CS Paper Radar

Daily listing

2026-11-07

429 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 169

Hydro-Dataset: A Multi-Modal Open-Source Pipeline for Nordic Hydrological Research and Foundation Model Benchmarking

cs.LG ·2026-11-07 ·Getnet Demil, Md Moinul Islam, Mourad Oussalah
benchmarkdatasetcode/data
Abstract · Accepted for presentation at ICIP Workshop 2026 and to be published as part of the conference proceedings

Despite world-class snow monitoring by the Swedish Meteorological and Hydrological Institute (SMHI) and Finnish Meteorological Institute (FMI), no station-anchored, multi-parameter snow benchmark dataset exists for the Nordic boreal domain. We present Hydro-Dataset v1.1, a reproducible open-source pipeline that harmonizes 582 SMHI meteorological observation stations across nine hydrological winters (2015/16-2023/24) into a CF-1.9 NetCDF product with 1.8~million annotated station-days. Six derived snow parameters, such as Snow Water Equivalent (SWE), snow phenology, melt rates, degree-day melt factor, rain-on-snow flags, and freeze-thaw cycles, supply benchmark-ready labels for hydrological foundation model training. ERA5-Land cross-validation (Pearson R=0.78, bias=-262mm) confirms observation-anchored SWE provides a more reliable benchmark than reanalysis for machine learning training.

0.66

An Analysis of Reinforcement Learning with Function Approximation

cs.LG ·2026-11-07 ·Francisco S. Melo, Sean P. Meyn, M. Isabel Ribeiro
top venuehighly cited
Abstract · This version of the paper corrects the one that was presented at the 25th International Conference on Machine Learning. We corrected conditi…

We address the problem of computing the optimal Q-function in Markov decision problems with infinite state-space. We analyze the convergence properties of several variations of Q-learning when combined with function approximation, extending the analysis of TD-learning in Tsitsiklis and Van Roy (1996) to stochastic control settings. We identify conditions under which such approximate methods converge with probability 1. We conclude with a brief discussion on the general applicability of our results and compare them with several related works.

0.63

Seeing Isn't Enough: Unlocking Visual Reasoning by Bridging the Reasoning-Access Gap with TIGER

cs.AI cs.CVcs.LG ·2026-11-07 ·Gregory Kang Ruey Lau, Minh Huynh Nguyen, Bryan Kian Hsiang Low
top venuecode/data
Abstract · Accepted at NeurIPS 2026. Gregory and Minh contributed equally

Multimodal Large Language Models (MLLMs) inherit strong textual reasoning capabilities from their language backbones and text pretraining, but often fail to apply the same level of competency to visual inputs. We show that this degradation cannot be attributed to only perception failures, by evaluating models on semantically-equivalent text and rendered-text image inputs: models can accurately read text in the images and recover near-text performance when provided their own transcriptions, but just cannot reason well on the images directly. We characterize this discrepancy as a modality reasoning-access gap, where models fail to route perceived visual content into existing reasoning machinery accessible with text. To bridge this, we propose TIGER (Text-to-Image Gap-targeted training for Enhanced Reasoning), which first mines synthetic modality counterfactuals where a model succeeds on a text problem but fails when provided the corresponding rendered image, and post-trains the model on the image input via Group Relative Policy Optimization (GRPO), along with SFT and DPO variants. Importantly, although trained exclusively on synthetic rendered-text images with no multimodal annotations, TIGER results in model performance improvements across diverse real-world visual benchmarks including geometric diagrams (MathVerse), multi-disciplinary sciences (EMMA, MMMU), complex charts (ChartQAPro), and embodied spatial reasoning (RoboSpatial-Home). TIGER also complements existing state-of-the-art RLVR visual reasoning methods, yielding further performance gains. Analysis involving reasoning-subspace activation and activation patching provide support indicating that TIGER helps to align visual-to-reasoning pathways toward text reasoning states. Our implementation is available at https://github.com/minhngh/Tiger.

0.60

JUBILEE: A Benchmark Environment for Off-Policy Evaluation in Debt Collections

cs.LG q-fin.RMstat.ML ·2026-11-07 ·Diego Silva Prado, Tiago Amaral Penteado de Jesus, Bruno Caldas de Souza, Cristiano Bevitori Maffia De Oliveira
benchmarkcode/data
Abstract · 28 pages, 3 figures, 35 tables. Code: https://github.com/alpha-synthesis/jubilee. Data: https://doi.org/10.5281/zenodo.24008544

Off-policy evaluation (OPE) estimates the value a candidate decision policy would achieve, using only data logged by the policy already in production. In debt collections, unrecorded factors can influence both actions and outcomes. This paper presents JUBILEE, a benchmark environment built around a debt-collections simulator that generates million-loan panels with competing risks, rare recovery events, and hidden factors driving both outcomes and assignment. The evaluation includes an 80-run comparison of tabular and gradient-boosted estimators under two synthetic outcome mechanisms and a 190-run parameter grid. Independent policy simulations provide reference values with estimated Monte Carlo uncertainty for a synthetic reward score and for simulated cash. With hidden factors influencing action assignment, several tabular methods agree closely while underestimating the field-visit policy's score, an error that also appears with boosted models. Intervals from tabular doubly robust estimation usually contain the reference without hidden assignment dependence but mostly miss it when that dependence is present. Estimates from randomized decisions do not show the same large error, although their limited precision leaves policy choice uncertain. The decision consequences depend on how policy value is defined. The field-visit policy ranks last under the score and first under simulated cash, which excludes channel costs. Under cash, log-based estimators miss this best candidate in nearly every run with hidden assignment dependence in the paired study. JUBILEE supports reproducible study of these failure modes under specified collections mechanisms.

0.60

Complementarity without ground truth?

cs.AI cs.HC ·2026-11-07 ·Andrea Ferrario, Alessandro Facchini, Federico Cabitza
Abstract · 15 pages, 2 figures

Human-AI complementarity captures the idea that a human assisted by an AI system can outperform both the human and the AI acting alone. In prediction tasks, this improvement is assessed against ground truth labels produced through a reference construction procedure and intended to represent the state of affairs to be predicted. Many human-AI interactions, including those where medical AI provides treatment recommendations, however, concern tasks other than predicting what obtains. Their relevant output concerns adequacy relative to a practical goal and depends on several evaluative dimensions. In this work, we examine the ground truth dependence of complementarity and extend it to human-AI interactions where ground truth is not the appropriate reference. We show that tasks beyond prediction can still admit a reference construction procedure that generates adequacy labels. We develop a route for extending complementarity to such settings. It replaces ground truth labels with context-dependent adequacy labels generated through a two-step scoring methodology inspired by multi-criteria decision analysis frameworks in healthcare and it makes the normative choices underlying reference construction explicit. Because the construction of adequacy labels embeds evaluative choices, we draw methodological inspiration from computational reliabilism to identify technical, scientific, and social factors relevant to supporting their construction. Our approach expands the empirical study of human-AI complementarity beyond prediction tasks, enabling its quantitative evaluation in decision-support settings where ground truth is not the appropriate evaluative reference.

0.52

Online Imitation Learning for Stabilizing Vlasov--Poisson Plasmas

cs.LG math.OCphysics.plasm-ph ·2026-11-07 ·Xiaofan Xia, Qin Li, Wenlong Mou
top venue
Abstract · Accepted at NeurIPS 2026. 43 pages, 9 figures, 2 tables

We study the problem of controlling instability in the Vlasov--Poisson plasma systems, a fundamental challenge for nuclear fusion control. Motivated by the partial observability of plasma states in practice, we investigate imitation learning algorithms that learn from a fully-observable expert controller, but operate under macroscopic measurements. We theoretically establish a separation between online and offline imitation learning --- even a small behavior cloning error leads to error compounding that exponentially amplifies instability over time. By way of contrast, the online imitation learning loss can polynomially control the stability of rollout trajectories. We then propose a DAgger-style algorithm that learns the stabilizing controller under partial observability. Simulation results on a 1D Vlasov--Poisson system demonstrate that our algorithm can effectively stabilize plasma instabilities over longer time horizons than behavior cloning.

0.52

FADE: Fractional Anomalous Dynamics Extrapolation for Training-Free Diffusion Transformers Acceleration

cs.LG cs.AI ·2026-11-07 ·Jinlong Yang, Lizilin, Jinke Wu, Yao Zhou
top venuecode/data
Abstract · Accepted at NeurIPS 2026

Diffusion Transformers (DiTs) have established themselves as the preeminent paradigm for high-fidelity generative modeling across various modalities, yet their practical deployment is hindered by the severe computational overhead of iterative sampling. While recent feature-caching methods mitigate this latency through feature reuse or forecasting based on integer-order operators, they remain limited in adapting to the non-stationary and highly curved dynamics of DiT latent trajectories. In this work, we reveal that DiT feature evolution exhibits anomalous diffusion characteristics and a certain degree of spatiotemporal heterogeneity, rendering static, integer-order forecasting operators inadequate and prone to error accumulation. To address this, we propose Fractional Anomalous Dynamics Extrapolation (FADE), a unified framework for training-free diffusion transformers acceleration. FADE models DiT feature evolution via anomalous diffusion and employs a Mittag--Leffler fractional operator to better adapt to high-curvature and non-linear latent trajectories of DiTs. Leveraging the empirical observation that these complex dynamics maintain cross-sample stability, we further construct a Look-Up Table with multiple dimensions through offline profiling. This enables negligible LUT-retrieval overhead, spatiotemporally adaptive parameter retrieval during online inference. Extensive evaluations across diverse architectures and modalities demonstrate that FADE achieves state-of-the-art acceleration, preserving exceptional generation fidelity under aggressive skip intervals with negligible latency overhead. Our project page is available at https://youngkinlon.github.io/fade-project-page/.

0.49

PIA: Agentic Physics-Informed Runtime Anomaly Detection

cs.LG cs.AI ·2026-11-07 ·Ali Mohammadi Ruzbahani, Abbas Yazdinejad, Hadis Karimipour
benchmark
Abstract

Runtime anomaly detectors for cyber-physical systems are either learned as black boxes from normal data, which miss stealthy manipulations that keep every signal inside its normal range, or derived by hand from physical models, which requires expert effort for every plant. We introduce physicsinformed agents (PIA), in which a large language model agent writes the detector itself as a set of physical residuals, such as pump characteristics, tank mass balances, and controller logic, and refines every candidate against normal operating data through an evaluation workbench before deployment. The result is a compact, human-readable detector that runs in milliseconds and names the component behind each alarm. On the BATADAL water distribution attack benchmark, PIA with Claude Opus 5.5 reaches an F1 of 0.878 on the test attacks against 0.504 for the strongest of six statistical and deep baselines, keeps detecting attacks concealed by replayed and offset sensor readings, and attributes every detected test attack to the attacked component. Ablations show that testing hypotheses against data is decisive: agents that write residuals from the plant description alone lose up to 0.28 F1, whereas agents that see only anonymized signals recover most of the physical structure.

0.49

Structurally Separated DAG Learning with Multi-Scale Normalized Closure

cs.LG cs.AI ·2026-11-07 ·Shuai Feng, Hao Yan
top venuetheory
Abstract · 33 pages, 7 figures. Accepted at NeurIPS 2026

Continuous DAG learning commonly enforces acyclicity through scalar functions of a weighted adjacency that aggregate evidence from weighted closed walks. We add a complementary layer: matrix-valued closure logits---explicit multi-hop structural relation scores for each ordered pair---supervised alongside data fit, sparsity, and a DAGMA-style barrier on the weighted adjacency. The distinction is not scalarization of the overall objective but how path evidence is represented and exposed before aggregation. SMNC (scale-normalized multi-scale closure) builds these scores on a structural gate by repeated squaring in log space and divides each scale by its path-length budget, counteracting product-style attenuation along long cycles; diagonal entries support node-wise cycle supervision and off-diagonals optional pairwise reachability cues. In the linear model, a bounded structure--magnitude factorization ties strong effective edges to strong gates; a nonlinear induced gate offers another construction with the same intent. For sufficiently strong cycles, normalized logits admit a length-independent lower bound; under a unique-active-cycle assumption, the hard final-scale readout yields a polynomial log-gate gradient factor rather than exponential product-style decay. On cue-free, unstandardized Gaussian linear and nonlinear ER benchmarks, these models improve over DAGMA, run with default hyperparameters at a reduced budget, on denser graphs and are close to it on sparse ones.

0.49

MMGraph-Agent: Agentic Multimodal RAG via Cache-Inspired Multimodal Knowledge HyperGraphs

cs.AI ·2026-11-07 ·Haoran Luo, Ziyue Zhu, Shangyang Wu, Linhao Luo, Yikai Guo +6 more
top venuecode/data
Abstract · Accepted by NeurIPS 2026 main conference

Driven by increasingly complex real-world applications, Retrieval-Augmented Generation (RAG) has evolved from textual pipelines to multimodal settings, comprising two complementary dimensions: knowledge organization and retrieval enhancement. However, existing methods still face challenges of high construction cost, context window limitations, and the lack of unified training across offline and online knowledge sources. To address these challenges, we propose MMGraph-Agent, a unified multimodal RAG framework based on cache-inspired multimodal knowledge hypergraphs. MMGraph-Agent enables extremely lightweight hypergraph construction with zero API cost and dynamic memory-based retrieval that alleviates long-context issues. Experiments across offline, online, and joint retrieval settings demonstrate consistent improvements in performance, efficiency, and architectural unification. Our software and data are publicly available at https://github.com/LHRLAB/MMGraph-Agent.

