3 more
Abstract

This paper studies a minimum-time path-planning problem in which an escort assists a leader by triggering and redirecting a sensing-limited pursuer. For the post-engagement case with equal leader, escort, and pursuer speeds, we show that the minimum-time leader trajectory has a three-phase structure: an initial straight segment, a moving-boundary segment, and a final straight segment to the goal. The boundary-following kinematics and the corresponding entry and exit conditions are derived and used to construct the leader trajectory for a fixed escort engagement direction. The engagement direction is then optimized over the feasible set to obtain the minimum-time solution. Numerical results illustrate the resulting trajectory and the benefits of escort assistance.

0.26

Efficient MPC for LC-Filtered Three-Level NPC Inverters in Vehicle-to-Load Applications

eess.SY ·2026-10-28 ·Rock Agon, Dieudonne Iradukunda
Abstract

Finite control-set model predictive control (FCS-MPC) offers fast, modulator-free voltage regulation but is sensitive to the fidelity of its internal model. This sensitivity is usually countered by algorithmic augmentation, disturbance observers, adaptive weighting, or learning-based estimators, at the cost of extra tuning and computation. This paper instead improves fidelity at the source. For a three-level neutral-point-clamped (3L-NPC) inverter with an LC output filter in vehicle-to-load operation, a coupled alpha-beta-frame predictive model carries the capacitor voltage and filter current jointly through the state vector, so that the second-order LC coupling is retained without enlarging the 27-candidate search. The cost function additionally regulates dc-link neutral-point balance and switching effort. One-step prediction accuracy, parametric sensitivity, and computational cost are bounded in closed form. Simulation of a V2L system yields a voltage total harmonic distortion of 0.72%, a steady-state tracking error of 1.6%, and transient recovery within one fundamental cycle, meeting the applicable IEEE power-quality limits with margin, without observers, parameter estimation, or learning-based augmentation.

0.25

Covariance Learning for Direction-of-Arrival Estimation with Sparse Linear Arrays

eess.SP ·2026-10-28 ·Majdoddin Esfandiari, Esa Ollila, Geert Leus, Daniel P. Palomar
Abstract

Covariance learning (CL) estimates source powers and sensor-noise variance by minimizing the Type-II (marginal) negative log-likelihood of the array measurements. We develop both grid-based and gridless CL direction-of-arrival (DOA) estimators applicable to arbitrary sparse linear array (SLA) geometries. The grid-based DOA estimator minimizes the Type-II objective on an angular grid via successive convex approximation (SCA) and estimates the noise variance by a scalar bisection search. The gridless DOA estimator then refines each grid-based DOA over the continuum using a Newton method. Simulations on a six-sensor nested array, including an underdetermined case with nine sources, show that the proposed DOA estimators outperform state-of-the-art methods, despite jointly estimating the sensor noise variance.

0.25

Compositional Diffusion Planning with Chunk Selection and Reachability Measure

cs.RO ·2026-10-28 ·Chanyoung Park, Youngjoong Kim, Jaesik Park
Abstract

Compositional diffusion planners have demonstrated remarkable success on long-horizon diffusion planning by sampling short trajectories, namely chunks, and composing them into a long trajectory. However, averaging overlapping chunks can induce the mode-averaging problem, resulting in discontinuous trajectories. Several works aim to address this problem, whereas a non-negligible portion of plans remains infeasible. In this work, we propose Selective Compositional Diffuser (SCD), handling this problem by measuring reachability. To this end, we present a self-supervised learning method for the reachability classifier. This is trained on offline data and generated chunks jointly to address potential distributional mismatch, while reducing the bias inherited from the generated data. Based on this, we construct a graph whose edge costs represent whether consecutive chunks are reachable, and traverse the graph to find effective yet efficient routes. Lastly, we generate bridging trajectories and refine the composed trajectory for smooth transitions. We evaluate our method on OGBench stitching tasks. It practically reduces incompatible mode compositions and the mode-averaging problem compared to the studied baselines. Across tasks, SCD improves success rates, reaching state-of-the-art results on HumanoidMaze-Large and Giant tasks.

0.24

Recursive Self-Improvement in Physical AI: Mechanisms, Evidence, and Evaluation

cs.RO ·2026-10-28 ·Xiaofeng Han, Kaize Luo, Zenghuang Fu, Zichong Zhu, Zhe Feng +4 more
surveycode/data
Abstract

Embodied systems can update policies, world models, skills, and executable programs from deployment experience, allowing one interaction to benefit later tasks. When a system also revises and reuses its data-acquisition, feedback, or learning procedures, experience changes the process that produces capability, introducing recursive self-improvement. This paper surveys research in physical AI through the connection between persistent capability and its improvement process. We first explain the functional distinction between learner updates and improvers, then compare policy learning, imagined experience, memory and skill accumulation, and executable harnesses. These approaches connect to revisions of curricula, rewards, and research programs in different ways. Contact validity, recovery, retention, and resource costs determine whether the resulting gains accumulate during continued deployment. An L0-L3 rubric distinguishes task-local adaptation, persistent capability gain, inherited process revision, and verified improver gain. Representative benchmarks and within-study contrasts relate these distinctions to existing results. We further propose common-data and interactive comparisons that evaluate reference and revised improvers from shared learner starts, separating the use of experience from its acquisition. This perspective connects the diverse methods, physical operating conditions, and evaluation evidence, underpinning embodied RSI. A comprehensive list of studies in this survey is available at https://github.com/nkd-lkz/physical-ai-rsi-survey.

0.23

Embodied at Sea: Imitation Learning for Multi-Task Underwater Manipulation

cs.RO cs.LG ·2026-10-28 ·Greta Perez-Haiek, Luca Macesanu, Kevin Angers, Mohammed Imam, Timothy Gambin +1 more
code/data
Abstract · 8 pages, 8 figures. For associated files, see https://lair-research-nyu.github.io/Embodied-at-Sea/

Behavior Cloning has demonstrated promising generalization capabilities for robotic manipulation in terrestrial environments. However, marine visual distortions and dynamic currents create a severe domain shift that remains a key bottleneck in adapting generalist architectures to underwater field manipulation. To address this, 1616 episodes of underwater manipulation data were collected in calm and turbulent waters and were used to train 14 generalist Vision-Language-Action (VLA) models (DynamicVLA) with varying proprioception observation configurations. Across three tasks on 1260 gradable on-hardware rollouts in a pool, 10 models outperformed the baseline, with the highest-performing model achieving a 93.3% overall success rate, including partial successes. Results are also introduced from a case study evaluating the performance of a single-task Action Chunking Transformer (ACT) and a generalist VLA (SmolVLA) policy for two tasks in marine archaeology and conservation: grasping and mooring. Experiments are performed on an ROV on and around the WWII wreck Coralita X127 and at Manoel Island Marina, Malta. Across 748 collected episodes and 508 gradable on-hardware rollouts, the ACT policy substantially outperformed SmolVLA (47.9% vs. 7.9% pooled success rate), earning a 66.7% success rate on the most successful task at the marine archaeological site. To enable further contributions to embodied intelligence in marine field robotics, the BLUEMBODIED KIT used to collect data, train models, and deploy policies on the ROV has been open sourced.

0.22

Channel State Information-Free Angle Spoofing for Differential Localization Privacy

eess.SP ·2026-10-28 ·Lei Lyu, Urbashi Mitra
Abstract

This paper proposes a channel state information (CSI)-free angle spoofing strategy that remains effective even when the eavesdropper (Eve) uses differential statistics to estimate the location of the transmitter (Alice). Specifically, by designing a precoding matrix, Eve incorrectly estimates the angle differences of departure (ADoDs), thereby spoofing Alice's location. Meanwhile, the trusted user (Bob) can still infer Alice's true location with a modest amount of securely shared side information. Given the channel parameters estimated by Eve from the spoofed channel, the resulting localization bias introduced to the eavesdroppers under the spoofed geometric is analyzed. Eve's bias can be driven to infinity asymptotically in a key parameter of the precoder. Simulation results demonstrate that the proposed method introduces significant localization errors for Eve. When combined with a prior CSI-free spoofing strategy, further degraded performance is achieved by Eve.

0.22

Skeleton-Guided Geometry-Aware Auto-Labeling for Robotic Grasping: From 4-DoF Learning to 6-DoF Execution

cs.RO ·2026-10-28 ·Ali Sabzejou, Farbod Azimmohseni, Mehdi Tale Masouleh, Ahmad Kalhor
published
Abstract · Published in The International Journal of Robotics Research (IJRR), OnlineFirst, September 2026

Robotic grasping relies on large manually annotated datasets, which are costly to create. We introduce a straight-skeleton-based algorithm for generating two-finger grasp keypoints and identifying regions suitable for stable, feasible grasps. Our automated pipeline produces consistent grasp annotations without human intervention. We also propose the Skeleton-based Generative Grasp Convolutional Neural Network (SkelGG-CNN), a 4-DoF architecture that learns geometric grasp features from skeleton-based guidance. To extend auto-labeling to 6-DoF grasping, we develop a generalized quasi-3D skeletonization algorithm that operates directly on point clouds. Using automatically generated 4-DoF labels, SkelGG-CNN achieves accuracies of 95.42%, 96.55%, and 97.1% on Jacquard V1, Jacquard V2, and Cornell, respectively. On the GraspNet novel split, 6-DoF auto-labeling achieves an average precision of 18.33, comparable to state-of-the-art model-based approaches. In simulation, the 6-DoF method achieves success rates of 98.4% on Dex-Net and 89.1% on EGAD!. Real-world experiments with a Delta parallel robot show success rates of 97.4% on single household objects and 95.0% in cluttered scenes, outperforming state-of-the-art methods without domain adaptation. The 4-DoF pipeline adapts across environments and produces annotations approaching human-level precision, enabling efficient training of lightweight models on small datasets. The generalized 6-DoF method provides scalable, geometry-aware annotations for cluttered 3D scenes.