0.49

EEGFaceSem: An EEG Benchmark with Paired Generative Latents for Semantic Visual Modeling

cs.LG ·2026-11-07 ·Jun Ma, Michiel M. Spapé, Keith M. Davis, Tuukka Ruotsalo
publishedbenchmarkdatasetcode/data
Abstract · 18 pages, 5 figures. Accepted at NeurIPS 2026, Evaluations and Datasets Track. Conference page: https://neurips.cc/virtual/2026/loc/paris/po…

Electroencephalography (EEG) is widely used to measure brain activity for brain-computer interface applications for its non-invasive nature and high temporal resolution. However, progress in modeling the neural responses to visual semantics has been limited by a reliance on fixed image collections that lack the fine-grained control needed to isolate fine-grained changes in visual features and their relation to perception. In this work, we present EEGFaceSem, a large-scale dataset pairing high-quality EEG with photorealistic face images synthesized from a generative model. We provide the generative latent vector for each stimulus, enabling research into the continuous mapping between a visual latent space and neural representations. Each participant's brain responses were recorded while they attended to specific semantic features (e.g., smiling, hair color) in a rapid serial visual presentation paradigm. The dataset contains EEG signals from 30 participants, totaling 67,200 time-locked EEG recordings. We present benchmarks for single-trial semantic classification under single-subject, cross-subject, and subject-adapted settings, as well as for cross-subject voting that aggregates per-stimulus predictions across participants. We further demonstrate the dataset's unique utility with two complementary applications: deriving semantic directions in the latent space for brain-supervised image editing, and conditioning a generative model on brain-decoded labels for brain-supervised image generation. Our dataset paves the way for advancing brain-computer interfaces (BCI) for semantic-level visual saliency detection and brain-supervised generative modeling. Our dataset and code are openly released at OSF (https://osf.io/2guk3/) and at HuggingFace https://huggingface.co/datasets/yefllower/EEGFaceSem.

0.48

GradReason: LLMs as Meta-Learners that Replace Training Epochs with Training-Data-Free Weight Updates

cs.LG cs.AI ·2026-11-07 ·Ali Mohammadi Ruzbahani, Abbas Yazdinejad, Hadis Karimipour
Abstract

Standard training recipes make a fixed number of passes over the training data, even when an epoch mostly adds mini-batch noise. We study whether a large language model (LLM) that reads the weight and gradient history of a run can replace some of these epochs with updates that compute no gradient and read no training image. In GradReason, an LLM agent is called every ten epochs. In a restricted sandbox, it analyses recent weight snapshots, epoch-mean gradients, and validation metrics, then submits a Python program that computes new weights and a plan marking each of the next ten epochs as data or virtual. A virtual epoch is kept only if it passes a forward-only check on 5,000 held-out validation images. In 120 runs on CIFAR-10 and CIFAR-100 with three architectures, Claude Opus 5.5 used 32.8% fewer data epochs than the 80-epoch recipe at 0.31 percentage points lower test accuracy (95% CI -0.72 to +0.09; 0.94 points on ResNet-18), on the recipe's own frontier of data against accuracy. Unprompted, Opus averaged recent weight snapshots in every run. This choice produced the savings: a fixed rule with the same acceptance check but extrapolations from prior work saved 5.4% of the data epochs, and the same rule with averaging added saved 25.0%. The LLM's contribution is to find a working update within a single run; once found, a fixed rule implements it equally well. An Opus run cost 0.68 USD, but at this scale the agent's latency exceeded the GPU time it saved.

0.46

Predictive Surprise as Self-Grounding Concept Bottleneck for Interpretable Time Series

cs.LG cs.AIeess.SPstat.ML ·2026-11-07 ·Sachith Abeywickrama, Emadeldeen Eldele, Min Wu, Xiaoli Li, Chau Yuen
publishedcode/data
Abstract · Accepted at NeurIPS 2026, Main Track, Poster

Time-series classifiers deployed in risk-sensitive domains must be (1) accurate, (2) interpretable, and (3) able to abstain when evidence is weak. Existing methods do not deliver all 3 at once, and the obstacle is structural, rather than logistical. Meaningful temporal patterns cannot be defined without knowing segment boundaries, yet meaningful boundaries cannot be placed without knowing which patterns to expect. Existing approaches break this circularity by using fixed windows or expert-annotated vocabularies, thereby sacrificing at least one of the 3 properties. We introduce ConceptTime, which closes the loop through a single self-supervised signal. A frozen probabilistic forecaster predicts a distribution over the next observation at every timestep of the sequence. We call the mismatch between its prediction and the realized signal the predictive surprise. This signal both locates segment boundaries and characterizes each resulting segment through the forecaster. Segments are thus grounded by how they behaved relative to expectation, not by what they look like. These summaries cluster into a vocabulary of dynamical regimes that we call concepts. A lightweight head classifies from the concept sequence, and the distance to the nearest concept yields reliability by design. Across 17 public benchmarks, ConceptTime is competitive with state-of-the-art black-box accuracy, beats every interpretable baseline, and retains near-98% accuracy on Epilepsy with 1% of labels. It further remains competitive with softmax, prototype-distance, and SHAP/TimeX/TimeX++ on OOD detection and deletion faithfulness, without a single annotation, language model call, or domain expert. Code is available at: https://github.com/Sachithx/ConceptTime.git.

0.46

Quantitative Assessment of Crystal Structure Prediction

cs.LG cond-mat.mtrl-scics.AI ·2026-11-07 ·Sergio Rincón, Gabriel González, Nicolás Andrade, Rafael Velásquez, Sebastián Ojeda +4 more
top venuebenchmarkcode/data
Abstract · Accepted at NeurIPS 2026, Track on Evaluations and Datasets. 49 pages, 21 figures, 17 tables. Code: https://github.com/BCV-Uniandes/OPAL

Crystal Structure Prediction (CSP) aims to predict a material's 3D structure from its chemical composition, with applications from energy storage to biomedical devices. Despite its importance, existing evaluation frameworks suffer from key limitations: they rely on threshold-sensitive or incomparable metrics, focus predominantly on geometric performance while overlooking physical plausibility, and fail to account for polymorphism, where a chemical composition can crystallize into multiple stable structures. In parallel, recent Symmetry-Informed CSP (SICSP) methods often report results alongside standard CSP baselines while receiving crystallographic templates at inference, which constrain the search space and make SICSP a fundamentally different task. To address these limitations, we introduce an Orientation-invariant, Polymorph-Aware, Large Benchmark (OPAL-Bench) that evaluates CSP and SICSP as separate tasks. We evaluate 7 CSP and 2 SICSP methods under OPAL-Bench across 4 datasets using threshold-free geometric and physical distance metrics. Our results show that no method dominates across all metrics, and that geometric performance is not strongly correlated with physical plausibility. We also release OPAL-CSP, an SE(3)-equivariant flow-matching baseline for the CSP track that performs competitively on geometric metrics and ranks among the best on physical ones. By unifying geometric and physical metrics under a polymorph-aware protocol, OPAL-Bench enables a more comprehensive assessment of generative quality in CSP.

0.45

SWXAI: A Unified Multimodal Explainability Library for Space Weather Forecasting and Heliophysics

cs.LG astro-ph.SRphysics.space-ph ·2026-11-07 ·Benedict Antonio Mervyn, Anli Ji, Manolis K. Georgoulis, Berkay Aydin
top venuebenchmarkcode/data
Abstract · 6 pages, 5 figures. Accepted at CIKM 2026 (Resource Track), Rome, Italy, November 7-11, 2026. Code: https://bitbucket.org/gsudmlab/swxai Doc…

Space weather forecasting and heliophysics increasingly rely on machine learning models that process heterogeneous inputs such as multi-viewpoint multi-wavelength solar images and rasters, multivariate time series, spatial annotations, and tabular event-level data. However, explainability workflows remain fragmented across libraries, model interfaces, and output formats, limiting reuse and reproducibility. We present SWXAI, an open-source library that unifies explainable AI (XAI) workflows for image, time-series, and tabular data in space weather forecasting and heliophysics. SWXAI provides a common, extensible interface across commonly used vision, time-series, and tabular learning models, with validation, adaptation, and persistent explanation objects. It also includes domain-grounded examples, reproducible notebooks, tests, and documentation, enabling users to inspect outputs, swap models or explainers, and extend workflows to new heliophysics datasets. SWXAI reduces modality-specific engineering overhead and supports comparable, reusable explanations across heterogeneous scientific machine learning pipelines. A benchmark against direct backend implementations shows that SWXAI substantially reduces the code users write to produce an explanation, with no practical runtime cost.

0.45

FedSGN: Federated Sparse Grassmann Newton Learning for IoT Anomaly Detection

cs.LG ·2026-11-07 ·Jingjing Liu, Xiansen Ju, Xianchao Xiu, Wanquan Liu
code/dataunusual
Abstract

Anomaly detection plays a critical role in securing the Internet of things (IoT). Although existing federated principal component analysis (PCA) methods can learn the subspace of normal samples without sharing raw data, they often suffer from feature redundancy that may degrade detection performance. To address this limitation, we propose a novel and effective federated sparse Grassmann Newton (FedSGN) learning method for IoT anomaly detection. Technically, our FedSGN leverages the $\ell_{2,0}$-norm constraint on the global projection matrix to eliminate redundancy and introduces an orthogonality penalty to preserve the intrinsic structure. Moreover, we develop an augmented Lagrangian solver integrated with a semismooth Newton scheme to enable parallel client updates and Grassmann projection for robustness. Numerical experiments demonstrate that the proposed method achieves favorable overall detection performance while reducing communication overhead. The source code is publicly available at https://github.com/xianchaoxiu/FedSGN

0.43

Robust Satisficing Ensemble: Scalable Model Aggregation Under Distribution Shifts

cs.LG cs.AIstat.ML ·2026-11-07 ·Ahmet Faruk Cetinkaya, Enes Agirman, Cem Tekin
top venuetheory
Abstract · Accepted at NeurIPS 2026 (Spotlight)

Standard ensemble methods can improve robustness by mitigating individual model stochasticity, but they frequently fail to generalize beyond the training distribution, leaving systems vulnerable to out-of-distribution (OOD) data. We introduce the Robust Satisficing Ensemble (RSE), a scalable probabilistic aggregation framework designed to secure model combinations by minimizing their fragility against distributional changes. By bridging a robust satisficing objective with a quadratic risk surrogate, RSE derives an analytical inner adversary. This fundamental reformulation enables highly efficient weight updates without the computational bottleneck of standard minimax optimization, crucially decoupling optimization complexity from sample size to scale effectively to large datasets. In particular, we prove a geometrically decaying truncation error for the fragility approximation via Lagrange-Burmann expansion and establish stationarity guarantees for RSE. We evaluate RSE across a comprehensive spectrum of distribution shifts, including adversarial label shift (SST-5, TREC), natural subpopulation and temporal shifts (CivilComments, HuffPost). Our results demonstrate that RSE consistently outperforms strong robust baselines, including GroupDRO, IRM, and SRM. These findings establish RSE as a practical, theoretically grounded safeguard for reliable model aggregation in dynamic environments.

0.43

Robust and Efficient Continual Model Merging via Global Singular Subspace Separation and Restoration

cs.LG ·2026-11-07 ·Juncheol Shin, Byeongho Yu, Semin Kim, Eunhyeok Park
publishedcode/data
Abstract · NeurIPS 2026. Code: https://github.com/ewsn1593/GS3R

Continual model merging incrementally integrates task-specific models without revisiting past models or data. However, repeated merging entangles past task components and distorts their singular geometry, making balanced task-wise allocation increasingly difficult. We propose Global Singular Subspace Separation and Restoration (GS3R), which augments the merged update with a compact restoration state. At each merge, GS3R restores the retained singular geometry, separates task-wise subspaces, and globally reallocates the available rank across tasks fairly. To reduce merging-time memory overhead, we further introduce Pre-merge Restoration and Flushing (PRF). Experiments on vision, NLP, and multimodal benchmarks show that GS3R provides a strong accuracy-efficiency trade-off.

0.42

SAMBoost: Spatially Aware Mesh Boosting

cs.LG stat.MEstat.ML ·2026-11-07 ·Mulah Moriah, Franck Vermet, Pierre Ailliot, Juliette Legrand
benchmarkcode/data
Abstract

We propose SAMBoost (Spatially Aware Mesh Boosting), a gradient boosting method that combines tree-based covariate modelling with a stochastic partial differential equation (SPDE) latent spatial field through a single penalised objective. At each boosting iteration, the predictor is decomposed into separate covariate and spatial components. The method is motivated by applications in which the spatial effect is not a nuisance to be removed but a quantity of interest that must be recovered alongside the covariates. In insurance, for example, practitioners need both flexible nonlinear covariate effects and an interpretable geographic risk surface that can be reported on its own. Existing approaches often involve trade-offs, offering flexible covariate modelling but no explicit spatial field, rigorous spatial inference with a fixed additive covariate structure, or a combination of both at computational costs that limit practicality for large datasets. SAMBoost addresses these goals within the boosting loop, and the underlying mesh produces a sparse precision structure whose cost is largely independent of sample size. On simulations and on an open benchmark, SAMBoost matches or improves on the predictive accuracy of joint Gaussian process and tree boosting approaches. It is also faster and recovers the spatial field and covariate effects more accurately. When covariate effects include interactions, it outperforms additive Bayesian spatial models. Applied to a flood insurance portfolio with 4.9 million records, SAMBoost trains on the full training set without subsampling and runs about sixteen times faster than a Bayesian reference, with a consistent spatial risk surface. A public Python implementation is available.

0.42

BlenderFORGE: Framework for Optimizing Reactive 3D-Graphics Editing Ability of MLLMs

cs.AI ·2026-11-07 ·Zilin Guo, Tianrui Zhang, Yi Rong, Yichen Liu, Yuxin Guo +4 more
top venuebenchmarkcode/data
Abstract · Accepted at NeurIPS 2026. 35 pages, including references, appendix, and checklist. Project page: https://sensetime-fvg.github.io/BlenderFORG…

Multimodal Large Language Models (MLLMs) have shown promise as interactive agents, yet precise, programmatic 3D graphics editing remains difficult to train because existing systems largely rely on closed-source model APIs and lack deterministic, state-verifiable execution environments. We present BlenderFORGE, a trainable data-environment-optimization framework for open-weight MLLM agents that edit explicit Blender scenes through Blender Python (bpy) code generation. BlenderFORGE targets four fundamental state-verifiable editing task families: object placement, shape-key editing, lighting adjustment, and material modification. It integrates three components: (1) a scalable perturbation-based data generation pipeline built from curated assets and large-scale 3D scene repositories; (2) an isolated Blender Sandbox for deterministic code execution, structured state export, and closed-loop reward computation; and (3) a two-stage training pipeline that uses offline teacher trajectories for cold-start Supervised Fine-Tuning (SFT) and sandbox-computed rewards for multi-turn tool-agentic Reinforcement Learning (AgentLoop-RL). Experiments across the four task families show that the resulting open-weight Qwen3-VL-8B and Qwen3.5-9B BlenderFORGE agents substantially improve over their base models and achieve competitive performance with several zero-shot proprietary MLLM baselines on selected metrics. Additional cross-benchmark evaluations and camera-reframing experiments provide evidence of transfer beyond the original training setting.

0.41

On Generalization in Bilevel Least Squares with Overparameterized Neural Networks

cs.LG stat.ML ·2026-11-07 ·Fares El Khoury, Edouard Pauwels, Samuel Vaiter, Michael Arbel
top venue
Abstract · 73 pages, 5 figures. Accepted at NeurIPS 2026

Bilevel optimization addresses machine learning problems in which the outer objective is a value function implicitly defined through the solution of an inner problem. While its algorithmic aspects are well developed, its generalization properties remain much less understood. In this paper, we study the generalization of inexact gradient methods when both objectives are least-squares regression problems and when the estimator of the value function's gradient, or hypergradient, is obtained by implicit differentiation in function space. This approach allows us to derive stationarity bounds for these methods in terms of the uniform generalization errors of the two regression functions needed to estimate the hypergradient. When these regression functions are defined as overparameterized two-layer ReLU networks trained by gradient flow with early stopping, we obtain explicit finite-sample bounds on these generalization errors in the Neural Tangent Kernel (NTK) regime. Combining these with our general inexact gradient analysis shows the effect of sample size on the overall convergence of the bilevel method. Numerical experiments on synthetic data illustrate our results. Beyond the NTK regime, further experiments reveal an initialization-dependent implicit bias induced by the bilevel structure, which can lead to feature learning at the inner level and overall performance improvements.

0.41

Bias, Measurement Error, and Double-Dipping: When Can GNN Convolutions Help Brain Connectome Prediction?

cs.LG stat.ML ·2026-11-07 ·Tommaso Castellani, Jiaqi Li, Muriah D. Wheelock, Rezwana Rosen Razzaque, Helmut Laufs +2 more
top venuecode/data
Abstract · Accepted at NeurIPS 2026. 32 pages, 9 figures. Code: https://github.com/tommicaste/B-ME-DD

Graph Neural Networks (GNNs) are an important tool for fMRI prediction, where empirical functional connectivity matrices serve both as features and for graph construction. However, recent studies and our own experiments show that these methods often fail to outperform simpler graph-agnostic baselines. We propose a statistical explanation for this failure by formulating connectome prediction as an errors-in-variables (EIV) problem: the true connectome is latent, while the observed covariance matrix is a noisy finite-time-series estimate. In a linear asymptotic setting, we decompose the effect of graph convolution into three mechanisms: smoothing-induced bias and variance reduction, graph estimation error, and a double-dipping effect arising when the same noisy matrix determines both graph and features. Under a latent community model adapted to neuroscience, we identify a narrow favorable regime, set by the brain community structure, the predictive signal, the graph construction and the recording length, in which message passing can improve prediction. Simulations validate this decomposition, while real-data experiments illustrate that connectome GNN pipelines often fall outside this regime. Our results provide a first step towards explaining when graph convolutions can help brain connectome prediction, and why they often do not.

0.40

The forces at hand: social learning is permissive, not instructive, for the evolution of phenotypic plasticity

cs.NE q-bio.PE ·2026-11-07 ·Nam H. Le
code/data
Abstract · 21 pages, 4 figures. Code and archived outputs: https://github.com/drnamlabs-gif/forces-at-hand

Evolution can favour a trait for two reasons: an instructive cause makes selection favour the trait; a permissive one lets it persist once selection no longer removes it. The distinction matters most for phenotypic plasticity and its replacement by inherited behaviour, genetic assimilation, the Baldwin effect. Social learning is credited with favouring plasticity on evidence that cannot tell the two apart: how much plasticity remains at the end of a simulation. Here we show, analytically and computationally, which role it plays in a minimal model of learning and evolution. Writing expected fitness as a function of plasticity, we find that environmental change is the only source of benefit from plasticity, and the social channel enters only its cost. Simulation confirms this: in a static environment selection against plasticity never reverses sign at any channel strength; when the environment changes, it reverses at a threshold the channel lowers, and without the channel the population collapses before reaching it. Social learning is permissive, not instructive: it supplies the means to be plastic, never the reason. For the debate over whether plasticity leads evolution, this is one case settled by measurement: environmental change directs, socially inherited information permits, and neither suffices alone.

0.40

Audit Before You Merge: Provenance, Probing, and Continual LoRA Composition

cs.AI cs.CRcs.LG ·2026-11-07 ·Lenny Aharon, Neta Glazer, Lior Aharon
publishedcode/data
Abstract · 16 pages, 4 figures, 17 tables. ICML 2026 Workshop on Continual Adaptation at Scale (CATS). Code: https://github.com/LennyAharon/audit_conti…

Post-hoc merging of public LoRA adapters is an attractive gradient-free path to continual adaptation, but reliable measurement of catastrophic forgetting in this setting depends on infrastructure that prior work has not addressed. We treat each adapter merge as a sequential task in a continual-learning setting, with the instruction-tuned base model's capabilities as Task 0, and study forgetting on Llama-3.1-8B-Instruct. Public adapter hubs contain modules that fail basic provenance checks: silent base-version mismatches and near-zero delta norms whose inclusion inflates measured interference by ~10pp. We propose a two-check audit (base-model match, non-trivial delta norm) and a leakage-free probing protocol as preconditions for any forgetting study in this setting. On an audited 4-adapter set, naive merging causes -10.5pp on GSM8K and -18.3pp on HumanEval, while Task Arithmetic at lambda=0.5 sits within a standard error of the unmerged baseline. A coordinate search including negative coefficients identifies the math adapter as probe-optimal at lambda=-0.5, outside the reach of positive-only mergers, suggesting forgetting has structure that negation-aware merging may exploit. The same protocol replicates on Mistral-7B-Instruct-v0.3, where 2 of 4 candidate adapters fail the audit and the audited merge reproduces the same merge-method ranking on GSM8K, supporting that our findings are not Llama-specific.

0.40

CoupledFlow: One-Step Neural Operators for Coupled Multi-Physics PDEs

cs.LG cs.AI ·2026-11-07 ·Trong Khiem Tran, Minh Long Bui, Phi Le Nguyen, Mrinal Sen, Sanjay Srinivasan +2 more
top venue
Abstract · Accepted NeurIPS 2026

This paper introduces COUPLED FLOW, a new neural operator learning framework for multi-physics PDE systems that captures the co-evolution of multiple physical processes by modeling their cross-process interactions. Unlike existing neural operator methods that primarily target single-physics settings, COUPLED FLOW employs a novel structure-constrained flow matching algorithm to learn process- specific mean velocity fields conditioned on the full multi-physics state. This enables fast and accurate one-step prediction of coupled PDE systems that would otherwise require expensive numerical simulations. A key challenge in training COUPLED FLOW however lies in sampling structure-constrained conditional flow interpolants that capture cross-process correlations and serve as training examples for flow matching. To address this challenge, we derive a multi-physics mean-flow consistency identity that characterizes cross-process interactions and defines the structure-constrained mean-flow matching loss. To provide informative training samples for flow matching with this loss, we develop a principled method for sampling correlated conditional interpolants based on posterior multi-output Gaus- sian processes, which capture richer cross-process correlations than the commonly used endpoint-conditioned linear interpolants. We also establish a theoretical re- sult showing that the Wasserstein-2 distance between the predicted and oracle PDE solution distributions is upper-bounded in terms of the structure-constrained flow matching loss and the smoothness of the learned flow. Experiments across diverse coupled PDE benchmarks demonstrate that COUPLED FLOW consistently outperforms existing state-of-the-art neural operators and generative PDE solvers.

0.40

Beyond Discrete Ancestry Groups: Distribution-Aware In-Context Learning for Cancer Outcome Prediction Across the Genetic Ancestry Continuum

cs.AI stat.MLcs.LG ·2026-11-07 ·Weizhi Li, Yan Cui
code/data
Abstract

Machine learning models for omics-based prediction of cancer outcomes are commonly built on aggregated data without distinguishing ancestry groups, or under an ancestry-discrete design, which partitions samples into ancestry groups and fits a separate model within each group. Recent evidence shows that this approach is inherently limited: predictive accuracy declines with genetic distance from the training samples, even within a single ancestry group, because the data distribution shifts continuously rather than only between predefined categories. In this paper, we propose distribution-aware in-context learning (DA-ICL) for ancestry-continuum modeling. Our method extends conventional ancestry-discrete modeling to ancestry-continuum modeling, capturing the nonstationary omics-endpoint relationship across the genetic continuum while enabling efficient modeling through an in-context learning foundation model. We provide theoretical analysis demonstrating the superiority of the ancestry-continuum over ancestry-discrete modeling, along with empirical results supporting the proposed framework. Source code is available at https://github.com/ai4pm/DA-ICL-Continuum.

0.39

Coverage-Verified Sparse Attention: Closed-Loop Quality Control for Long-Context LLM Decoding

cs.LG cs.CL ·2026-11-07 ·Yan-Cheng Li, Chun-Yi Lee
top venue
Abstract · Accepted to NeurIPS 2026. 35 pages, 12 figures, 35 tables

Sparse attention accelerates long-context decoding by reading a subset of the KV cache, but existing methods are open-loop: the kernel commits to its selection with no signal indicating whether enough attention mass was preserved and no recovery path when it was not. The needed signal already exists inside the kernel. The fraction of softmax mass on the selected tokens, which we call coverage, satisfies an exact algebraic equality with per-step output error and serves a dual role as the acceptance criterion and the interpolation weight for recovery. Coverage-Verified Sparse Attention (CVSA) turns coverage into a training-free, closed-loop verify-then-correct decode loop that accepts high-coverage drafts and recovers only the heads that fall short, with no sampling, no draft model, and no speculative-decoding infrastructure. Quality varies by only $0.14$ points across the threshold sweep; budget and scheduling ablations characterize the deployed operating point. On LongBench at 7B and 70B, CVSA comes within $1.13$ points of dense attention, statistically matching it at 70B, and it loses $0.8$ points on RULER; measured median decode-step speedup reaches $6.21\times$ at 128K context.

0.39

Two-Sided Conformal Evidence Sets for Retrieval-Augmented Generation

cs.LG cs.CL ·2026-11-07 ·Ben Jenkins, Mihaela Cardei
publishedbenchmark
Abstract · Accepted at the 2026 IEEE International Conference on Tools with Artificial Intelligence (ICTAI 2026). 8 pages

Retrieval-augmented generation (RAG) can surface passages that support a claim, contradict it, or fail to address it, yet existing calibrated grounding methods are typically one-sided and model only supporting evidence. We introduce Conformal Evidence Sets (CES), a two-sided framework that returns separate support and refutation sets with finite-sample inclusion coverage of at least $1-α$ on each side under exchangeability; a Bonferroni union yields joint coverage of at least $1-2α$. On labeled candidate pools from the contrastive VitaminC benchmark, CES closely tracks nominal coverage across $α\!\in\!\{0.05,0.10,0.15,0.20,0.30\}$, with maximum deviations of $3.0$ percentage points for support and $5.4$ for refutation, and reaches essentially the same empirical coverage--size frontier as a test-tuned fixed-threshold heuristic without any test exposure. Stress tests mark the boundary of that guarantee: under FEVER distribution shift both sides under-cover, by up to $12.7$ percentage points, and support-side group-stratified recalibration substantially reduces the support gap. Scorer swaps, hard-negative retrieval, cross-dataset transfer, and a live end-to-end pilot expose a clear separation: when exchangeability holds and relevant evidence is retrieved, evidence inclusion remains calibrated even with an imperfect scorer, whereas the support/refute verdict tracks scorer quality directly. Calibrated evidence coverage and reliable verdicts are therefore distinct properties and should be reported separately.