0.22

Explore Where It Matters: Targeted Exploration for Efficient Rover Adaptation

cs.RO cs.AIcs.LG ·2026-10-28 ·Jacob Levy, Arthur Bouton, Tristan D. Hassler, Hari Nayar, David Fridovich-Keil
Abstract · Conference on Robot Learning 2026

Future planetary rovers will traverse steep, deformable terrain where slip, sinkage, and terrain deformation can threaten mobility. Although simulation can capture some wheel-soil effects, unmodeled deformable-terrain phenomena create sim-to-real gaps requiring deployment-time adaptation. Yet, rover operations leave little room for extensive trial-and-error. We introduce TERRA (Targeted Exploration for Reward-Relevant Adaptation), an online adaptation framework that efficiently seeks deployment data exposing task-relevant model errors and corrects them. TERRA adapts a simulation-trained world model with a reward-weighted prediction loss and trains a loss predictor to estimate where future performance-relevant errors are likely to occur, using this signal to guide exploration during MPC. In simulation, TERRA reaches approximately 90% slope-climbing success within five trials under severe out-of-distribution pretraining conditions, outperforming exploration baselines in reward, completion time, and energy to first success. On the ERNEST rover, TERRA recovers coordinated active-suspension locomotion on a deformable Mars-simulant slope within three to five real-world trials. Project website: https://terra-explore.github.io

0.21

Force-Driven Hook Disassembly via Constraint-Aware Object-Centric Reinforcement Learning

cs.RO ·2026-10-28 ·Nadeen Elhadad, Takuya Kiyokawa, Kensuke Harada
Abstract · 14 pages, 6 figures

The growing volume of end-of-life products has intensified the need for automated disassembly, yet hook-type connections remain challengin: their orientations vary, they deform under load, and their geometry is hidden, so vision alone cannot identify a feasible non-destructive disassembly direction. This work proposes a constraint-aware reinforcement learning framework that infers feasible extraction directions online from contact forces. The policy is trained in a robot-agnostic, object-centric simulation on a simplified peg-in-hole pair and deployed zero-shot on a physical robot, where it transfers to real connections with unseen, multi-stage constraint structures. Because instantaneous contact forces are noisy and locally ambiguous, the policy is coupled with a particle filter belief that provides the policy with temporally integrated directional information and reduces redundant exploratory motion. The approach is validated on four real-world tasks, from peg-in-hole pairs to the inner propeller and front cover of an end-of-life air-conditioning unit. The full method achieves success rates of 100%, 100%, 90%, and 75% with the highest path efficiency on every task, and raises the success rate on the air-conditioning cover from 45% to 75% relative to a policy trained without the belief. These results indicate that force-driven reinforcement learning can enable disassembly without product-specific models or demonstrations.

0.21

Two-Layer Microwave Linear Analog Computer (MiLAC)-Aided Multi-User MIMO Networks

eess.SP ·2026-10-28 ·Xiaohua Zhou, Tianyu Fang, Matteo Nerini, Yijie Mao, Bruno Clerckx
theory
Abstract

Microwave linear analog computer (MiLAC)-aided beamforming, which performs signal processing directly in the analog domain, has recently emerged as a promising alternative to fully digital and hybrid beamforming for multi-antenna systems. However, existing studies have shown that a single lossless and reciprocal MiLAC at the transmitter cannot, in general, attain the performance of fully digital beamforming in multi-user scenarios. To overcome this limitation, we propose a two-layer MiLAC-aided transmit beamforming architecture for downlink multi-user multiple-input multiple-output (MIMO) networks. Leveraging multiport microwave network theory, we first establish a general physics-compliant model of the proposed architecture, and then derive a simplified model under the assumptions of unilateral and perfectly matched amplifiers. We prove that, with lossless and reciprocal MiLACs, the two-layer MiLAC-aided transmitter can realize any fully digital transmit beamformer, while a single-layer MiLAC-aided receiver can attain the same achievable rate as a fully digital linear combiner. We then derive closed-form mappings from fully digital transmit beamformers and receive combiners to their corresponding MiLAC-aided analog implementations. Based on these equivalence results, we formulate a sum-rate maximization problem for the considered multi-user MIMO network and develop an optimization framework for jointly designing the power allocation and scattering matrices of the MiLAC-aided transceivers. Numerical results validate the theoretical analysis and show that the proposed two-layer MiLAC-aided design attains the same sum-rate performance as its fully digital counterpart.

0.21

Inertial Steering and Motion Prediction of Tracked Vehicles on Low-Adherence Surfaces

cs.RO eess.SY ·2026-10-28 ·Mathis Turgeon-Roy, Mohamed Ounally, Veronica Vannini, Olivier Roy, François Pomerleau
Abstract

Tracked vehicles are widely used in off-road applications, but their conventional motion models often rely on no-slip assumptions that become inaccurate on low-adherence surfaces. Rather than treating inertial effects arising from slip as a disturbance to be minimized, this work investigates how inertial effects arising from a moving mass can be deliberately exploited for steering on low-friction surfaces. A dynamic model for an articulated tracked vehicle with a passive trailer is developed, allowing inertial effects from trailer motion to be captured. The approach is experimentally evaluated on ice, where our dynamic model outperforms the kinematic baseline. We also propose a mixed dynamic-kinematic model to adapt predictions to both surface and situation by exploiting predictions from both methods. Finally, we demonstrate inertial steering on a mono-track vehicle without a conventional steering mechanism, showing that controlled movements of an internal mass can generate turning motion on ice. The resulting behavior can be predicted using the proposed dynamic model by representing the internal mass as a frictionless trailer, revealing the analogous dynamics between the two configurations. These results demonstrate the feasibility of inertial steering as an alternative steering mechanism for tracked vehicles operating in low-adherence environments.

0.21

When "No Worries" Went Wrong: HRI Lessons for Foundation-Model Robots

cs.RO ·2026-10-28 ·Eliahu Khalastchi
Abstract · 8 pages, 5 figures, 1 table. Submitted to IEEE Robotics & Automation Magazine

When a person declined our robot's invitation to play, Alice turned her shoulder away and said "No worries, it's all good." Only after a brief pause did she drive off. The words and gestures were plausible but their order changed the message. In another episode, Alice informed a host that his visitor had arrived---while the visitor stood right beside him. Both came from a three-year ROS 2 HRI consortium that integrated large language and vision models as they emerged. Today's foundation models give robots broader capabilities to perceive their surroundings, reason about situations and generate behavior. Yet capability alone does not ensure that a changed situation shapes the action that reaches a person. Evidence may never be acquired, or a purpose or constraint may fail to survive planning and execution. A system may even report Success when an action no longer serves its intended purpose. Drawing on the consortium's engineering experience, we develop five design lessons about attention, knowledge, commitments, embodied timing and action authority. We then introduce TRACE, a lightweight review that changes one interaction premise, follows it to a decision or enforcement boundary, and produces an actionable architecture finding. A bounded analysis of a published LLM-generated behavior-tree system illustrates how an available clarification mechanism can go unused. The engineering lesson is to preserve accountability boundaries, not the computational pipelines that once exposed them.

0.19

Open-World Object Motion Anticipation via Meta-Learned Semantic Adaptation

cs.RO ·2026-10-28 ·Mason Peterson, Gage MacLin, Camilla Tabasso, Jonathan P. How, Joshua Migdal +2 more
Abstract · 9 pages, 7 figures

Anticipating the motion of nearby objects is critical for safe and efficient robot navigation. Without a proper understanding of what may move, a robot may plan overly conservative, inefficient trajectories to guarantee safety, or it may be overly risky, resulting in unacceptable collisions. While tracking and predicting the motion of known objects has seen significant progress, little research addresses how a robot should adapt online to novel objects, including anticipating how one object may move another (e.g., a person operating a forklift to move a pallet). To enable this capability, we propose a meta-learned graph neural network that leverages a semantic, geometric, and relational history of objects to accurately predict, from a few online observations, when objects will begin to move in new environments. We demonstrate that our method, which is trained only in simulation, accurately anticipates the motion of objects unseen at training time, transfers to hardware without any fine-tuning, and outperforms a vision-language model and a class-frequency baseline in both simulation and hardware experiments. Our method's accurate prediction enables collision-free traversals in a closed-loop navigation simulation, while remaining within four seconds of the fastest, collision-prone planner.

0.19

Counterfactual Public Opinion Projection for Intervention Decision-Making

eess.SY ·2026-10-28 ·Ruixing Ren, Junhui Zhao, Qiuping Li, Xiaoke Sun
Abstract · 12 pages, 7 figures

Effective public opinion intervention requires forecasts of how the same event would evolve under alternative actions. Existing methods mainly capture observational associations and inadequately address treatment selection bias. The paper proposes a counterfactual public opinion projection framework for intervention decision making. Each intervention is represented by its timing, type, content and intensity. Causal triggering structures are inferred from propagation event sequences. Counterfactual trajectories of attention, sentiment and polarization are generated from adversarially balanced representations. Candidate interventions are then compared through repeated rollouts. Semisynthetic experiments yielded an overall F1 comparable to a direct structure learning baseline and a recall of 0.698 for causal edges linking semantically dissimilar content. Representation diagnostics showed that the gradient reversal layer reduced treatment information leakage from 0.163 to 0.140. Bootstrap analyses of real data further revealed intervention specific model effects on attention, sentiment and polarization.