0.39

MambaLoc: Lightweight Indoor Localization Using Cross-Modal Knowledge Distillation

cs.AI cs.NI ·2026-11-07 ·Mohab Bahnassy, Omar Saqr, Hamada Rizk, Moustafa Youssef
published
Abstract

Accurate indoor localization on resource-constrained edge devices remains challenging due to the tension between modeling fidelity and computational efficiency. Wi-Fi Channel State Information (CSI) enables low-cost deployment using existing Wi-Fi infrastructure but is heavily distorted by multipath propagation. Transformer-based solutions mitigate these effects but their quadratic complexity limits real-time edge deployment. Ultra-Wideband (UWB) provides geometrically precise ranging yet requires dedicated infrastructure. We present MambaLoc, a cross-modal distillation framework that improves the accuracy-efficiency balance in CSI-based localization. MambaLoc uses a linear-time Selective State Space Model (Mamba) as a lightweight CSI student, guided during training by a UWB-Transformer teacher. A Feature Alignment Module bridges the representational gap between heterogeneous RF modalities, allowing the CSI model to inherit UWB-derived geometric structure without UWB hardware at inference. A probabilistic Gaussian Mixture regression head models multi-modal spatial ambiguity from multipath effects, avoiding the physically implausible averaging of deterministic regression. Evaluations on a real-world dataset show MambaLoc achieves a median localization error of 0.230 m, reducing error by 22.2% over a standalone Mamba baseline, while requiring 65.1% fewer FLOPs and 58% lower inference latency than Transformer-based approaches.

0.39

GEAR: A GPU-Accelerated Global Solver for Bounded Continuous Nonlinear Programs via Linear Bound Propagation

cs.LG math.OCcs.MSmath.NA ·2026-11-07 ·Duo Zhou, Hesun Chen, Xiangru Zhong, Grani A. Hanasusanto, Huan Zhang
top venuecode/data
Abstract · Accepted by NeurIPS 2026

We present a novel GPU-accelerated global solver for constrained nonlinear programming (NLP) inspired by the efficient linear bound propagation framework in neural network (NN) verification. Existing work on NN verification, a problem that can be cast as feasibility detection for a nonlinear objective, demonstrated orders-of-magnitude speedups with the linear bound propagation-based framework when compared to traditional solvers. Extending this paradigm to global nonlinear programming requires addressing many challenges: a global constrained NLP solver must effectively exploit all, possibly unstructured, nonlinear constraints to improve the objective, whereas NN verification typically handles simple input box constraints over a well-structured NN and does not directly optimize an objective function. We address these challenges by fully exploiting the linear bounds produced by bound propagation: these linear bounds can be viewed as relaxations of nonlinear constraints and used to provide lower bounds, check infeasibility, and guide the search for a primal solution. In addition, we propose tighter linear relaxations for bilinear, quotient, and other commonly used nonlinear functions in NLP problems, and a projected augmented Lagrangian method tightly coupled to our dual solving procedure to produce primal solutions. Our method solves 421 of 505 bounded and continuous NLP problems in the GAMS and MINLPLib benchmarks under a 180-second limit, more than each of the six global solvers we compare with, including Gurobi and BARON (commercial), and SCIP (open-source). Sixteen of these problems are certified by none of the six global solvers; all of them contain exponential, logarithmic, or trigonometric terms, which are often challenging for traditional approaches. Project webpage is at https://verified-intelligence.github.io/GEAR.

0.38

Generating Diverse and Verified Synthetic Clinical Notes with Labels by Construction

cs.LG cs.AIcs.CL ·2026-11-07 ·Matthias Lehmann, Madeleine Werthebach
publishedbenchmarkcode/data
Abstract · Accepted at the GenAI for Health Workshop @ NeurIPS 2026. 38 pages including appendix. Code: https://github.com/Matt00n/phi_discharge_letter…

Under strict privacy regulations in the clinical domain, de-identification is necessary to enable the use of clinical data such as for natural language processing. However, we currently lack sophisticated ways to tell which de-identification models actually are suitable. Specifically, we need benchmarks that are challenging, diverse, and clinically plausible, all while ensuring label correctness. To this end, we present SPARCLING, an LLM-based pipeline to generate synthetic discharge letters with character-exact PHI span annotations. Our pipeline combines structured attribute sampling, template generation, and red teaming to obtain diverse PHI-labeled discharge letters. In our experiments, we demonstrate this superior content diversity over existing public datasets. Further, we show SPARCLING's downstream utility as a benchmark to uncover failure modes of de-identifiers in the German clinical domain. Finally, our experiments show that training on our synthetic data can yield more capable de-identifiers than training on the scarce public data alone. Our code is available at https://github.com/Matt00n/phi_discharge_letter_generator.

0.38

Pruning DNA Sequence

cs.LG q-bio.GN ·2026-11-07 ·Faias Satter, Shaker Islam, Aminul Islam
code/data
Abstract · 23 pages, 7 figures, 18 tables

Data pruning can substantially reduce training cost by removing examples that contribute little additional learning signal, but it has not been studied for DNA language-model pretraining. We introduce Sequence Pruning (SP), a dynamic pruning framework that removes genomic sequences as they become easy for the model to predict. SP uses per-sequence pretraining accuracy computed directly from the model's forward pass, requiring no separate scoring model or additional pass over the corpus. We evaluate SP across five pretraining settings spanning two corpora, three model families, and both MLM and CLM objectives, with downstream evaluation on five genomic benchmarks. SP reduces pretraining data processing by up to 96% while largely preserving downstream performance. At moderate savings, pruning improves performance by up to 1.2 MCC points over full-data pretraining; at the highest savings, pretraining uses only 4% of the baseline GPU hours with little performance loss. SP also outperforms the dynamic pruning methods InfoBatch and BPAS across most comparisons. Beyond accuracy, our framework supports alternative criteria such as perplexity and EL2N, although accuracy yields the strongest downstream performance at matched savings. These results show that genomic pretraining does not require repeatedly processing all available sequences and that dynamic sequence-level pruning can substantially improve its data and computational efficiency. Code is available at https://github.com/FaiasPromit/pruning-dna-sequence.

0.38

A Gamified Framework for Conflict-Driven Design: Discovering Strategies for Collaborative Resilience with Multi-Agent Learning

cs.AI cs.LGcs.MA ·2026-11-07 ·Jin Gao
publishedbenchmark
Abstract · 10 pages, 11 figures, 1 table. Author version of a paper published in the proceedings of ACADIA 2025: Computing for Resilience

Conflicting stakeholder objectives make it difficult to coordinate investments in urban resilience. We model conflict-driven urban design as a non-zero-sum digital board game in which stakeholder agents allocate scarce resources and balance individual payoffs against shared urban development and resilience objectives. We use multi-agent reinforcement learning (MARL) to train policies and examine their behavior across three settings. On a 4 x 4 board, learned policies reach about 84% of the total return found by a simulated annealing benchmark. On an 8 x 8 board with spatial and resource constraints, trained agents favor projects that improve resilience in self-play. In a 12 x 12 simulation initialized from historical maps of Tatsumi, Tokyo, trained agents achieve a mean total score of 452.1, compared with 357.3 for a random-policy baseline across 10,000 simulation runs. Gains in adaptability, sustainability, flood resistance, and greenery come with reductions in vitality and density. A human-in-the-loop interface supports replay and lets participants take over stakeholder decisions, providing a testbed for exploring these trade-offs in participatory urban planning.

0.38

Interpretable Cyber-Physical Attack Detection in Smart Grids: A Leakage-Controlled Multi-Testbed Evaluation of Fusion, Robustness, and Observability Limits

cs.LG cs.AIcs.CR ·2026-11-07 ·Muhammad Ridwanul Hoque, Sarah Sharif, Reza Saeed Kandezy, Yaser Banad
code/data
Abstract

Cyber-physical smart-grid detectors are typically treated as black boxes and validated exclusively on clean, temporally shuffled datasets, which risk overly optimistic performance claims. This work presents a multi-testbed evaluation study built around an interpretable, time-windowed fusion Random Forest, adopting the pessimistic result whenever evaluation protocols disagree. (i) On a public paired cyber-physical dataset, shuffled cross-validation is optimistic: macro-F1 falls from 0.96 to 0.88 +/- 0.06 under rolling-origin temporal evaluation. A topology-free graph convolutional network (GCN) serving as a control confirms that this lightweight model is competitive on these splits. (ii) The adaptive and adaptive-joint feature-space attacks, which are infeasible as realistic grid manipulations, degrade macro-F1 to 0.67 and 0.34, respectively. Adversarial training and randomized smoothing partially restore performance, and a physics-constrained evolution-strategy evasion on the IEEE 14-Bus testbed shows the same vulnerability pattern. (iii) On real IEC-104 traffic, the cyber view reliably catches structural attacks (recall of 0.87 to 0.98) but largely fails on semantic value manipulation (recall of 0.11 to 0.35). An observability taxonomy on the IEEE 14-Bus system reveals that stealthy false data injection defeats all three observation views. (iv) On a full AC weighted least-squares state-estimation testbed, a forecasting-aided temporal test is evaded by slow ramps precisely where a rate-invariant physics-manifold test sustains detection at a common 5% false-alarm target. A full end-to-end evaluation on IEEE 118-Bus confirms the ordering but bounds its strength: the manifold test detects every trial down to 0.05 degree/step and 0.48 at 0.02 degree/step, against 0.14 for the temporal test. All code and benchmarks are released.

0.38

Stochastic Port-Hamiltonian Gaussian Processes for Learning from Noisy Dynamics on Lie Groups

cs.LG ·2026-11-07 ·Thribhuvan Rapolu, Luke McLennan, Chandrajit Bajaj
code/data
Abstract · 54 pages, 14 figures

Learning predictive models for dynamics is difficult because observations are often sparse and noisy, while the underlying dynamics evolve on nonlinear configuration manifolds and are affected by environmental dissipation and stochastic disturbances. Standard neural dynamics models can fit short-term trajectories but often violate physical structure, accumulate error over long horizons, or rely on parameterizations that introduce singularities such as gimbal lock. We introduce LieSPHGP, a stochastic Gaussian-process framework for learning rigid body dynamics with port-Hamiltonian structure and stability-aware robustness. LieSPHGP represents motion on Lie groups and Lie algebras, enabling coordinate-consistent optimization without singular parameterizations. Learned mass, potential energy, dissipation, and control components retain physical roles and a separate diffusion term represents stochastic forcing. Each component is modeled by Matérn kernel Gaussian processes which capture uncertainty induced by noisy observations. To improve physical validity, Lie group exponential map updates preserve rotational constraints, while positive-definite damping ensures nonnegative deterministic energy dissipation. The implicit-explicit integrator additionally provides unconditional stability for the frozen linear damping subsystem. The resulting model learns predictive dynamics that are robust to noise, respect rigid body geometry, and capture both intrinsic motion and environmental dissipation. Experiments on various noisy rigid body benchmarks show improved prediction accuracy, stronger long-horizon stability, and better satisfaction of manifold constraints compared with neural dynamics baselines. Code, data and interactive results are available on the project website: https://thribhuvanrapolu.github.io/liesphgp/

0.37

No Second Run: Does RSI Stop at Frontier Pre-training?

cs.AI ·2026-11-07 ·Xiaojie Cai, Jiayi Chen, Junwei Chen, Yu Chen, Sikai Cheng +54 more
code/dataunusual
Abstract

Pre-training is where every foundation model begins; bringing recursive self-improvement (RSI) there would let AI improve the process that creates AI. Yet public RSI work has so far targeted only post-training, which starts from an existing base model, runs trials costing GPU-hours, and returns verifiable scores. Pre-training breaks all three conditions at orders-of-magnitude higher cost: a run consumes hundreds of thousands of GPU-hours over months and cannot be repeated; its objective, post-training headroom, cannot be read off any single score; and what must improve is not a model but the whole production system, from data pipelines and training infrastructure to the teams that run them. To our knowledge, we are the first to apply RSI to an entire frontier pre-training run from scratch and to report it in detail. RSI does not stop at frontier pre-training, but it changes form: since no end-to-end loop can close over a single unrepeatable run, we decompose it into smaller loops that can, each repairing a step the full run breaks. The result is 801-30B-A3B-Base, a 30B-A3B MoE trained from scratch on 11.8T tokens in three months by a newly formed team. We release it as a base model; the lightweight SFT reported here (6B tokens, no RL) only probes post-training headroom. As a base model, 801-30B-A3B outperforms Qwen3-30B-A3B and Nemotron-3.5-Lightning-30B-A3B, trailing only the larger Qwen3.5-35B-A3B, and keeps this position under identical SFT; with lightweight SFT alone, it achieves the highest average on four agentic benchmarks against fully post-trained models, including Qwen3.6-35B-A3B. Since post-training RSI presupposes a strong base model, bringing RSI into pre-training acts on the foundation every later stage builds on. We open-source 801-30B-A3B-Base and the lightweight SFT data at https://github.com/GAIR-NLP/801-30B .

0.37

Monitoring One Channel Is Not Communication Control: Causal Evidence for Adaptive Reallocation of Joint Message--Route Codes

cs.AI cs.IT ·2026-11-07 ·Liu Hung Ming, Tian Xia
code/data
Abstract · 23 pages total: 8-page main paper plus 15-page supplementary material; 1 main-paper figure and 3 main-paper tables. Equal contribution. Code…

Monitoring a visible channel can change how learned agents communicate without removing why communication is useful. We test this distinction in an end-to-end game where a sender and receiver learn a discrete protocol and route policy without communication supervision. In two independent 20-seed batches, 20/20 and 19/20 protocols form. Online partial coverage increases shadow-route share over no monitoring by .701 [.574,.826] and .575 [.412,.737], while task accuracy remains .905/.913 and signed frozen message contribution remains .592/.563. Complete coverage removes most selective migration without removing communication. Frozen interventions, including a fresh pre-outcome confirmation, show that route is semantic: joint message--route shuffling changes .668 of receiver predictions and route alone retains .388 accuracy versus .25 chance. A corrected explicit-silence cost makes 18/20 policies fully silent and lowers mean sending to .050, but its strict complete-suppression gate fails because two seeds retain communication. Failed semantic-evasion and operational gates further bound the result. The evidence supports adaptive reallocation of a joint message--route code under partial monitoring, not open-world route discovery, emergent encryption, or a universal defense.

0.37

Elastic Representations via Hyperbolic Geometry

cs.LG cs.CV ·2026-11-07 ·Arjun Ramesh Kaushik, Rudrasis Chakraborty, Nalini Ratha, Venu Govindaraju
top venuecode/data
Abstract · Accepted at NeurIPS 2026

Learned representations (or embeddings) are the foundation of modern machine learning systems, yet they are typically trained as fixed-size embeddings, without accounting for varying downstream resource or task constraints. Matryoshka Representation Learning (MRL) addresses this limitation by learning nested representations, but requires costly end-to-end retraining and does not support post hoc expansion, while Contrastive Sparse Representations (CSR) reduce training overhead via lightweight adaptors but operate at a 4x larger dimensionality. Additionally, all prior works restrict their framework towards compression only. In this work, we introduce Hyperbolic Elastic Representation Learning (HERL), a framework for learning dimension-adaptive embeddings (compression and expansion) through a lightweight adaptor without retraining the base encoder. Our key idea is to exploit the exponential capacity of hyperbolic geometry by learning a non-linear mapping from the encoder space to hyperbolic space. We train a simple MLP to downsample the representations before projecting them onto a Lorentzian and learn adaptive representations that respect the Lorentzian geometry. Our approach is model-agnostic and effective in both supervised and unsupervised settings. Through extensive experiments across image and text modalities, we show that HERL preserves downstream performance across a wide range of embedding sizes, exhibits smooth interpolation between dimensions, and achieves these benefits at a fraction of the computational cost. Code: https://github.com/arjunrkaushik/herl.

0.36

PatternBloom: Empowering Agentic RAG with Externalized RL-Distilled Graph Patterns

cs.AI ·2026-11-07 ·Yi Ge, Heng Huang
top venuecode/data
Abstract · Accepted at NeurIPS 2026. 46 pages. Code: https://github.com/EllenYiGe/PatternBloom

Reinforcement-learning-trained agents for retrieval-augmented generation retrieve content (passages, knowledge-graph fragments, or hyperedges) and supervise training with outcome rewards on the final answer. Across these paradigms the reasoning route through retrieved content never leaves the language model's hidden state, conflating declarative knowledge with the procedural knowledge that recurs across queries of the same logical type; content-only methods consequently plateau as reasoning depth grows. We introduce PatternBloom, an agentic RAG framework that shifts the unit of retrieval from content (what to read) to logic (how to reason). A first RL stage scores the agent's constructed evidence graph by a frozen oracle's information gain through the Information-Density Reward (IDR), a size-normalized mutual-information surrogate. High-reward trajectories are distilled into a Graph Pattern Memory (GPM) of type-abstracted reasoning skeletons, and a second RL stage uses the Pattern-Augmented Reward (PAR) to couple policy to memory, turning procedural reasoning into a learnable, externalized structural prior. A 7B PatternBloom outperforms $27$ baselines by $+15.0$ Avg-OOD EM, with the gain compounding with reasoning depth precisely because procedural knowledge, once distilled, is reused rather than re-derived at every query. We release our code at https://github.com/EllenYiGe/PatternBloom.

0.36

Causal Heteroscedastic Structure Learning From Incomplete Temporal data

cs.LG cs.AI ·2026-11-07 ·Abdellah Rahmani, Pascal Frossard
top venuetheory
Abstract

Discovering causal structures in multivariate time series (MTS) is critical in domains such as finance and neuroscience, yet remains challenging in practice due to three distinct obstacles. First, MTS are routinely corrupted by missing values, where naive impute-then-discover pipelines distort the underlying data distribution and yield spurious causal graphs. Second, identifying instantaneous effects is difficult even on complete data, and missingness further entangles imputation with graph recovery. Third, real-world data frequently exhibits heteroscedastic noise with variance driven by both instantaneous and lagged causes, violating the core assumptions of standard discovery methods. Existing approaches fail to address these challenges concurrently, typically presuming complete data, homoscedastic noise, or strictly time-lagged relationships. Crucially, we show through ablation studies that resolving these issues sequentially, rather than jointly, leads to a substantial degradation in performance. To overcome this, we introduce Cheesefill, an optimal transport framework for causal structure learning under missingness and heteroscedastic noise. Cheesefill parameterizes causal mechanisms using conditional normalizing flows and jointly learns a stochastic correction map that refines naive imputations into trajectories consistent with the inferred causal mechanisms. Theoretically, we prove that optimizing over stochastic correction maps is equivalent to a conditional kernel reformulation of the Kantorovich problem, from which our neural parameterization yields a tractable objective. Extensive experiments on synthetic and real-world benchmarks demonstrate that Cheesefill consistently outperforms existing baselines across diverse missingness regimes.

0.36

MACaQ: Fast, Model Agnostic Calibrated Quantization for Low-Precision Weights and Activations

cs.LG ·2026-11-07 ·Lorenz K. Muller, Philippe Bich, Chiara Boretti, Hyun-Min Chang, Jiawei Zhuang +1 more
code/datacommunity pickpopular code
Abstract

Weight-activation quantization can reduce the memory and compute cost of neural-network inference, but recent methods often require expensive optimization and architecture-specific propagation of basis transforms across layers. We introduce MACaQ, an MXFP4 W4A4 post-training quantization method that treats every linear layer independently. MACaQ applies a per-channel scale and a block-Hadamard transform online to each layer's activations, while absorbing the corresponding inverse transforms into its weights. Retaining these transforms online avoids cross-layer dependencies and enables a simple streaming implementation of the quantization process, in which only one layer must reside in memory at a time. To improve over round-to-nearest quantization, we introduce Flip-Refinement, a Hessian-aware coordinate-descent procedure that greedily moves individual FP4 codes to adjacent representable values. Despite executing the activation transforms at inference time, MACaQ adds only 2.3% end-to-end overhead over pure MXFP4 on Qwen3-32B in vLLM. MACaQ reaches lower perplexity than DuQuant++ and MR-GPTQ, and already matches their best perplexity on Llama-3-8B in 11 times less quantization time. We further apply MACaQ without architecture-specific modifications to an image diffusion model, illustrating the generality of its layer-local design and show quantization of a 122B parameter model in 64 minutes, using only 32.5GB GPU RAM. Code is available at https://github.com/huawei-csl/MACaQ .

0.36

Trustworthy Recursive Self-Improvement with Statistical Guarantees

cs.AI stat.MLcs.LGstat.ME ·2026-11-07 ·Zihe Gao, Zhengling Qi, Yifan Cui
theory
Abstract · 44 pages, including appendix

Recursive self-improvement offers a promising path toward AI agents that improve through experience by refining their own harnesses without retraining their underlying large language models (LLMs). Making this self-improvement process reliable, however, requires more than generating new updates: agents must decide which changes are worth keeping. A greedy gate that accepts updates based solely on higher validation scores can mistake an observed gain for genuine progress and lead to performance degeneration, whereas overly conservative statistical gates may reject beneficial updates and hinder performance improvement. We introduce Trustworthy Recursive Updates with Statistical Testing (TRUST), a framework that adapts an online multiple hypothesis testing procedure to control the rate of false evolution across rounds, with an LLM analyzing historical context to help determine how much evidence is required to update the incumbent. Under interpretable assumptions, we prove that TRUST limits the proportion of accepted updates that result in unacceptable performance loss. We further derive a lower bound on the final performance gain. Across the benchmarks HotpotQA, DROP, and APPS with the models DeepSeek-V4-Flash and Qwen-3.8-Flash, TRUST reduces the average ratio of accepted inferior updates from the greedy gate's 45.59% to 2.78%, while achieving higher final scores than the state-of-the-art statistical gate baselines for recursive self-improvement. These findings demonstrate that TRUST strikes a balance between statistical reliability and performance gain in recursive self-improvement.

0.36

The Surprising Effectiveness of Deleting Weights in LLM Post-Training

cs.LG cs.CL ·2026-11-07 ·Jack Lu, Zhenbang Yang, Mike Lasby, Tejas Pote, Yani Ioannou +1 more
publishedpopular code
Abstract · NeurIPS 2026

How little of a pretrained large language model (LLM) has to change for it to acquire new capabilities? We find that zeroing fewer than 0.05% of its weights, with no other modification, matches full fine-tuning (FFT) at scale across three LLM post-training regimes: reinforcement learning with verifiable rewards, supervised fine-tuning, and on-policy distillation. Our method, Deletion Tuning, learns a binary mask over the gate projection of transformer blocks. Two practical advantages follow: the trained mask is hundreds to thousands of times smaller than the corresponding FFT checkpoint, and Deletion Tuning forgets far less, which addresses a central challenge in continual learning. Under reinforcement learning, its prior-task accuracy stays near the base level while FFT, LoRA, and random sparse fine-tuning lose up to 5.4 points. To understand why masking alone suffices, we analyze the geometry of the learned weight changes under reinforcement learning using stable rank, spectrum drift, and principal-weight overlap. Compared with LoRA and random sparse fine-tuning, Deletion Tuning is closest to FFT across the three measures.

0.35

Toward universal excited-state machine-learned potentials: a local excitation Hamiltonian for size consistency and transferability

cs.LG physics.chem-phcond-mat.mtrl-sci ·2026-11-07 ·Pascal Friederich
code/data
Abstract · 24 pages, 6 figures, 3 tables, Supplementary Information (19 pages) included. Dataset: https://doi.org/10.5281/zenodo.24042143. Code and tra…

Universal machine-learned interatomic potentials could make excited-state simulations across chemical space routine, but excitation energies, unlike ground-state energies, are not sums of atomic contributions. The excited states of noninteracting molecules are the union of their states, which pooled multistate readouts reproduce at best for the lowest state. We introduce LEH-MACE, whose excitation energies are the eigenvalues of a local excitation Hamiltonian assembled from atom-centred blocks, so this union holds exactly. We generated TD-DFT energies and forces for about 300,000 equilibrium, thermal, and excited-state geometries of over 100,000 molecules. LEH-MACE's advantage over direct readouts grew with the distance from the training data and was largest above $S_1$, with about half their $S_2$ and $S_3$ errors for long conjugated oligomers. For molecule pairs, only LEH-MACE kept its single-molecule accuracy, while a softmin readout remained superior for $S_1$ dynamics of larger chromophores. Exact size consistency is thus one step toward such potentials.

0.35

Applied bifidelity material response model construction and uncertainty quantification with a sampling hypernetwork

cs.LG ·2026-11-07 ·Reese E. Jones, Cosmin Safta, Dan Bolintineau, Craig M. Hamel, Gianluca Geraci +1 more
Abstract · 29 pages, 16 figures

Multifidelity modeling offers a practical route to accurate prediction and uncertainty quantification when high-fidelity data are scarce but lower-fidelity data are comparatively abundant. This work develops a bifidelity Langevin sampling hypernetwork framework that constructs probabilistic neu- ral response models by learning a posterior distribution over model parameters rather than a single deterministic fit. The method combines a flexible shared-latent data representation for low- and high-fidelity responses, enabling parameter sharing and transfer learning between the fidelities. We introduce several methodological enhancements to improve the robustness and efficiency of this ap- proach, including stability-promoting drift models, staged and conditional training strategies, adap- tive initialization of Langevin trajectories, and training procedures designed to better exploit corre- lations between fidelity levels. These additions improve convergence behavior, reduce the sampling effort required to obtain representative posterior ensembles, and provide more reliable predictive uncertainty in data-limited regimes. The resulting framework makes more effective use of available multifidelity data while preserving interpretability through the learned posterior structure, yielding improved predictive performance, better uncertainty characterization, and a more stable workflow for constructing uncertainty-aware surrogate models. The method is demonstrated on experimental, high-fidelity simulation and illustrative synthetic data.