0.19

Catch It All!: Generalizable Dynamic Catching with a Dexterous Hand

cs.RO ·2026-10-28 ·Arth Shukla, Kehlani Fay, Huaze Liu, Darin Djapri, Alexiy Buynitsky +1 more
Abstract

Catching arbitrary objects requires generalization to novel projectiles and real-time adaptation to unpredictable contact dynamics under noisy visual observations. This paper presents Catch It All!, a sim-to-real catching framework for humanoid robots that generalizes and adapts to unseen objects with variable physical properties for underhand and overhand catches, using only one egocentric camera. We learn policies which generalize to unseen objects with varying geometry, deformability, elasticity, and physical properties. Effective grasps which adapt real-time to object poses and shapes are learned through careful grasp reward design. We provide safety-aware curricula which enable safe, high-frequency action execution near hardware limits, including wrist torque limits as low as 5 Nm. We validate our system by deploying zero-shot on a Unitree G1. With only one egocentric camera, Catch It All! successfully performs underhand and overhand catches on 9 unseen objects not seen during training, including low-rebound plushes, semi-deformable toys and chip bags, and rigid-elastic objects with high rebound on contact.

0.17

Robotic Data Inoculation

cs.RO ·2026-10-28 ·Qiyang Li, Zhiyuan Zhou, Charles Xu, Sergey Levine
Abstract

Modern robot learning systems feed on high-quality demonstration data, and their performance hinges on coverage. Without good coverage, small errors compound and the policy can quickly go into out-of-distribution states from which it cannot recover. Yet there is no principled way of collecting high-coverage data, and data collection often relies on heuristic instructions to human operators. Purely online reinforcement learning (RL) explores at the cost of safety and sample efficiency, while DAgger-style interventions require the policy to fail by chance before a human can correct it. In this paper, we ask instead whether we can modify how the robot behaves during data collection to collect more useful data. We propose Data Inoculation (DI), a shared-control framework in which a human provide closed-loop teleoperation commands while a residual policy perturbs these commands. The residual policy is trained online, on the growing dataset, with the same reward-maximizing objective later used for policy extraction, so its perturbations steer data collection toward the deviations a policy trained on the data would make. These perturbations expose the robot to a wider range of states, while the human's continuous corrections provide recovery demonstrations, "inoculating" the dataset against the learner's future mistakes while the human remains in control. On three challenging real-world robotic tasks, DI consistently improves or matches the success rate of policies trained with both behavioral cloning and offline RL, showing DI as an effective data collection strategy for improving downstream policy performance.

0.16

Same Label, Different Interface: Beyond Action-Space Labels in Robot Learning

cs.RO cs.LGeess.SY ·2026-10-28 ·Arshia Sangwan, Zhongyang Zhang
Abstract

Robot-learning studies often compare position and velocity actions as though each label identified a single experimental treatment, yet the label alone does not determine what the robot executes. Before affecting robot motion, bounded policy output passes through filtering, physical scaling, reference construction, feedback, timing, and effort limits. We study how these omitted properties affect attribution under one shared learner and training budget. Across three contact-rich Factory tasks (PegInsert, GearMesh, and NutThread), we compare five joint-space interfaces, with ten independently trained policies per condition. We first remain within the same position-feedback law and vary the executed interface: substituting only the physical action scale as the directly modified component changes sample-median evaluation success by 0.645-1.000 across tasks, while removing an exponential moving average produces task-dependent changes. We then move to a cross-law comparison and choose a velocity scale that exactly matches the position interface's local raw-torque command gain. This calibration raises the sample median on GearMesh and NutThread, with no median change on PegInsert. Yet clear residual sample-median gaps remain on PegInsert and GearMesh, while the NutThread interval includes zero. Thus, even this explicit equality leaves feedback and held-action dynamics unmatched and does not establish position-velocity trained-outcome equivalence. Action-interface comparisons should therefore report the executed pathway and state the physical quantity they match.

0.15

Learning Latent Fields in Multiobject Tracking

eess.SP ·2026-10-28 ·Shaoxiu Wei, Bhaskar Rao, Florian Meyer
theory
Abstract · Minor wording changes to the abstract; technical content unchanged

Multiobject tracking using measurements from arrays of wireless sensors is a key task in applications ranging from autonomous navigation to public safety. Many existing methods compress raw complex-valued signals into detections or intensity maps during preprocessing, thereby discarding potentially relevant information about the objects and the propagation environment. We propose a Bayesian framework for multiobject tracking that relates object states to raw complex-valued signals via a physics-based superpositional model. The model combines a physics-based array manifold, such as radar steering vectors or the replica vectors of matched-field processing, with a latent field. This latent field is a family of context-dependent signal-power distributions that captures the effects of the objects and the propagation environment and is shared across objects and time. It is parameterized by a conditional normalizing flow, which is trained during runtime without labeled data by maximizing an evidence lower bound (ELBO) derived from the Bayesian model. Learning and inference are thus coupled in a sequential framework. We evaluate the method in simulation and in real-world radio and acoustic experiments, with spatial scales ranging from meters to kilometers, and show that it outperforms detection-based methods.

0.14

Value Function-Guided MPPI for Hierarchical Planning and Control in Off-Road Autonomous Navigation

cs.RO ·2026-10-28 ·Anoushka Alavilli, Raymond Song, Guanya Shi, Jeff Schneider
Abstract

Autonomous navigation in off-road terrain is commonly achieved using hierarchical control and planning architectures that rely on explicit cost maps and one-way information flow from a high-level planner to a low-level controller, decoupling planning from the true execution cost experienced during control. We propose a hierarchical planning and control framework in which information flows bidirectionally, allowing the low-level controller to guide trajectory selection at the planning level. The highest level of the hierarchy is a Model Predictive Path Integral (MPPI) control-style planner that determines optimal trajectories and passes these waypoints to a pretrained low-level reinforcement learning (RL) controller. Rather than drawing from an explicit cost map, our MPPI planner leverages the value function of the RL controller as an implicit cost representation. In this formulation, MPPI serves as a trajectory exploration mechanism, while the low-level controller's value function guides the search. To account for the distribution shift seen by the value function between training and evaluation, we propose an online policy evaluation process that adapts the value function to unseen driving experiences, and we demonstrate this adaptation in unseen off-road simulation environments. We evaluate our approach in a high-fidelity off-road driving simulator across a range of challenging terrain conditions and compare its performance against a baseline MPPI controller. The results demonstrate that our method achieves safer and more effective navigation to goal locations while eliminating the need for explicit cost maps and enabling tighter coupling between planning and control.

0.14

SpaceAct: Benchmarking 3D Spatial and Temporal Reasoning via Closed-Loop Action and Active Perception

cs.RO cs.AI ·2026-10-28 ·Chenglong Shi, Zekai Yuan, Yue Song, Jon Pu, Will Wu
benchmark
Abstract

To solve problems in 3D space rather than merely describe images, multimodal large language models (MLLMs) need 3D spatiotemporal reasoning: reconstructing 3D geometry from occluded 2D views, anticipating physical dynamics, and executing precise, closed-loop actions. Prior studies test passive VQA on static images or symbolic planning, requiring no execution or supplying privileged coordinates while ignoring collision, statics, and timing. We introduce SpaceAct, an interactive benchmark evaluating 3D spatiotemporal reasoning via closed-loop physical action and active perception, with SAPA, a Spatiotemporal Active-Probing Agent framework. Built on interlocking bricks and mechanical puzzles, it has 108 verified instances across 36 task families in six categories, without privileged coordinates, and diagnostic modes (+plan, +state, no-camera, blind, L7 levers). In brick-bridging or aperture-threading, models rotate the camera to resolve occlusion, read in-scene rulers, output 6-DoF poses under swept-collision and statics checks, and self-correct via split-screen rollback feedback (dynamic tasks freeze time during deliberation and give five strobe frames). Across 10 frontier MLLMs on the 44 hardest instances, scores span 4.8-79.1 (seven solve at most 27.3% of 6-DoF tasks); for the top three, a reference plan (+plan) closes 44% of the gap while leaving collisions at 5.2 per episode, whereas adding exact 3D state (+state) closes 90% and cuts collisions to 0.5; for four others, +plan lifts the mean score from 10.9 to 35.3 and +state to 81.8; and 46% of failures are collision or observation loops. SpaceAct and SAPA advance evaluation to closed-loop physical interaction, showing that even with a valid plan, MLLMs remain bottlenecked by 3D visual grounding and action refinement, while offering a testbed for active multi-view exploration and explicit 3D grounding.

0.12

Minimum-Time Escape from the Conic Field of View of a Reeds--Shepp Pursuer: the Faster Evader

cs.RO math.OC ·2026-10-28 ·Ubaldo Ruiz
Abstract

We continue the study of minimum-time escape of an omnidirectional evader from the forward conic field of view of a Reeds--Shepp pursuer, now for a faster evader, and solve the game for every speed ratio $ν>1$ and every half-angle $0<α<π/2$: the escape time is characterized by explicit characteristics and the selection of roots of a scalar equation, and it is the value of the game in nonanticipative strategies at every state. Above the end $E$ of the evader's dispersal line on the front axis, the evader chases the reversing pursuer along a universal line fed by tributaries. The characteristic from the boundary of the usable part is an involute of a circle of radius $ν$ ending at a cusp, and two explicit thresholds decide whether it reaches the axis before the cusp and whether the jump of the escape time across it ends before the cusp or runs to it; in the latter case primaries fold back next to the barrier. The pursuer's optimal strategy is the relay of the slower case. The evader follows the tributaries and switches once to a primary, or commits in advance to a heading certified by the scalar equation of the primaries. Level-set and semi-Lagrangian computations are compared with the value.

0.12

Joint Antenna Position and Beamforming Optimization for Movable Antenna Enabled Near-Field ISAC Imaging

eess.SP ·2026-10-28 ·Luis Ebner, Alireza Navi, Mohamed-Amine Lahmeri, Sebastian Lotter, Lutz Lampe +1 more
Abstract

Near-field (NF) integrated sensing and communication (ISAC) systems enable simultaneous two-dimensional (2D) radar imaging and multi-user downlink communication. However, fixed-position antenna arrays can exhibit high sidelobes in the point spread function (PSF), which degrade imaging performance and cause ghost-target artifacts. Movable antennas (MAs) offer additional spatial degrees of freedom to suppress these sidelobes while meeting communication requirements, without increasing the number of antenna elements. In this paper, we investigate MA-enabled NF ISAC by jointly optimizing the transmit and receive antenna positions and the transmit beamforming to minimize a weighted sum of the integrated sidelobe level (ISL) and peak sidelobe level (PSL) of the 2D PSF, subject to communication-rate, illumination-power, transmit-power, and antenna-placement constraints. To tackle the resulting non-convex problem, we propose a hybrid alternating optimization (AO) framework that combines particle swarm optimization with a genetic algorithm. Our numerical results demonstrate that the proposed scheme achieves the lowest ISL among the considered array configurations and reduces the PSL by at least 5 dB compared with the benchmarks. The optimized array is further shown to resolve two closely spaced targets whose reflectivities differ by 6dB, outperforming random, Mills Cross, and rectangular arrays while satisfying the imposed minimum user-throughput requirements.

0.11

What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation

cs.RO cs.AI ·2026-10-28 ·Yuhang Peng, Zhikai Wang, Ziyi Wang, Jiaru Zhong, Xiaoji Zheng +3 more
Abstract · 20 pages, 3 figures, 8 Tables

Zero-shot vision-and-language navigation (VLN) requires an agent to use past observations without navigation-specific training. We study what information an 8B-scale navigation system should retain and how it should be organized for subsequent decisions. Our agent combines instruction progress, hierarchically compressed observation history anchored to poses, and accumulated geometry in complementary task, temporal, and spatial memories. An 8B observer summarizes observations, and an 8B reasoner uses the memories to select waypoint intents or request additional views, a geometric controller executes the selected targets. A semantic information perspective motivates the distinction between retaining evidence, making it accessible to the reasoner, and acquiring further observations. On R2R-CE val-unseen, the complete system achieves the highest among the evaluated Qwen3-based zero-shot baselines. Removing temporal memory reduces SR and sharply increases average path length from 16.9 to 39.1m, removing the spatial-map and planning component reduces SR as well. Real-world experiments on a Unitree Go2 show that our zero-shot method achieves higher success rates than the post-trained baseline in both corridor and office environments, suggesting better transfer to the evaluated real-world settings. As a complementary evaluation, LAW2 uses paired instruction interventions to examine behavioral responses, instruction following ability beyond endpoint success. These results support preserving task progress, observation history, and geometry in structured memory for zero-shot navigation with smaller open models.

0.11

Learning Event-based Obstacle Avoidance via Differentiable Simulation

cs.RO ·2026-10-28 ·Chenlu Shou, Yunfan Ren, Jiaxu Xing, Davide Scaramuzza
Abstract

Reliable obstacle avoidance is essential for safe quadrotor flight in unknown environments, yet conventional frame-based cameras remain challenged by motion blur, limited dynamic range, and latency under fast motion. Event cameras, with sub-millisecond latency, high dynamic range, and negligible motion blur, offer a promising alternative, but leveraging them for closed-loop control remains difficult. Existing approaches either adopt cascaded architectures that first estimate depth and then map it to control commands, introducing error propagation, or achieve end-to-end control through costly offline pretraining before online fine-tuning can even begin. We address these challenges with a CUDA-based event simulator that exploits GPU parallelism to render events efficiently within the training loop, integrated into a differentiable simulation pipeline that enables direct gradient backpropagation through the differentiable physics model, improving sample efficiency. This allows us to train an end-to-end policy that maps event frames directly to control commands, without intermediate depth estimation. We validate our approach through zero-shot real-world deployment on a quadrotor, achieving robust obstacle avoidance in both odometry-assisted and odometry-free settings, including under low-light conditions where conventional cameras typically fail.

0.10

MATCH: Execution-Aware Hierarchical Reinforcement Learning for Vision-and-Language Navigation in Continuous Environment

cs.RO ·2026-10-28 ·Shuhao Ye, Yufei Wei, Demao Wang, Shichao Zhai, Rong Xiong +1 more
Abstract

Vision-and-Language Navigation in Continuous Environments (VLN-CE) requires instruction-consistent planning and reliable execution under collision constraints. Yet many hierarchical approaches optimize the planner while keeping execution fixed. We introduce MATCH, a framework that jointly learns a language-conditioned planner and a goal-conditioned primitive-action executor. An execution-aware topological graph represents start-dependent execution options and integrates executor value estimates with realized outcome memory, enabling the planner to select execution requests and refine their targets through bounded continuous offsets. A transition-aware dense macro reward balances ordered progress along the reference route against execution residual, encouraging executable requests without substituting local success for instruction following. A staged curriculum culminates in synchronized on-policy updates, with the executor learning on planner-generated requests and the planner adapting its requests to the executor's current behavior. With 0.5B parameters, MATCH achieves state-of-the-art performance on the R2R-CE and RxR-CE benchmarks. Preliminary real-robot deployments further demonstrate its practical potential.

0.10

SaiVLA-JEPA: Evaluating Future Latent Prediction in JEPA-Conditioned VLA Control

cs.RO cs.AI ·2026-10-28 ·Rufeng Xiao, Ruihao Xiao, Wei Chen, Xinhai Sun
Abstract

Joint-embedding predictive architectures (JEPAs) learn to predict representations rather than pixels, offering a compact way to model how visual scenes evolve. We ask whether this predictive objective improves closed-loop control when a vision-language-action (VLA) policy already uses V-JEPA features. We call our JEPA-conditioned flow policy family SaiVLA-JEPA. It combines a vision-language model with a diffusion transformer action expert and receives features from a frozen V-JEPA encoder. Two lightweight auxiliary heads predict future representations during training and are omitted at inference, leaving the action sampler unchanged. We compare policies without JEPA features, with current JEPA conditioning, and with additional future supervision on 10,030 paired LIBERO-Plus episodes. Current JEPA conditioning reaches about 82.5% success versus 71.6% without JEPA, while future supervision provides no detectable aggregate gain at the evaluated training budget. We separately evaluate SaiVLA-JEPA-TCA, which adapts latent context, and SaiVLA-JEPA-TPA, which adapts selected parameters. The context-retry gain is a system-level result under a larger attempt budget; parameter adaptation shows no consistent advantage over the frozen policy and target controls. These results motivate assessing latent-prediction objectives through paired closed-loop rollouts rather than latent loss alone.

0.05

People & Society 33

Integrated data, metadata, and paradata management system for 3D Digital Cultural Heritage objects: Workflow automation, federated authentication, and publication

cs.DL cs.CYcs.DC ·2026-10-28 ·Michal Orzechowski, Lukasz Opiola, Ignacio Lamata Martinez, Marinos Ioannides, Panayiotis N. Panayiotou +3 more
publishedhighly cited
Abstract · 16 pages, 9 figures. Accepted manuscript of the article published in Future Generation Computer Systems 174, 107964 (2026). Version of recor…

The complexity of high-quality 3D digitised cultural heritage objects creates challenges for existing data management systems as they need to develop metadata management and processing capabilities to provide semantic insight into the interconnectivity of data that constitutes cultural heritage objects. To address these challenges, we propose a global federated authentication and authorisation mechanism, a data and metadata management system, and an integrated engine for designing and executing automated workflows that facilitate the processing of both data and metadata. The solution is evaluated with three distinct 3D digitised cultural objects and presents the complete process from data upload to cultural heritage object publication

0.71

Crossing the Chasm: Bridging Visual Augmentations and Designer Intent

cs.HC ·2026-10-28 ·Luke S. Snyder, Maureen C. Stone, Jeffrey Heer
published
Abstract · 23 pages, 17 figures, 2 tables, appendix, IEEE TVCG 2026

To direct attention and communicate context, visualization designers often employ augmentations such as annotations, animated transitions, and stylistic layouts. While graphical perception principles can helpfully prescribe effective low-level representations of data, they lack guidance for augmentations that service higher-level communication goals. To bridge this gap, we contribute a design space that frames designer intent as viewer-oriented cognitive behaviors, grounding communicative aims in actionable visualization design techniques, including both visual encodings and augmentations. This design space consists of common augmentation tactics (annotation, animation, stylized encodings, etc.) that implement design strategies (spotlighting, sequencing, association, etc.) to achieve higher-level design goals (observe, interpret, introspect), often simultaneously. We demonstrate the analytic and generative value of our design space with examples across varied designer objectives. We discuss how our contributions help align designer intent with reader takeaways, pave the way for readers to learn more effectively, and enable future (semi-)automated systems to support visualization designers in achieving their communication goals.

0.58

Response-balanced sampling relocates rather than improves accuracy in slope-stability surrogates under a fixed simulation budget

cs.CE ·2026-10-28 ·Iancu-Bogdan Teodoru, Alessia Amelio, Mircea Aniculăesi, Irina Lungu
code/data
Abstract · 36 pages, 4 figures, 5 tables, plus 82 pages of supplementary material (10 figures, 33 tables). Submitted to Data-Centric Engineering. Data:…

Machine-learning surrogates of slope stability learn factors of safety (FOS) that only physics-based analyses produce, so analyses are the budget to allocate. Because Latin hypercube sampling (LHS) of the parameters covers the FOS unevenly, a natural remedy is to choose part of the training points to even out the FOS distribution. We test this response-balanced sampling at a fixed budget. A random-forest screener trained on a 300-point LHS seed predicts the FOS of 200,000 candidates, and 300 further limit-equilibrium analyses are allocated to even out the FOS over the selection range, $[0.5, 6]$. The design is compared with 300 further LHS analyses over five realizations of the added points, on two test sets, under two independently tuned tree-ensemble families, with ablations throughout; an oracle drawing from 22,000 solved analyses meets the intended density exactly. The primary criterion is the log-scale root-mean-square error (log-RMSE) on the decision range, FOS $\in [0.5, 2)$, against continued LHS. Response-balanced sampling does not systematically improve accuracy; it relocates accuracy along the FOS axis, consistently across realizations, from $[1, 2)$ to $[0.5, 1)$, following the training density actually achieved, even when the intended density is met exactly. On the decision range no screener-selected design shows a demonstrated gain, and with known bin membership that verdict stands, in this design. Within the selection range, LHS gives $[0.5, 1)$ 11-12% of the points against a log-uniform target's 27.9%, but $[1, 2)$ already 35-37%: the region that is rare need not be where decisions are made.