0.34

SituRecBench: Evaluating Frontier Models as Situated Recommendation Assistants in 3D Environments

cs.AI cs.CLcs.CVcs.IR+1 ·2026-11-07 ·Jingtong Yang, Dongding Lin, Jian Wang, Wenjie Li
top venuebenchmark
Abstract · Accepted by NeurIPS 2026 Evaluations and Datasets Track

Recommender systems are transitioning from static retrieval engines into interactive assistants grounded in physical environments. However, current studies largely neglect the combined complexities of spatial navigation, visual grounding, and multi-turn social interaction. To bridge this gap, we introduce SituRecBench, an interactive benchmark for evaluating frontier vision-language models (VLMs) as situated recommendation assistants within high-fidelity 3D environments. We reformulate recommendation as a situated assistance process in which an assistant interacts with a user while sharing the user's egocentric view. Our benchmark simulates real-world shopping environments (e.g., supermarkets and home furnishings stores), requiring assistants to elicit preferences, ground recommendations in the environment, and guide the user to relevant locations through dialogue. The benchmark evaluates dialogue quality, recommendation utility, and navigation effectiveness through a unified interactive protocol. Extensive experiments reveal persistent challenges in spatial grounding and preference elicitation, highlighting limitations in their ability to integrate global contextual information with local observations. SituRecBench provides a rigorous foundation for advancing the development of situated assistants in the real world.

0.34

Spectrogram-Based CNN Autoencoder for One-Class Anomaly Detection in Machining Vibration Signals

cs.LG ·2026-11-07 ·Sebastian Zeleny, Christian Gawron
published
Abstract · 29 pages, 14 figures

This work presents a one-class anomaly detection framework for vibration-based condition monitoring in machining processes using a convolutional neural network autoencoder. Three-axis vibration signals acquired during industrial broaching are transformed into time--frequency spectrograms and combined into a compact three-channel representation, enabling convolutional feature learning on multivariate time-series data. The autoencoder is trained exclusively on normal process data and detects anomalous strokes based on reconstruction errors. Labeled validation data are used for threshold calibration and an independent test set is reserved for final performance evaluation. To improve sensitivity to short-lived process instabilities such as chatter, a localized residual-based metric termed Segmented Vertical Error (SVE) is introduced. Experimental results on approximately 10{,}000 broaching strokes show a clear separation between normal and anomalous process conditions under strong class imbalance. The processing pipeline is additionally evaluated on an external public turning dataset as a proof of concept, indicating preliminary applicability beyond the original broaching setup. The results suggest that spectrogram-based autoencoder models can provide a practical and scalable basis for one-class anomaly detection in industrial vibration monitoring.

0.34

The Lagrangian Brain: A Constraint Mechanics of Mind and Machine

cs.AI ·2026-11-07 ·Yuhang He, Si Wu
Abstract · 58 pages, including appendices

A brain is in one state at a time, yet most of what decides its next state is absent from it: the patterns its world never produces, the statistics of a lifetime, a goal that lies ahead. In mechanics a wire acts on a bead only through a force that appears when the bead presses on it, and Lagrange computed such forces with one multiplier for each constraint. We propose that a brain follows the same mechanics. Everything a circuit knows, is told or wants is a constraint on its single state and acts through its own multiplier, which accumulates the violation and pushes the state back. Neurons and synapses carry the multipliers, each with its own time constant: inhibitory interneurons carry what never happens, adaptation the cost of staying in one state, synapses the statistics of experience, and feedback to apical dendrites the future. From this hypothesis follow Dale's law, the reciprocal wiring and balance of excitation and inhibition, the paradoxical responses of cortex and the frequencies of brain rhythms. Belief becomes the time one trajectory spends in each state, learning becomes sleep proposing laws and waking refuting the false ones, and planning becomes a wave from the goal meeting a wave from the present. Connectionism and symbolism are the limits of frozen and of instantaneous multipliers, and between them the time a circuit runs becomes the depth of its computation. A machine built on these principles, trained without gradients on 1000 solved puzzles, solves all 422,786 Sudoku-Extreme test puzzles with one trajectory each. It rediscovers the rules of Sudoku by sleeping and waking, plans shortest paths in mazes of 64 times the training area, and tracks permutations exactly on strings 25 times longer than in training, where Transformers fail. A mind, on this view, is life that learns its own constraints.

0.34

A Geometric Theory for Nonequilibrium Transformer Dynamics

cs.LG math-phmath.PRnlin.AO+1 ·2026-11-07 ·Aythami Bethencourt de León, Darryl D. Holm
theory
Abstract · 29 pages Submitted to Proc Roy Soc A

We develop a variational and geometric theory based on statistical physics for the joint evolution of Transformer attention and token representations at fixed model parameters. To this end, we introduce a free energy for the attention and model the evolution of the attention weights by its Fisher--Rao gradient flow. At fixed token representations, we prove that the softmax matrix is the unique free-energy minimiser and that every positive attention trajectory converges to it. For prescribed time-dependent representations, we show that the attention dynamics becomes linear after a change to logit variables. This result yields tracking bounds relative to the instantaneous softmax and sufficient conditions for attention to converge to a fixed matrix. Hamiltonian dynamics for the representations and their momenta are then coupled to the attention relaxation driven by the same energy. This coupling results in an energy-dissipation law. We also develop two alternative Hamiltonians incorporating the attention value map. We prove global well-posedness of the attention flow and all three coupled systems.

0.34

The Delta Gap: Cellular World Models for Perturbation Response Prediction Are Far Harder Than Global Correlation Implies

cs.AI cs.LG ·2026-11-07 ·Junhui Gao
benchmarktheoryhot direction
Abstract · 35 pages, 16 figures, 17 tables

Cellular world models aim to predict how cells respond to genetic perturbations, analogous to world models in reinforcement learning. Perturbation response models are evaluated almost exclusively by global Pearson correlation between predicted and observed transcriptomes. We show this metric is dominated by thousands of non-responsive genes and severely inflates apparent performance. On the Replogle K562 screen (509 unseen perturbations, 5000 genes), GEARS achieves global Pearson 0.977, but its delta Pearson is only 0.076. A trivial baseline predicting mean training delta achieves 0.267, which is 3.5x better than GEARS. Our primary contribution is methodological: we identify and quantify this delta gap as a systematic evaluation artifact, and propose delta-space evaluation as the primary benchmark. We provide two post-hoc corrections (RWR on co-expression graphs and kNN matching) and show that 2.8x of the remaining gap lies in perturbation representation geometry, not decoder capacity. We further show the gap is partially addressable end-to-end: delta-aligned training (lambda = 3.0) raises delta Pearson to 0.179 on Replogle (2.35x over GEARS) and surpasses the trivial baseline on direction accuracy (0.680 vs 0.567) and gene recovery (1.52 vs 1.19); combined with kNN it reaches 0.217. An amplitude-calibration variant (eta = 0.3) further raises gene recovery to 2.16 while reaching 0.218. Critically, these representations transfer zero-shot across cell lines: from Replogle K562 to unseen Norman A549, the delta-aligned model reaches 0.328 on 741 shared genes, 4.7x a trivial baseline and 2.1x standard GEARS. This is the first systematic quantification of the delta gap, and the first diagnostic framework localizing the residual gap to perturbation representation geometry via an oracle upper bound.

0.34

STABLE: A Continual Learning Optimizer with Adaptive Drift Control

cs.LG cs.AIcs.CV ·2026-11-07 ·Xiaoyan Li, Lanpei Li, Massimo Coppola, Vincenzo Lomonaco, Antonio Carta
top venuepositioncode/data
Abstract · Accepted at the 40th Conference on Neural Information Processing Systems (NeurIPS 2026)

Continual learning suffers from catastrophic forgetting, leading to the loss of past knowledge and unstable adaptation across task streams. Existing approaches typically address this issue through replay buffers, architectural changes, or additional regularization terms. We argue that forgetting is a consequence of unstable learning dynamics, which may benefit from an optimizer-level solution. The main challenge is that learning dynamics are time-dependent, as demonstrated by empirical phenomena such as the stability gap. First, we formally prove that, for distance-based classifiers, bounded parameter drift guarantees bounded margin erosion under a local score-stability condition. Based on this insight, we propose STABLE, a primal-dual optimizer with adaptive drift control. STABLE enforces a drift budget by increasing the dual variable when the budget is violated and decreasing it otherwise. This results in high stability without penalizing plasticity. Experiments on long-horizon unimodal streams and CLIP-based multimodal streams show that vanilla PEFT baselines trained with STABLE match or outperform specialized continual learning methods. In the multimodal setting, STABLE further preserves zero-shot transfer, matching or exceeding dedicated multimodal continual learning methods. Code is available at https://github.com/atlas-luiss/STABLE.

0.33

Unified Factored Representations Resist Catastrophic Forgetting

cs.LG cs.AI ·2026-11-07 ·Joel Lehman, Kenneth O. Stanley
Abstract

This paper explores an unexpected connection between the quality of a neural network's (NN's) internal representation and its vulnerability to catastrophic forgetting. The representational forgetting hypothesis is that networks with unified, factored representation (UFR) can resist catastrophic forgetting without any modification to the underlying fine-tuning algorithm or network architecture. The key insight is that catastrophic forgetting may largely result from fractured entangled representation (FER): fine-tuning on only part of the input domain results in disruptive changes for other input patterns, because of how individual features are globally entangled. In contrast, if a gradient update is compatible with how a UFR network factors its representation, such an update may result in coordinated change across the input domain, alleviating catastrophic forgetting. Experiments in two domains provide evidence for this hypothesis: (1) an illustrative 1D regression task, in which UFR networks constructed by hand are compared against networks trained from scratch with SGD, and (2) an image-regression task with NNs previously shown to exhibit UFR. In both settings, UFR networks forget substantially less than their FER counterparts when repeatedly fine-tuned on UFR-compatible variations. Further experiments show that a training loss that distills UFR representations into fresh NNs reduces forgetting; along with several additional controls, these results help isolate representation as the critical factor. Together these results suggest that representational quality is an underappreciated contributor to catastrophic forgetting, and that the encouragement and maintenance of such quality may require attending to what patterns a particular NN is ready to learn.

0.33

Proximal Preference Optimization: A Unified Framework for Stable and Reward-Free Alignment

cs.LG cs.AI ·2026-11-07 ·Qisai Liu, Zhanhong Jiang, Zahid Hasan, Soumik Sarkar
theorycode/data
Abstract · 10 pages, 4 figures, IEEE BigData 2026

We introduce a proximal preference optimization framework that unifies the implicit reward formulation of Direct Preference Optimization (DPO) with the trust-region stability of Proximal Policy Optimization (PPO). Our framework yields two complementary algorithms. P$^2$O (Proximal Preference Optimization) introduces a clipped importance ratio directly into the DPO preference margin, producing a reward-model-free algorithm that supports safe dataset reuse, bounded policy divergence, and monotonic preference improvement guarantees. PKTO (Proximal Kahneman-Tversky Optimization) extends P$^2$O by replacing the DPO margin with KTO's prospect-theoretic utility margins, adding a self-regulating adaptive baseline that tightens the effective trust region as the policy drifts, and an asymmetric weighting that handles imbalanced or asymmetric preference datasets. We provide a unified theoretical analysis, including Kullback Leibler (KL) trust-region bounds, a policy-improvement lower bound, and convergence guarantees under multi-epoch proximal updates. Empirically, P$^2$O and PKTO match or exceed DPO and KTO in alignment quality while maintaining better stability under repeated dataset reuse across multiple LLMs on three preference benchmarks. The code base is available here.

0.33

NeuroHorizon: Long-Horizon Forecasting of Neural Population Activity with Hierarchical Memory

cs.LG q-bio.NC ·2026-11-07 ·Peng Jiang, Xiaoxuan Jia
top venue
Abstract · Accepted at NeurIPS 2026 (Main Track). 25 pages, including references, appendices, and the NeurIPS paper checklist

Long-horizon forecasting of neural population activity tests whether a model has captured dynamics that remain predictive well beyond the observed history. Prediction quality typically declines as the horizon grows, yet neither short-horizon accuracy nor teacher-forced evaluation, which supplies the true activity preceding each predicted step, reveals how well models sustain predictions from an initial history alone. We compare direct multi-step prediction and autoregressive rollout on spiking recordings across horizons and introduce NeuroHorizon, which pairs event-based spike encoding with a causal decoder for population firing rates. Its hierarchical prediction memory keeps recent predictions at full temporal resolution and compresses earlier ones into segment summaries, giving the decoder both local detail and longer-range context. Under teacher-forced training this memory holds ground-truth activity, whereas during rollout it holds the model's own predictions. To reduce this mismatch, we train with scheduled sampling, progressively exposing the decoder to its own predictions. From 500 ms of macaque motor-cortex activity, NeuroHorizon achieves the highest forward-prediction score among the evaluated autonomous models at all four horizons from 100 to 1000 ms, and its per-bin predictions remain above a mean-firing-rate baseline across nearly the full 1 s window. Complementary experiments on macaque sensorimotor and mouse multi-region recordings show that performance generally increases with model and data size, alongside improved prediction after cross-subject pretraining and fine-tuning. Together, these results clarify how forecasting performance changes across horizons and offer an effective approach to long-horizon forecasting of neural population activity.

0.32

Learning Biological Hierarchies in Single-Cell Foundation Models

cs.LG cs.AIq-bio.QM ·2026-11-07 ·Xiangyu Guo, Ricardo Henao
top venue
Abstract · Accepted to NeurIPS 2026

Cell types are organized by taxonomic relationships that define lineages in the so-called cell ontology. However, most existing foundation models typically ignore cell-type lineages encoded in this cell ontology. We introduce a biologically informed foundation model called scOntoFM, which embeds hierarchical cell ontology into representation learning. Using Lowest Common Ancestor (LCA) distances from the ontology graph, our framework pairs efficient offline triplet sampling with hierarchical cell-ontology learning. This approach jointly preserves local neighborhoods and shapes the global embedding geometry to reflect the ontology structure. Across diverse cell- and gene-level benchmarks, the model consistently improves performance, with especially strong gains in zero-shot settings. Our model also balances batch integration with biological structure preservation, providing complementary perspectives to existing embedding evaluation frameworks and supporting scalable biological discovery.

0.32

Confronting Uncertainty in Mechanism-AI Coupled Dynamical Systems via Optimal Transport and Information Geometry: Part I. Distribution Learning

cs.LG math.DSmath.MGmath.PR+1 ·2026-11-07 ·Pengfei Song, Yanni Xiao, Shuangshuang Yin
hot direction
Abstract

Mechanism-AI coupled dynamical systems are increasingly used to study complex biological systems. However, uncertainty associated with noisy, sparse, and partially observed data limits their reliability. This paper presents a systematic mathematical formulation for treating these uncertainties from the perspective of optimal transport and information geometry. We formulate the uncertainty quantification problem as the minimization of an energy functional on a probability density space equipped with a Riemannian metric, with the loss evaluated through a differential equation, and give a unified description of the gradient flows of the energy functional under specific metrics, such as the Wasserstein and Fisher-Rao metrics. Furthermore, we present corresponding numerical methods for solving these gradient flows using abstract adjoint sensitivity analysis. The formulation connects a range of Bayesian inference methods for dynamical systems; numerical experiments compare posterior approximation and predictive uncertainty under a common computational budget. Moreover, it also provides a geometric perspective on dynamical modeling and the study of heterogeneity evolution.

0.32

CHARM: Certified Homology-Aware Reverse-scale Mapping via Diffusion-Guided Quotient Towers

cs.LG ·2026-11-07 ·Zhe Li
benchmark
Abstract

Conventional dimensionality-reduction methods optimize geometric fidelity but do not return auditable homology certificates or identify the input observations associated with nontrivial classes. Direct high-dimensional persistent-homology computation is costly, and its representatives are rarely linked to a shared planar view. We introduce Certified Homology-Aware Reverse-scale Mapping (CHARM), which augments two-dimensional coordinates with algebraically certified, sample-traceable structural records. Diffusion-guided cover proposals and homotopy-safe contractions compress the input neighborhood flag complex. A Cover Quotient Tower supports cross-scale representative tracking and exact $H_k$ certification when algebraic tests complete. Reverse-scale layout connects quotient-chain preimages to the same coordinates used for geometric exploration. Contraction preserves the input neighborhood flag-complex homotopy type; the quotient tower inherits a multiplicative 3-interleaving with coarse-hop Rips, and conditional distortion bounds connect it to original Euclidean Rips. In a fixed-seed benchmark of 22 datasets (41,637 samples), CHARM achieves the highest share of first- or second-place scores across 15 metrics. On nonsynthetic data, it ranks first or second on every metric and first on all four $H_0/H_1$ persistence-diagram metrics. CHARM couples strong nonsynthetic geometric performance with an auditable structural layer.

0.32

Too Big to Inspect? Ethically Auditing Large Datasets Through Small Samples

cs.AI cs.DB ·2026-11-07 ·Giovana Marchetti, Jhessica Silva, Tiago Torrent, Carlos Caetano, Sandra Avila
Abstract · 12 pages, 6 figures

Big techs and researchers increasingly claim that their training datasets are too large to inspect for bias and ethical problems. However, in the race to build ever-large models, they have relied on automated web scraping and no human curation. Datasets built this way absorb the stereotypes, exclusionary norms, and historical prejudices of their sources, and the models trained on them can reproduce and amplify those patterns. Because manually auditing billions of samples is humanly impossible, we need practical, effective methods to identify biases in these datasets. This paper studies the feasibility of ethically auditing large datasets through small samples and proposes a practical method for identifying structural biases by applying AI Ethics Tools (AIETs) to random samples. We applied Datasheets for Datasets, Data Statements, Data Nutrition Labels, and Data Cards to a 1% sample of Framed Multi30K, and Datasheets for Datasets to two 0.00006% samples of relaion2B-en-research-safe. Our results show that evaluating small samples with AIETs can reveal ethical problems, structural biases, and technical inconsistencies. Although industry practice assumes that identifying problems requires analyzing the entire dataset, we show that a random subset supports a viable qualitative analysis without overwhelming computational or human resources. Demonstrating that a dataset is biased does not require examining every instance: if a bias appears in a random sample, it is a flaw of the creation process. The AIETs' questions acted as provocateurs, evoking different perspectives and analysis methods for the same dataset, but the analysis and its results depended fundamentally on the human factor. We hope these findings contribute to urgent public discussions about ethics, transparency, accountability, and AI regulation by offering a practical pathway for responsibly auditing massive datasets.

0.31

Bayesian Low-Rank Posteriors for Scalable Membership Inference

cs.LG ·2026-11-07 ·Khanh Nguyen, Antonio Emanuele Cinà, Andrey Barsky, Alejandra Reinares, Fabio Roli +1 more
published
Abstract · NeurIPS 2026

Membership inference attacks (MIAs) aim to determine whether a sample was used during the training of a target model. Strong MIAs rely on reference (or shadow) models to estimate target-conditional score distributions, but training such ensembles is computationally expensive for modern large-scale Vision--Language Models (VLMs). In this work, we propose a scalable alternative that replaces explicit reference-model training with a Bayesian approximation over low-rank adaptation parameters. Specifically, we model uncertainty only within the LoRA subspace and construct a posterior from a single training run, from which we sample a diverse set of \emph{virtual reference models}. We instantiate this approach using Stochastic Weight Averaging Gaussian (SWAG), enabling efficient approximation of target-conditional score distributions at a fraction of the cost of conventional shadow-model ensembles. We evaluate the resulting attack on three downstream-adapted VLMs across two privacy-sensitive visual question answering tasks, namely Document VQA and Medical VQA. To isolate the effect of pretraining knowledge, we additionally introduce a controlled synthetic dataset. Our approach outperforms standard reference-based attacks while requiring only a single reference-model fine-tuning run, making membership inference more practical for large VLMs.

0.31

Agentic discrete-event simulation for manufacturing system design: an open 3D simulator, a benchmark and an industrial case

cs.AI ·2026-11-07 ·Javier Pernas-Álvarez, Diego Crespo-Pereira
benchmarkcode/data
Abstract · 55 pages, 15 figures, 8 tables. Software: https://github.com/simulean/SimuLean (doi:10.5281/zenodo.23565564)

The design of manufacturing systems relies on discrete-event simulation (DES) models that experts still build in tool-specific interfaces. AI agents can now write simulation code, but every model they write from scratch is new, unverified software. We present SimuLean, an open-source 3D DES simulator on the Unity engine designed to be programmed by AI agents: its modelling semantics are verified once, in the platform, and the agent builds and studies one model through domain-level commands, in the 3D scene or headless; physically simulated durations are calibrated for fast headless studies. We evaluated it on 18 manufacturing modelling tasks, from single machines to published production lines. A frontier agent built valid models in 93 of 108 attempts (86%), in about two minutes each; on the benchmark's interface version, two frontier agents built 105 of 108 models through domain commands alone, so that each comes with a record of how it was built. Every residual failure was silent - a model that builds and runs but answers wrongly - and only comparison with reference models revealed them all. In an industrial case, one agent working alone rebuilt, verified and studied a robotised welding workshop from a published study, in SimuLean and in SimPy. Comparing them with the published, expert-built model exposed an error in it; against the corrected model, the agent-built models agree within 1%. One agent can already build complete, realistic and traceable models for the design of manufacturing systems; independent reference models are what make them trustworthy.

0.31

Dynamic State Estimation from Sparse PMUs via Physics-Informed Graph Neural Operators

cs.LG eess.SY ·2026-11-07 ·Huy Hoang Le, Haoguang Wang, Christian Moya, Marcos Netto, Guang Lin
benchmark
Abstract · 10 pages, 10 figures, 4 tables

This paper proposes a physics-informed graph neural operator for power system dynamic state estimation. From a sparse, intermittent measurement window at a few generator buses, the operator estimates the rotor-angle and speed trajectories of all machines, unmeasured ones included, in one forward pass. A graph-attention encoder processes the measurements, and the swing equations are imposed as a hard constraint or as a soft one with a learned correction that absorbs model errors. It is trained on simulated measurements, without state labels. With several phasor measurement units, masked training hides one from the input but keeps it in the loss. On four transmission systems, simulations show that the operator is the most accurate estimator that neither optimizes per scenario nor uses state labels, except on the smallest system at large model errors, where tuned Kalman smoothers are better. Networks trained on simulated states are more accurate on the two smallest systems but not on the 118-bus system, where the operator is faster and more accurate than hard-constrained moving-horizon estimation. Under model errors, it matches or outperforms an optimization-based benchmark at a fraction of its cost. The results hold for a detailed machine model with exact parameters seen only through voltage phasors.

0.31

IIT Patna at "ICPR 2026 TVRID Competition on Privacy-Preserving Person Re-Identification from Top-View RGB-Depth Camera"

cs.AI ·2026-11-07 ·Abhinav Kumar, Swagatam Pati, Advait Deshmukh, Aditya Mehrotra, Bibek Das +4 more
Abstract

Privacy-aware person re-identification (Re-ID) from top-view cameras replaces frontal appearance with anthropometric shape, while depth sensing further reduces color-based identity cues. The ICPR 2026 TVRID competition evaluates this setting using 86 identities captured by four synchronized Intel RealSense D455 cameras with RGB and Depth streams across flat, ascent, descent, and oblique viewpoints. Performance is measured using mAP and CMC@1 across within-camera in/out, ascent-descent, and flat-oblique scenarios. The competition had 147 registered participants and 33 final submissions. Three teams from IIT Patna participated: 'abhinavkumar03' and 'sjkdgfhl_' ranked 5th and 7th among 31 ranked participants, and 'vivek katuri' reached 77.67% CMC@1 / 85.20% mAP on the RGB track. This paper presents the methods of the two top-10 teams. Methodology 1 (abhinavkumar03) achieves 93.51% mAP and 90.42% CMC@1 on Depth, ranking 4th of 28, while Methodology 2 (sjkdgfhl_) reaches 99.63%/99.35% on RGB and 86.26%/80.34% on Depth. Both substantially outperform the 57.19% track-average Depth mAP, with Methodology 1 yielding a 7.3-point gain over Methodology 2. These results demonstrate the importance of depth representation and supervision for privacy-preserving Re-ID.

0.30

DFT-Agent Bench: Scaffold Effects and Execution Provenance in Autonomous DFT

cs.AI cond-mat.mtrl-sci ·2026-11-07 ·Shijie Liu, Jin Yang, Weilin Wan, Jingtao Han, Jia LI +1 more
benchmark
Abstract

Large language models are increasingly deployed as autonomous agents for first-principles materials simulation. We report empirical findings from DFT-Agent Bench, a benchmark of 200+ Quantum ESPRESSO tasks spanning relaxation, electronic structure, equation-of-state, and elastic-tensor pipelines over multiple material tiers. Each task carries a QE-derived ground truth with public-reference cross-checks where available, and a provenance audit checks whether the required QE execution contract was met. Under one execution harness, we compare a near-bare ReAct baseline with a bundled v5.1 scaffold on two models. The complete scaffold improves bands+DOS by +0.08/+0.11 and d-electron vc-relax by +0.17 on both models, while high-scoring relaxation and equation-of-state cells leave little room for improvement. The elastic result is model-dependent (0.18 to 0.40 for deepseek-v4-pro and 0.62 to 0.59 for claude-opus-4-8). A fixed 100-task analysis with retrieval disabled supports the aggregate pattern: the fully frozen DeepSeek comparison gains +0.17, while the Opus comparison gains +0.10 against a historical pool baseline (p < 0.001). The provenance gate catches incomplete or unmarked executions that can receive high rubric scores; it is an execution-contract check, not a test for intentional fraud. These results characterize the evaluated DFT workflows. The anonymous review artifact contains the frozen diagnostic slice, prompts, scorer, and solver revisions: https://anonymous.4open.science/r/231e31re91287h/README.md.