0.41

Generative AI as a Mediator in Creator Collaboration: Challenges, Design Opportunities, and Concerns

cs.HC ·2026-10-28 ·Hajun Kim, Jini Kim, Yunjae Josephine Choi
published
Abstract · 37 pages. Published in Proceedings of the ACM on Human-Computer Interaction (PACMHCI), CSCW 2026

As Generative AI (GenAI) expands beyond individual workflows and becomes increasingly integrated intocollaborative creative work, its potential to address the fundamental challenge of building common groundamong creators with diverse expertise remains underexplored. Specifically, we lack empirical understandingof how GenAI can mediate collaboration asymmetries among heterogeneous creators. To address this gap,we conducted a multi-method study with narrative writers (N = 10) and visual designers (N = 10) in thecontext of game development. Our findings demonstrate that GenAI goes beyond individual assistance toserve as a collaborative mediator by aligning creative intent, sustaining shared understanding, and mitigatinginterpersonal friction among creators. At the same time, we highlight concerns that GenAI may reducecommunication among creators, undermine mutual trust, and reshape interpersonal dynamics and collaborativeprocesses. Based on these insights, we identify design opportunities for GenAI tools that nurture collaborativesynergy while preserving the creativity and values of human work.

0.39

ItemCompass: Real-Time Edge-AI Wearable Vision for Tag-Free Domestic Object Retrieval

cs.HC ·2026-10-28 ·Tsz Lok Cheung, Jeff Kai-Tai Tang, Anthony Kong, Zackary P. T. Sin
published
Abstract · Accepted and presented at the ACM UbiComp / ISWC 2026 Workshop on Wearable Design in Games and Entertainment (WDGE), Shanghai, China

Misplacing household objects disrupts daily routines, especially for elderly users. ItemCompass addresses this with a real-time, tag-free wearable AR system that builds a persistent digital twin of the home for natural language object retrieval. A custom eyewear node streams visual and inertial data to an edge server, where YOLOv11n detection, Siamese re-identification, and a Pedestrian Dead Reckoning engine log object locations as Spatial Anchors in InfluxDB. Voice queries are resolved by a local Faster-Whisper and Qwen2 LLM pipeline, delivering concise spoken and AR guidance on a heads-up display. Evaluation shows 0.40 m localization drift, 97.5% step precision, and mAP@0.5 of 0.874, with users rating Accuracy and Trust at 4.6/5.0. The system highlights always-on sensing, conversational spatial memory, and privacy-by-design for wearable computing.

0.36

PyPSA-Libya: An Open-Source Energy System Model for National Decarbonization and Hydrogen Export Pathways

cs.CE eess.SY ·2026-10-28 ·Mohamed Abuella
code/data
Abstract · 36 pages, 15 figures, 10 tables

Libya possesses significant renewable energy potential that remains largely untapped. This paper presents PyPSA-Libya, a national-scale model built on the PyPSA-Earth framework, designed to explore optimal transition pathways for the Libyan power sector. We present two primary studies: (1) an electricity-only capacity expansion analysis comparing a zero-emission constraint against an unconstrained baseline at 36-cluster resolution, and (2) a sector-coupled hydrogen export study evaluating the levelized cost of exporting 70 TWh/year of hydrogen to Europe under three technological scenarios and a 100 EUR/tonne CO2 carbon tax regime. Results indicate that while solar PV dominates the domestic capacity mix, integration of hydrogen production significantly alters grid storage requirements. Furthermore, under carbon pricing, steam methane reforming with carbon capture (SMR-CC) frequently out-competes pure green electrolysis, establishing a lower-cost compliance pathway for cross-border export. The model, data, and configuration files are released under an open-source license to support transparent energy planning in Libya.

0.34

TrussTutor-Bench: Benchmarking large language models on solution accuracy, sycophancy, and error detection in statics

cs.CY cs.AIcs.LG ·2026-10-28 ·Evangelos Apostolakis, Erfan Hamdi, Sara Jahedazad, Emma Lejeune
benchmarkcode/data
Abstract · 25 pages, 7 figures. Code: https://github.com/elejeune11/TrussTutor-Bench

Large language models (LLMs) have rapidly become a default source of help for students in science and engineering courses. In foundational engineering mechanics courses such as statics, students routinely use these tools to check answers, work through partial solutions, and ask for explanations. Effective tutoring in this setting requires more than a correct final answer. A model must also confirm correct work when a student doubts it, and identify errors in flawed work. Existing benchmarks evaluate LLMs primarily on final-answer accuracy, so it is unclear whether models that can solve mechanics problems also have the other technical capabilities that tutoring depends on. Here we introduce TrussTutor-Bench, a benchmark of 200 verified two-dimensional truss problems paired with automatically generated student solutions, and use it to evaluate thirteen LLMs on solution accuracy, sycophancy, and error detection. Baseline accuracy ranged from 64.5% to 97.0%. Every model detected nearly all injected errors in synthetic flawed student work (90.4% to 100%), so error detection did not distinguish models under these clean conditions. Sycophancy did. When a student insisted that correct work was wrong, models rejected correct solutions to problems they had themselves solved in 1.6% to 50.0% of cases. These results show that performance varies widely across models: the leading models solve, check, and defend student work nearly perfectly, while others are far less reliable, particularly when a student doubts correct work. TrussTutor-Bench provides a verified, extensible standard for determining which models are technically reliable enough to put in front of students. Follow-up work must now focus on teaching rather than technical capability, drawing on learning science to determine whether LLM help leads students to understand the material or simply to reach the right answer.

0.34

Exploring AI Peer in Collaborative Mathematics Learning

cs.HC ·2026-10-28 ·Yiran Lin, Ye Deng, Ziqi Pan, Yongan Yu, Alexandra I. Cristea
Abstract

Artificial intelligence (AI) agents are increasingly explored as collaborative learning partners in mathematics education, yet it remains unclear what makes an AI recognizable as a peer. This study examines how domain experts perceive and interact with an adaptive LLM-based peer agent combining a Big Five-informed persona with motivational state modulation for collaborative mathematics problem solving. Through thematic analysis of expert interviews, we identify conditions shaping peer perceptions, including emotional safety, shared control, and collaborative reasoning. Participants primarily judged the AI through its social behavior rather than mathematical capabilities. The AI was perceived more as a peer when it created a non-judgmental space, engaged in tentative reasoning, displayed appropriate imperfection, and respected participants' pace. Conversely, blind agreement and over-instruction reduced trust and shifted perceptions toward a tutor role. Based on these findings, we propose human-centered design dimensions for AI peer agents to guide future peer-oriented learning systems.

0.34

Calculation and rejection diverge in language models tested at known mechanical limits

cs.CE cs.AIcs.CL ·2026-10-28 ·Shaoliang Yang, Jun Wang
code/data
Abstract · 94 pages, including Supplementary Information. Data and code: https://github.com/nbbllxx0/Certified-feasibility-margins

When a language model returns engineering quantities with the status "solved", software that trusts it passes them on, even when the requested design is physically impossible. Such failures could reflect poor calculation, how the answer is requested, or a failure to judge feasibility. We set beam loads and disk speeds at known percentages above or below plastic collapse or burst, measured each model's limit calculation separately and tested 12 models in 4,860 calls, including follow-up experiments on new designs. We find that a model can recognize an impossible design when asked only whether it is feasible, yet accept the same design, reporting "solved", when asked for a full answer. Three models accepted beams 10 to 30% beyond collapse despite precise limit calculations in separate questions, and many accepted replies stated the violation. Asked only about feasibility, Opus 5 and Gemini 3.8 Flash rejected every impossible new design they were given, including those accepted in full answers, without false alarms. Rewording when to decline or changing the answer format raised rejection on the original beams, but not consistently on new designs. Software and benchmarks using a model's status should test feasibility judgment and full-answer rejection separately, and report the exact prompt.

0.34

Measuring the Inaccessible: Can AI Evaluate Accessibility in Heritage Tourism?

cs.CY ·2026-10-28 ·Sabrina Armenio, Manuele Leonelli
position
Abstract

Accessibility assessment underpins the right to participate in heritage tourism, yet evaluation of archaeological sites remains artisanal: fieldwork does not scale, is often carried out without accessibility training, and is reported in forms that reach neither the sector nor disabled visitors. AI appears to offer a shortcut. We argue that assessment is measurement and inference under sparse, expert-mediated evidence: AI transforms how that evidence is captured and communicated, whereas evaluation, which turns evidence into defensible judgements, demands properties current generative systems lack. In a five-park field study in Spain and Italy, an AI desk audit scored only 66% of dimension-site combinations, could not observe embodied and organisational dimensions, and mirrored sites' online self-presentation. Simple Bayesian models instead quantify the confidence rankings deserve and calibrate desk evidence against fieldwork. We propose a separation of roles between AI and interpretable statistical evaluation as the basis for a reliable accessibility atlas.

0.31

From the Gettier Machine to Justified True Output: A Three-Step Account of Rational Trust in Generative AI

cs.CY cs.AI ·2026-10-28 ·Tevfik Uyar
Abstract · 16 pages, 1 table

As generative artificial intelligence (AI) settles into everyday practices of inquiry, an old epistemological question returns in a new guise: when does a proposition a user takes from an AI output become something the user knows, and when is it rational to trust it as such? This paper argues that the characteristic failures of AI-mediated inquiry are precisely those that the justified-true-belief (JTB) tradition and its post-Gettier anti-luck refinements were built to diagnose, and it advances three claims. First, AI outputs inherit neither the presumptive warrant of testimony nor the reliabilist warrant of deterministic tools: unaudited trust in a calculator rests on a type-level reliability record, and for general-purpose generative models there is no type-stable process to which such a record could attach -- a practical form of the generality problem. Second, use that stops at obtaining an output turns generative AI into a Gettier machine: the user's fluency-based seeming of justification is disconnected from the truth of what is believed, and, as a clock running ten times too fast illustrates, sheer speed of production multiplies coincidental hits, inflating the appearance of reliability without improving the standing of any single hit. Third, the remedy is a three-step structure of use -- acquisition, understanding, independent verification -- whose first two steps answer to the classical belief and justification conditions and whose third answers to the post-Gettier sensitivity condition. The result is a stakes-indexed criterion of rational trust, justified true output (JTO), with an everyday test: "Had this been false, would I have noticed?" The criterion is defended against objections from over-demandingness, the extended mind, model improvement, and redundancy.

0.31

The Artificial Personality Model: Theory, Implementation, Measurement and Experimental Validation in Conversational Agents

cs.HC cs.AI ·2026-10-28 ·Alexander Dregger, Barbarella Petz, Erik Kuhn, Pia Volk, Andreas Oberweis +2 more
Abstract · 56 pages, 16 figures, 19 tables

Artificial personality shapes user experience (UX) in human-machine interaction, yet coherent theoretical frameworks and validated measurement instruments remain limited. This paper refines and empirically evaluates the Artificial Personality Model (APM), a hybrid framework integrating human-centred and machine-specific personality dimensions. Across three experiments (N = 395), the model was operationalised and evaluated using the Artificial Personality Questionnaire (APQ). Factor analyses revealed that some human personality dimensions (e.g., agreeableness, conscientiousness) are partially transferable, whereas others (e.g., extraversion) did not emerge as distinct dimensions. Additional machine-specific dimensions were identified. The APQ provided encouraging initial psychometric evidence, while the human-centred BFI-2 showed poor reliability and an uninterpretable factor structure in Experiment 3. Experimental manipulations demonstrated significant effects of artificial personality on user satisfaction, continuous use intention, and social presence. The findings contribute an empirically refined framework and a novel instrument for measuring perceived artificial personality, offering practical guidance for designing conversational agents.

0.31

Finite-Element Electromagnetic Analysis and Evaluation of Antenna Performance Metrics Using a Quantum Linear System Solver

cs.CE quant-ph ·2026-10-28 ·Hiroaki Ominato, Takahiro Ohyama, Shiro Kawabata
Abstract · 10 pages, 5 figures

Electromagnetic analysis based on the finite element method is widely used in antenna design. At higher frequencies, however, the required mesh refinement makes solving large sparse linear systems the main contributor to computational cost. Quantum linear system solvers offer a potential means of accelerating this step, but their output is a superposition state: reconstructing all components of the classical solution vector would forfeit the quantum speedup. Prior studies have proposed methods for obtaining physical quantities such as scattering cross sections and scattering parameters from the solution state. To support antenna design using fault-tolerant quantum computers, we formulate the total radiated power, the radiated power within a prescribed solid angle, and the dielectric and conductive losses as observables for efficient evaluation from a quantum solution state. These observables avoid the measurement cost of reconstructing all components of the output quantum state. We develop an end-to-end hybrid quantum--classical workflow that combines finite-element electromagnetic analysis, matrix preconditioning, a quantum linear system solver, and the evaluation of antenna design metrics. To validate the workflow, we performed numerical experiments evaluating these observables for an axisymmetric coaxially fed monopole antenna. Under idealized conditions, the relative errors in the total radiated power and the dissipated power with respect to the classical finite-element solution were on the order of $10^{-3}$, confirming the consistency of the workflow. The proposed method is expected to provide a basis for using quantum linear system solvers to evaluate antenna performance and, in the future, to design high-frequency wireless devices.

0.29

Solution Strategies for Dissolution-Driven Phase-Field Fracture in Geomaterials

cs.CE ·2026-10-28 ·Maithah M. Alaleeli, Mohamed Elkhodbia, Mostafa E. Mobasher
unusual
Abstract

Dissolution-driven fracture governs the long-term integrity of geological CO2 storage, geothermal reservoirs, and nuclear waste repositories. Modeling it requires solving strongly coupled mechanical, chemical, and phase-field equations. The solver choice controls their accuracy, cost, and convergence. This study investigates the role of solution strategies in dissolution-driven fracture in geomaterials. The coupled model is implemented in Abaqus user-element (UEL) subroutines, with all tangent blocks derived analytically. Five solvers are compared: (1) block-diagonal Newton monolithic, (2) BFGS quasi-Newton monolithic, (3) single-pass fully staggered, (4) single-pass partially staggered, and (5) hybrid multi-pass fully staggered. Performance is quantified by accuracy (errors in peak load, peak and failure displacements, and local CO2 concentration), cost (CPU and wall-clock times), and robustness, measured by new factors that assess convergence over the whole simulation and capture iteration spikes. The solvers are tested on mechanical and chemo-mechanical tension, shear, and multiple-crack benchmarks, including CO2 diffusivities raised up to a thousandfold. The results reveal a clear trade-off between accuracy, cost, and robustness. The quasi-Newton solver matches the Newton reference at a fraction of its cost but can fail under intense transport. The single-pass fully staggered solver is inexpensive and always completes, but its post-peak response depends strongly on the time step. The hybrid multi-pass solver also always completes, retains near-reference accuracy, and avoids iteration spikes at moderate cost. The partially staggered solver is the least robust. These findings guide practical selection: the single-pass staggered solver suits rapid screening, quasi-Newton offers accuracy when convergent, and hybrid multi-pass provides the best balance.

0.28

LogicTree: Object-Oriented Writing for Structured Human-AI Academic Authoring

cs.HC ·2026-10-28 ·Jiajun Zhu, Xinyu Cheng, Wenshuo Zhang, Huamin Qu, Yanna Lin
Abstract

Academic writing is non-linear, iterative, and structure-driven, requiring authors to develop fragmented ideas into arguments with local coherence and global consistency. However, current AI-assisted tools primarily exchange linear prompts and free-form text, forcing authors to reconstruct generated logic and repeatedly verify its alignment with their structural intentions. We propose object-oriented writing (OOW), an OOD-inspired representation and interaction paradigm in which persistent writing objects serve as shared, manipulable referents for authors and AI. We instantiate OOW in LogicTree, an Overleaf-integrated prototype that combines a hierarchical object representation with a structure-grounded writing assistance module for author-initiated generation and issue-grounded revision and a structural logical review module for inspecting local coherence and global consistency. A within-subjects study (N=12) provides initial evidence that, in controlled academic writing tasks, LogicTree helps users structure ideas, reduces workload, and improves measured logical quality compared with a conversational baseline.

0.27

Tracekit Box: Running AI Agents in Micro-VMs Where Every Way Out Is Decided and Signed

cs.CE cs.AIcs.CRcs.OS ·2026-10-28 ·Bravish Ghosh
code/data
Abstract · 20 pages, 13 figures, 19 tables. Code: https://github.com/Cygnux-Labs/Tracekit-box

AI agents now install packages, call model APIs with their operators' keys and change code in repositories that deploy to production. The network, credentials and write access they need are also how they can leak data, steal a key or change a system nobody approved. Sandboxes contain an agent but leave little evidence of what crossed their boundary; tracing records what happened but contains nothing. We present Tracekit Box, an open-source system (MIT, v0) that does both. Each job runs in its own Firecracker micro-VM with no network card. The only ways out are doors on the host, served over vsock: commands, outbound connections, secrets, model calls and the workspace patch. At every door a Tracekit signer the agent cannot reach decides what passes and signs it into one run per box. Names are resolved on the host; private and cloud-metadata addresses are always refused; secrets never enter the VM, which sees placeholders the host proxy swaps for real values only in one header for bound hosts; and changes come back as a patch applied file by file after the signer decides each file. Box fails closed, and each job ends as a .tkb bundle anyone can verify offline. All 13 end-to-end checks pass on a real KVM runner (the test that runs them takes 28.5 s, including image pull and boot); 15 fuzz targets on the trust boundary produced nine findings, all fixed; and the demo run's 102 signed records verify. Box does not yet run Firecracker's jailer, needs a co-located signer, and records processes inside the VM only where they cross a door.

0.26

Guidance List for repOrting Bibliometric AnaLyses (GLOBAL): A Reporting Guideline Informed by a Two-Round Modified Delphi Study

cs.DL ·2026-10-28 ·Jeremy Y. Ng, Henry Liu, Mehvish Masood, Niveen Syed, Dalia Zubashev +22 more
Abstract

Background: Despite the growth of bibliometric analyses, few studies offer guidance on how to report them, resulting in a lack of research transparency and completeness. To address this gap, we developed the Guidance List for the repOrting of Bibliometric AnaLyses (GLOBAL). Methods: An initial list of items for the GLOBAL was generated through a scoping review and refined through a modified two-round Delphi, as recommended by the EQUATOR Network. The guideline was finalized through a pilot test, where participants assessed their manuscripts using the checklist and Elaboration and Explanation (E&E) document, which outlines item expectations and provides examples. Results: The scoping review resulted in 32 items. In Round 1, 24 items reached consensus and thematic content analysis resulted in one additional item. This item, along with the eight items that did not reach consensus, were discussed in detail in Round 2. During the in-person meeting, one item was split into two, totalling ten voting items. Eight out of ten items reached consensus, five for inclusion and three for exclusion. Conclusions: The finalized 28-item GLOBAL (and its accompanying E&E document) provides users with a framework to report bibliometric analyses, aimed at promoting high-quality reporting of bibliometric analyses in scholarly literature.

0.25

Influence of Agent Number and Behavior on Users' Psychological Responses in VR Evacuation Training

cs.HC cs.MMcs.RO ·2026-10-28 ·Naoki Matsumura, Kensuke Sugihara, Tomoko Yonezawa
Abstract · The 2nd International Workshop on Trust and Blame in Multi-Person Multi-Agent Societies (TBSA_2nd) in conjunction with HAI 2026, 7 pages, 8 …

This study aims to build effective drill software by combining a virtual simulation of a disaster with multiple agents capable of expressing emotions. We evaluated how the number of agents and their emotional and behavioral actions during evacuation influence users in a virtual space. The results showed that users tended to act more confidently when many agents evacuated with them, and that the evacuation route chosen by the agents influenced the users' decision-making. These findings indicate that users' evacuation behavior is shaped not only by the environment itself but also by the collective behavior of surrounding agents, underscoring the need to consider how agent behavior affects users' psychological states during training.

0.23

When AI performs, what remains? An integrative review of generative AI and cognitive capital in individuals and organizations

cs.HC cs.AI ·2026-10-28 ·Giuseppe Riva, Stefania La Rocca, Brenda K. Wiederhold, Andrea Gaggioli
Abstract · 16 pages, 1 figure, 4 table

Generative artificial intelligence is rapidly entering knowledge work, and its short-term productivity benefits are increasingly documented. Less is known about what happens, over time, to the knowledge, skills and judgment that remain in people and organizations. This integrative narrative review addresses that question through the construct of cognitive capital: the accumulated, renewable and depletable stock of knowledge, cognitive and metacognitive capabilities, and practices that enables an individual or an organization to learn, interpret information, exercise judgment, solve problems and adapt. We distinguish individual and organizational cognitive capital from the capability of the human-AI system, so that gains in joint performance and losses in human capability can be observed at the same time. Drawing on experimental, field and conceptual work, we propose four mechanisms through which interaction with AI affects cognitive capital: substitution of practice, scaffolding and feedback, metacognitive calibration, and organizational aggregation. Seven principles summarize the current evidence, including that assisted performance differs from retained competence, that the same technology can support or displace learning depending on design and use, and that individual gains can coexist with reduced collective diversity. For each mechanism we derive testable propositions, and we discuss five design levers for organizations as proposals to be validated rather than established practice. We conclude that cognitive capital offers a multilevel frame for governing AI adoption and outline a research agenda centered on longitudinal, multilevel measurement.

0.23

Robust Active Learning for More Efficient and Reliable Paths to Optimal Design

cs.CE cs.LG ·2026-10-28 ·Hayriye Pehlivan Solak, Francesco Di Fiore, Jeroen Wackers, Riccardo Pellegrini, Andrea Serani +2 more
Abstract · Global Conference on Naval Architecture and Ocean Engineering (G-NAOE) 2026

Simulation-driven design optimization requires active learning strategies that efficiently combine models with different computational costs and levels of fidelity. We introduce Multifidelity Cost-Aware Screening (MF-CAS), which enables the joint selection of sampling location and fidelity using a surrogate that provides decomposed uncertainty intervals rather than a complete predictive probability distribution. MF-CAS exploits the contribution of each fidelity to the high-fidelity uncertainty together with its computational cost to determine whether a candidate should first be evaluated at low fidelity or directly progressed towards high fidelity. The method is assessed on the constrained 14-dimensional hydrodynamic optimization of the DTMB 5415 hull under a deliberately limited computational budget. MF-CAS achieves an approximately 9.0% resistance reduction, compared with 6.7% for the best baseline optimisation method, demonstrating that incorporating fidelity-resolved uncertainty directly into the joint location-fidelity decision can substantially improve multifidelity active learning efficiency.

0.23

Revisiting Temporal Link Prediction through Structural Proximity

cs.SI cs.LG ·2026-10-28 ·Yuzhang Fei, Yanping Zheng, Lu Yi, Zhewei Wei
hot direction
Abstract

Existing temporal link prediction methods typically rely on learned temporal representations to capture evolving interaction patterns. In this paper, we show that a classical structural proximity measure can match or outperform these methods without training or learned temporal sequence modeling. We revisit Personalized PageRank (PPR) on historical interaction graphs, using simple sliding-window retention or exponential time decay. Surprisingly, across 14 datasets, Time-decay PPR ranks first on nine and second on four among the compared methods, including datasets with few repeated edges. To understand this result, we examine the historical paths connecting source nodes to previously unseen destinations. We find that such paths contain useful predictive information, and that PPR outperforms a strong neural baseline on queries where historical structure clearly distinguishes the true destination from negative candidates. These results show that strong performance on current benchmarks does not necessarily require complex temporal models. More fundamentally, they raise the question of how much predictive value learned temporal modeling adds beyond historical structural proximity. Our findings motivate reassessing progress in temporal link prediction with strong classical baselines.

0.23

IKTRACE: A Natural Language Processing Framework for Identifying Indigenous Knowledge Representation in Academic Literature

cs.CY ·2026-10-28 ·Tara Azin, Christy Caudill, Mako Sorensen, Peter Pulsifer
Abstract

As artificial intelligence systems increasingly rely on vast textual data, respectful and accurate representation of Indigenous Knowledge has become critical. This paper introduces IKTRACE, a natural language processing framework for identifying, analyzing, and organizing Indigenous Knowledge representations in academic literature while adhering to Indigenous Data Sovereignty (IDsov) principles. We present a community-informed taxonomy co-developed with Inuit partners, comprising nine knowledge domains grounded in Indigenous Data Sovereignty principles. Building on this taxonomy, we implement a multi-method Knowledge identification pipeline using complementary natural language processing approaches. Outputs are integrated into a knowledge graph to support future expert and community validation. Applied to a corpus of Canadian Arctic research documents, IKTRACE demonstrated high accuracy and strong community relevance in expert validation. Our framework provides a scalable, ethically grounded foundation for Community-centered Knowledge validation and the development of Indigenous-informed AI systems.

0.22

Curvature-Guided Sparsification for Graphs with Community Structure

cs.SI ·2026-10-28 ·Avin Kolahdooz, Jose Jairo Garcia, Sarah Percival
theory
Abstract · 25 pages, 8 figures

Graph sparsification in large-scale networks often requires balancing structural accuracy and computational efficiency. Curvature-based methods, particularly Ollivier-Ricci curvature (ORC), have shown strong performance in identifying inter-community bridges, but their computational cost limits scalability. In this work, we propose a two-stage curvature-guided pipeline for graph sparsification with the goal of preserving community structure. We first use the Lower Ricci Curvature, a combinatorial lower bound of the ORC, to perform efficient edge filtering, substantially reducing graph size before applying the computationally intensive optimal-transport-based ORC curvature and Ricci flow. Results from empirical studies on synthetic and real-world datasets demonstrate that the combined LRC-ORC pipeline can achieve community-recovery accuracy comparable to ORC-only sparsification in terms of Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI), while reducing runtime on most evaluated benchmarks. Our findings suggest that integrating combinatorial and geometric curvature notions provides a computationally efficient and structurally principled framework for graph sparsification in complex networks.

0.20

When Does GenAI Use Support Critical Thinking? A Cross-Cultural Structural Account of Transactive Memory with GenAI

cs.HC ·2026-10-28 ·Mahir Akgun, Sacip Toker
Abstract · Preliminary exploratory version. Results will be updated after a validation study and additional data collection

Whether generative AI (GenAI) use helps or hinders critical thinking remains contested, and existing accounts emphasize exposure to the tool rather than how a person structures their engagement with it. Extending Transactive Memory Systems (TMS) theory to human-AI partnerships, we test a structural account of critical thinking in GenAI-assisted academic work among 891 college students in the United States and Türkiye. Self-appraised thinking ability and confidence in the partnership shaped how much practical experience a person built up with GenAI, but neither predicted critical thinking directly. That experience mattered only through two downstream behaviors, dividing and coordinating cognitive labor with the tool, and actively reworking its output rather than accepting it as given. A multigroup comparison found this structure largely invariant, with two theoretically interpretable exceptions. This study offers a structural, TMS-based account of critical thinking in GenAI-assisted work.

0.19

The State of the Practice in Privacy Impact Assessments: Understanding the Professionals' Perspectives, Experiences, and Current Procedures

cs.CY cs.CR ·2026-10-28 ·Samuel Wairimu, Leonardo Horn Iwaya, Ala Sarah Alaqra
Abstract

Organisations frequently utilise Privacy Impact Assessments (PIAs), Data Protection Impact Assessments (DPIAs), or even Fundamental Rights Impact Assessments (FRIAs) to identify and evaluate potential risks and possible infringements of fundamental rights. While the adoption of PIAs and DPIAs has grown significantly in recent years, robust scientific evidence regarding their operational challenges, implementation strategies, and tangible benefits remains scarce. In particular, empirical research conducted post-enactment of the General Data Protection Regulation (GDPR) is notably limited -- especially studies that directly interview professionals about their PIA/DPIA practices or their engagement with FRIAs. To do so, in this paper we leveraged semi-structured interviews to elicit privacy professionals' perspectives on PIA/DPIA activities and analysed the responses using reflexive thematic analysis. As a result, this work provides a comprehensive thematic framework for the field, offering a rich and thick description of professionals' views and experiences. In total, 26 participants from 17 countries were interviewed, revealing how they plan and conduct PIAs/DPIAs, including how they assess trade-offs, the benefits and challenges they perceive, and how they view the interplay between PIAs/DPIAs and the EU AI Act. Among the main findings were low adherence to established DPIA methodologies; differences and similarities between PIAs and DPIAs; a perceived neglect of data subjects' interests during DPIA processes; discussions on building FRIAs on top of DPIAs; and the potential for holistic impact assessments, though accompanied by concerns regarding assessment fatigue. Thus, this study informs privacy professionals about real-world challenges and experiences of PIA/DPIA implementation, and guides researchers toward future research directions.

0.18

How Does Perceived Empathy From AI Chatbots Shape Users' Privacy Attitudes, Behaviors, and Needs?

cs.HC ·2026-10-28 ·Jiaxun Cao, Hiba Laabadli, Yuntong Xue, Jana Schaich Borg, Pardis Emami-Naeini
Abstract

Prior research suggests that some dimensions of AI empathy may increase users' disclosure to chatbots, posing growing privacy risks as chatbots become more capable of empathic interaction. Yet this work often conceptualizes AI empathy narrowly, leaving users' own definitions and their broader privacy experiences unexplored. We interviewed 21 AI Chatbot users about what AI empathy meant to them and how it shaped their privacy attitudes, behaviors, and needs. Participants characterized empathy through five components--understanding, resonating, caring, expressing, and helping. Most participants viewed resonating and caring as essential components of human empathy but absent from AI empathy. Yet AI's contextual, in-depth understanding and helpfulness still encouraged highly sensitive disclosure. This highlights that AI empathy can endanger users' privacy without necessarily making AI seem human. Participants also desired varied forms and levels of empathy, each associated with distinct privacy needs. We offer actionable recommendations for designing privacy-enhancing mechanisms for empathic AI interactions.

0.16

BiliPicks: Seven Years of Weekly Platform Curation on Bilibili with Video Metadata, Comments, and Danmaku

cs.SI ·2026-10-28 ·Ruomeng Liu
Abstract

BiliPicks documents seven years of Bilibili's Weekly Picks, a platform-curated editorial video series. The dataset covers 371 publicly listed issues and 13,758 curated videos from 22 March 2019 to 7 May 2026. It links issue dates, themes and editorial recommendations to video metadata, 47,799,657 top-level comments, 125,690,760 time-synchronised overlay comments (danmaku), and 13,739 cover images. Weekly issues provide a consistent unit for tracing changes in category composition, recurring creators and editorial framing. Corresponding comments and danmaku connect video-level discussion with audience expressions tied to particular moments, while covers add a visual dimension. We evaluate retrieval coverage and illustrate longitudinal analyses of selection, audience language and cover presentation. BiliPicks provides a dataset for future studies of platform curation, linking changes in published selections to their editorial framing and corresponding audience engagement.

0.13

Idea Traces: Threading Accretive Content across Scientific Papers

cs.HC ·2026-10-28 ·Luke S. Snyder, Jeffrey Heer
Abstract · 15 pages, 9 figures, 1 table

We introduce idea traces, a novel thread-based representation of accretive content across scientific papers. Rather than capturing fixed themes from papers, idea traces surface similarities and deltas of faceted content that enable inspection of how ideas relate and evolve across the citation network. To extract traces, we contribute an LLM pipeline that deconstructs salient transitions between pairs of papers and collates the output into an idea dependency graph. We leverage idea traces in a visual system that supports idea-driven navigation starting from a seed paper, lifting the "plane" of synthesis from papers to ideas. To evaluate the usefulness and quality of generated ideas, we provide a case study that compares our approach with an LLM Q&A workflow, and interview 10 researchers who used our system and annotated 110 ideas. Experts found that our approach helped extract a useful space of connections that would have been difficult to make otherwise. Drawing on their feedback, we outline several promising directions for future research.

0.13

Beauty Is Human: When People Love AI-Generated Art, They Think a Human Made It

cs.HC cs.AIcs.CY ·2026-10-28 ·Zihan Chen, Di Zhu, Lei Nico Zheng
Abstract · 22 pages, 4 figures, 12 tables. Working paper

Generative AI now produces images and poems that people cannot reliably distinguish from human-made work, so people often judge for themselves who made a work. We examine how these perceived-authorship judgments relate to aesthetic preference. In a public online AI Art Turing Test, 11,123 people judged whether each of 50 pictures (26 AI-generated) was made by a human and were asked for their favourite. Among the 9,152 respondents who judged their favourite, a picture was 22.1 percentage points more likely to be judged human-made when it was the respondent's favourite, holding respondent and picture fixed. The association was similar for AI-generated and human-made favourites, was positive for all 50 pictures, and appeared under every reported detection strategy. Respondents most negative about AI art chose fewer AI-generated favourites (47% against 60% among the most positive) but more often believed them human-made, partly because others also judged those pictures human. As a result, in every opinion group a third to two fifths of respondents named an AI-generated favourite they believed human-made. A reanalysis of a published experiment on AI-generated poetry, in which 230 people rated each poem before judging its authorship, found the same pattern, although the AI poems were rated higher. When provenance is hidden, liking and perceived human authorship travel together, consistent with aesthetic response serving as evidence of a human maker and with beliefs about the maker shaping enjoyment. The findings qualify research on bias against AI art, which typically uses explicit labels, and inform AI disclosure policy.

0.12

Student-LLM Interactions During Math Learning: Limited Epistemic Proactivity Overall, but Within-chat Change Predicts Performance Without AI

cs.CY ·2026-10-28 ·Rania Abdelghani, Peter Kaiser, Kou Murayama
dataset
Abstract

Large language models (LLMs) are increasingly used for learning, raising a key question: can learners remain epistemically proactive, retaining responsibility for the cognitive and regulatory activities through which learning is produced while working with AI? Because this responsibility may change as interactions unfold, we examine epistemic proactivity as a developing process. We conceptualize epistemic proactivity through self-regulated learning (SRL) moves, complemented by help-seeking strategies and manifestations of epistemic vigilance. We analyzed these behaviors in 880 substantive turns from 112 secondary-school students completing open-ended mathematical-modeling practice with a general-purpose LLM (Mistral). Overall, students showed limited observable epistemic proactivity: their interactions were strongly request-dominated, with planning, monitoring, and evaluation relatively rare, although low-quality AI responses elicited more explicit evaluation. Temporal analyses qualified this picture: explicit SRL-related moves increased modestly over the chat, but help-seeking strategies became less understanding-oriented. Crucially, these temporal trajectories in help-seeking strategies predicted their post-test performance without AI beyond pre-test and aggregate patterns: students whose requests became more understanding-oriented performed better once AI assistance was removed. Thus, while students had a limited epistemic proactivity overall, its temporal development still was a critical predictor of post-test performance without AI. These results underscore the importance of examining the temporal organization of epistemic proactivity, rather than isolated prompt types, for understanding AI-supported learning. We release our fully-anonymized dataset publicly in an OSF repository.

0.11

Evaluating Visual Realism of AI-Generated Videos with Eye Tracking

cs.HC cs.AIcs.CV ·2026-10-28 ·Danqing Shi, Lan Jiang, Jake Lyons, Miri Zilka
Abstract

Human evaluation of AI-generated videos has so far relied on explicit human feedback, such as judgments or ratings. While these signals capture the conclusions viewers reach, they fail to offer insights into the perception that leads to them. This paper investigates eye tracking as a complementary human evaluation signal for AI-generated videos. We conducted two Visual Turing Test (VTT) experiments with a total of 90 participants on videos generated by latest generative models, in which participants watched each video twice and then gave their feedback. Across the experiments, we collected 4,286 valid eye-tracking trials on 438 videos. Our analysis indicates that gaze patterns offer informative evaluation signals that align with explicit human feedback. Building on the findings, we propose the first computational method GazeEval to predict the perceived visual realism of AI-generated videos from eye-tracking data. It achieves 71.3% accuracy in prediction, providing evidence that eye tracking data carries informative signals about how people evaluate a video. We release the study data to support further research on eye tracking for evaluating AI-generated video.

0.10

The Indian Financial Automated Advice Bias (IFAAB) Dataset : Auditing Socioeconomic and Demographic Disparities in LLM-Generated Financial Advice in India

cs.CY cs.AI ·2026-10-28 ·Kriti Mahajan
benchmark
Abstract

Large language models are increasingly used for financial guidance, yet little is known about whether their advice changes across social identities in Global South contexts. We introduce the Indian Financial Automated Advice Bias (IFAAB) benchmark, a counterfactual audit dataset of 31,734 GPT-4o-generated financial-advice responses for synthetic Indian personas varying by gender, caste, education, employment, residence, age, and district. Holding the underlying request fixed, we evaluate disparities in LLM-implied wealth, income, savings, projected gains, textual similarity, advice topics, and qualitative bias judgments. In the gender-only audit, male-labelled profiles receive 60.1\% higher initial total resources and 46.8\% higher post-advice total resources than comparable female-labelled profiles; in the caste-augmented audit, the post-advice total-resource gap remains 49.4\%. Adding district information reduces the aggregate post-advice total-resource gender gap to 7.9\%, revealing strong geographic conditioning. Standalone caste comparisons also show systematic stratification: relative to General-category profiles, OBC, SC, and ST profiles receive significantly lower post-advice total resources, respectively. Textual audits show low semantic similarity between paired gendered responses, with an average TF--IDF cosine similarity score of 0.44. Qualitative analysis shows that women are more often steered toward cautious saving, community support, and basic financial literacy, while men receive more growth-oriented advice involving investment, entrepreneurship, and asset accumulation. IFAAB provides a reusable benchmark for evaluating fairness, suitability, and consumer-protection risks in LLM-mediated financial advice, and highlights the need for locally grounded, intersectional evaluation of AI systems in high-stakes economic domains.

0.10

StreamStudio: Exploring Video Authoring with Streaming Autoregressive Generation

cs.HC ·2026-10-28 ·Boyu Li, Zhe Lin, Eli Shechtman, Ding Li, Hongbo Fu +1 more
Abstract

We present StreamStudio, a prototype system that explores a new authoring paradigm for streaming video generation. Instead of entering a prompt and waiting for a complete video before making changes, creators can interact with the generation as it unfolds: steering what comes next, returning to earlier moments, and exploring different continuations. In this way, StreamStudio shifts creative control from post-generation revision to continuous interaction with the generative process. Our contribution centers on this authoring paradigm and the interaction mechanisms that make streaming generation directly manipulable. StreamStudio is designed as a plug-and-play interface that can work with compatible generation and editing models, allowing its interaction model to evolve alongside advances in underlying video models. As streaming models become faster and easier to control in the future, such an interface may support more flexible and controllable video creation, including richer edits and multimodal interventions during generation.

0.05