0.30

SetPrint: Black-Box LLM Fingerprinting via Setwise Correspondence of Response Changes

cs.AI ·2026-11-07 ·Jinmo Han, Ju Yeon Kang, Jimin Hong, Sunghwan Ahn, Yonghyeon Jun +1 more
Abstract

Black-box LLM fingerprinting aims to determine whether a target model was derived from a source model when only the output responses to input queries are observable. One way to reveal this relationship is to perturb each query and observe the resulting responses of both models, yielding two corresponding sets of response variations. These sets provide two complementary sources of information: the relations among response changes within each set, and the correspondence of response changes between the two sets. Existing approaches, however, have not fully exploited both. We introduce \textbf{SetPrint}, a training-free method that jointly captures both within-set relations and between-set correspondence. SetPrint expresses each response change relative to the mean change of its own set, and then compares changes from the two models that were induced by the same perturbation. On LeaFBench, SetPrint consistently outperforms existing training-free black-box fingerprinting methods. Further analyses attribute these gains to both within-set relations and between-set correspondence.

0.30

Predictive Modelling of Solid Waste Generation through Time Series Analysis

cs.LG ·2026-11-07 ·Mitra Tithi Dey, Punyasha Chatterjee, Amlan Chakrabarti
unusual
Abstract · 16 pages, 5 figures

Reliable forecasting of Municipal Solid Waste (MSW) production is pivotal for sustainable urban management, as it enables efficient resource planning, optimized collection schedules, and reduced environmental impacts. Existing works on MSW forecasting have some limitations, such as unavailability of granular data from lower administrative units, small datasets, limited or sometimes no consideration of important factors that can affect MSW generation, including holidays, weekends, seasonality, changes in temperature or humidity. This work presents a comprehensive comparative analysis of four popularly used time series forecasting models, namely Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors (SARIMAX), Prophet, Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU), using a real-world MSW dataset from Sri Lanka spanning 2012-2018. A detailed factor-wise analysis is conducted to examine waste generation behaviour considering holidays, weekday-weekend variations, seasonal patterns, temperature and humidity. The results show lower waste generation during holidays and weekends, higher generation during Summer and Winter, and comparatively lower generation during Monsoon and Post-Monsoon. Temperature and humidity do not exhibit a direct linear relationship with waste generation, although their seasonal variations are associated with the changes in waste generation across different seasons. All model performances are evaluated using Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and the coefficient of determination (R2). Among the evaluated models, LSTM and SARIMAX achieve the best overall forecasting performance.

0.30

Learning a Unified Cross-Model Semantic Dictionary via Gated Bottleneck Sparse Autoencoders

cs.LG cs.AIcs.CV ·2026-11-07 ·Jinchi Zhu, Tie Luo
publishedcode/data
Abstract · The 40th Annual Conference on Neural Information Processing Systems (NeurIPS 2026)

Foundation models pretrained on similar data distributions tend to learn overlapping semantic structure, but their representations are distinct due to different model architecture, pretraining objectives and protocols. This prevents unified interpretability and cross-model feature reuse. Existing approaches extract interpretable concepts from individual models and align these concepts in a shared coordinate system using cross-reconstruction objectives, but the alignment only applies to latent coordinates while the concepts themselves remain model-specific and cannot be shared or reused. This paper introduces the Gated Bottleneck Sparse Autoencoder (GB-SAE), which casts unified cross-model interpretability as a representation factorization problem and solves it by simultaneously learning a unified shared dictionary and multiple private model-specific dictionaries via a gated bottleneck sparse autoencoder framework. The shared dictionary encodes the cross-model common semantic structure in unified concepts that are reusable across all participating models (and beyond), while each private dictionary comprises residual concepts that capture model-specific biases. We evaluate GB-SAE with distinct foundation models including ViT, SigLIP, DINOv2, and ConvNeXt. GB-SAE demonstrates superior reconstruction capability, unified and faithful interpretability, strong reusability and generalization to unseen models and downstream tasks. The code is available at https://github.com/trustAIoT/GB-SAE_NeurIPS.

0.29

Sharp Logarithmic-Product Bounds for List Replicability of Sign Matrices

cs.LG math.CO ·2026-11-07 ·Guangjian Zhang
theory
Abstract · 292 pages

List replicability measures how many accurate hypotheses suffice to contain the output of a learning rule with arbitrarily high confidence. We prove that every nonempty class of $m$ binary concepts on $n$ domain points has list replicability at most $805\ln(2m)\ln(2n)$. This resolves the logarithmic-product conjecture of Blondal, Hatami, Hatami, Lalov, and Tretiak. Hanner matrices give lower bounds of the same order, for square matrices and for explicit families with unequal logarithmic aspect ratios. The upper bound is obtained by constructing nested acyclic sets in the faces of a cube. A critical-point estimate supplies highly connected regions, weighted cell attachments keep the sets compatible across faces, and a cubical carrier theorem converts them into accurate finite covers and learning rules. Appending one constant coordinate turns list replicability into an exact geometric quantity, the face width, and graph-orientation classes satisfy exact product formulas. In particular, there are explicit classes with list replicability $4k+1$ and equivariant index $3k$, so the additive gap between the two is unbounded. We also develop entropy-tilting and two-sided Gibbs constructions, give combinatorial upper bounds adapted to structured matrices, compute exact values on classical families, and study the index of Paley matrices.

0.29

Detent: Batch-Invariant LLM Serving and RL Rollouts at Near-Stock Speed

cs.LG cs.DC ·2026-11-07 ·Peixu Xin, Runchuan Feng, Jiazhen Guo, Kan Wu, Ming Yin
Abstract · 18 pages, 14 figures, 7 tables

Reproducible evaluation, verifiable inference and on-policy reinforcement learning (RL) need an LLM engine to give a request the same output bits whichever requests share its batch. Batch-invariant kernels exist, but engines still select each kernel and its parameters from the batch. Two kernels that compute the same sum can round it differently. The engines' deterministic modes remove the choice by keeping one conservative kernel per operator and pay for it in latency. Many fast kernels are batch-invariant by themselves, so it is enough to fix the few batch-dependent values that kernel selection reads. Detent expresses kernel selection as a program that the engine executes. It finds the fewest values in this program that must be fixed and tunes them for the workload before deployment. On multi-GPU deployments of Llama-3.1-70B and Qwen3-30B-A3B in vLLM and SGLang, Detent is batch-invariant. Tuned per load, it serves at 1.04-1.17x the latency per output token of Stock (each engine with its standard settings), against 1.67-10.3x for the deterministic modes. In SGLang it matches the GSM8K accuracy of SGLang's deterministic mode. Detent covers multiple deployment combinations, including multiple engines, GPU generations, tensor-parallel degrees, and dense and MoE models. Seeded RL rollouts of models up to Llama-3.1-70B replay bit for bit under other batches and cache splits. A trainer on Detent's kernels and values matches every rollout logprob bit for bit at 1.03x Stock's rollout time per token. BF16 GRPO then stays stable where Stock's run collapses.

0.29

Safe Offline Reinforcement Learning using Behavior Regularisation and Latent Feasibility Guidance

cs.LG ·2026-11-07 ·Mahesh Keswani, Parth Bhardwaj, Ankur Kumar, Raunak P. Bhattacharyya
top venuebenchmark
Abstract · Accepted to NeurIPS 2026

Safe offline reinforcement learning (RL) seeks to learn policies that maximize return while satisfying safety constraints, using a static dataset. To characterize state-wise safety, reachability analysis identifies the feasible region, defined as the set of states from which constraints can be satisfied at every future step. Policies are commonly extracted from learned value functions with advantage-weighted regression (AWR). However, AWR relies on exponentiated weights that amplify value estimation errors, whereas behavior-regularized policy gradient methods directly utilize the action gradient of the value function and often make more effective use of learned critics. We propose Feasibility-Optimized Conditional Actor Learning (FOCAL), a method that incorporates behavior-regularized policy gradients into a conditional variational autoencoder (CVAE) actor. FOCAL optimizes this policy using a first-order objective partitioned by feasibility, updating the actor to increase return on transitions estimated to be feasible and to reduce constraint violations on transitions estimated to be infeasible. During inference, FOCAL dynamically adjusts the scale of the latent prior according to the estimated feasibility of the state to adjust the conservatism of action sampling. Theoretically, we show that for AWR, the policy update variance depends exponentially on the advantage error, whereas for FOCAL it scales linearly with the variance of the value estimation error. Across the DSRL benchmark, FOCAL satisfies the constraints on all evaluated tasks while achieving competitive or higher returns compared to baselines.

0.29

Receiver-Resolution APS-PDP Maps for Wireless Channels: Benchmarking and Coarse-to-Fine Modeling

cs.AI eess.SP ·2026-11-07 ·Peilin Zheng, Haozhe Jia, Xiucheng Wang, Wenshuo Chen, Nan Cheng
benchmark
Abstract · 18 pages, 6 figures

Angle Power Spectrum (APS) and Power Delay Profile (PDP) are receiver-facing spectra that govern beam selection and wideband equalization. Most ray-tracing benchmarks provide supervision as variable-size multipath component lists. This format does not match practical receivers, whose angular and delay resolutions are limited by array aperture and system bandwidth. As a result, learning targets are ambiguous and evaluations can change with reconstruction conventions. We introduce UrbanADPSMap, a benchmark that converts ray-traced supervision into fixed-grid APS and PDP targets by explicitly modeling bandwidth-limited and aperture-limited response kernels. By fixing the receiver-domain kernels, discretization, and validity mask, UrbanADPSMap removes reconstruction-induced target ambiguity and stabilizes downstream evaluation across methods. Built on this benchmark, we propose SpecFlowAR, a coarse-to-fine generator based on multi-scale autoregressive flow matching. The model first captures coarse dispersion structure and then refines beam-relevant peaks and low-power tails. The number of function evaluations controls the sampling budget, enabling latency and quality to be adjusted for dense spectrum queries. Experiments on the test set show that SpecFlowAR reduces BPL@5 to 2.03 dB under a unified NFE=80 budget, while improving beam utility and reducing errors of 3GPP dispersion statistics.

0.29

Hallucination-Guided Unlearning: Using Hallucination Traces to Reveal Overfitted Memories

cs.AI ·2026-11-07 ·Jiayu Zhang, Ziqi Zhong, Yuliang Gai, Canran Xiao
published
Abstract · 40th Conference on Neural Information Processing Systems (NeurIPS 2026)

Selective unlearning is increasingly important for the safety and compliance of large language models. Harmful or memorized content is often entangled with useful knowledge and difficult to enumerate. We use the model's own failures to identify candidates for unlearning. We propose Hallucination-Guided Unlearning (HGU), which uses hallucinations as diagnostic signals. HGU localizes unsupported spans via evidence and internal consensus, traces a sparse set of high-responsibility MLP neurons, and edits them with compact gates under a utility-preserving KL constraint. On TOFU, HGU achieves the best MU-FE trade-off, with Avg scores of 0.7221 (forget05) and 0.7699 (forget10), and the highest SepS H-Avg (0.4382/0.4715) under mixed forget/retain prompts. On the WMDP auxiliary task, HGU yields the strongest forgetting (e.g., 14.2 on Economics) while improving Retain (52.7) and MMLU (57.2). These results suggest that tracing hallucinations enables targeted, post-hoc unlearning that better balances forgetting efficacy and general utility without full retraining.

0.29

An LLM-Orchestrated Multi-Modal Chatbot for Tomato Disease Consultant

cs.AI cs.ET ·2026-11-07 ·Ngoc-Son Le, Khanh-Nghi Phan, Ngoc Hong Tran
Abstract · 14th International Conference on Multidisciplinary Research (ICMR 2026), ISBN: 978-974-474-120-2, Thaksin University Publisher, pp. 37-50, 2…

This paper introduces a chatbot developed to assist in diagnosing common diseases in tomato plants through natural language interaction. Users can describe symptoms in text. Then, the chatbot responds with the most likely disease, and basic treatment suggestions and related image-based reasoning. The system leverages pre-trained Mistral and Llama 3.2 large language models. The system is also built using LangChain and Streamlit to provide a lightweight and offline-friendly interface. This work targets to support tomato growers a fast and accessible tomato plant health guidance when the technical expertise is not available nearby, especially in the remote areas. The experimental results are promising and encourage further improvement of the chatbot for future real-world deployment.

0.28

Clinical Evidence-Guided Residual Learning for Medication Recommendation with Dynamic DDI Regularization

cs.LG cs.IR ·2026-11-07 ·Zesheng Li, Chengchang Pan, Honggang Qi
code/data
Abstract · 13 pages, 6 figures, 8 tables. Supplementary PDF and detailed result tables are included as ancillary files. Code: https://github.com/LIZESH…

Medication recommendation must combine patient-specific treatment evidence with the interaction structure of a prescription. The central challenge is to determine which evidence supports each medication and how strongly to regularize interactions while preserving predictive quality. We propose EviDrug, a clinical evidence-guided residual learning framework. Its clinical-memory backbone integrates a self-trained visit-aware encoder, training-case memory and one medication-set interaction block to establish a patient-specific prescription. A gated refinement head combines the backbone states with clinical associations and history-conditioned medication identities through bounded logit corrections. Drug-drug interaction (DDI) feedback adapts regularization strength using training-query predictions. Across MIMIC-IV, NWICU and SICdb, EviDrug leads twelve published-method baselines in AP, F1 and Jaccard, improving AP by 2.26, 2.06 and 1.71 percentage points over the strongest comparator. These gains come at lower parameter and runtime costs than MetaDrug. Matched controls demonstrate DDI reductions with stable predictive quality. Code is available at https://github.com/LIZESHENG13/EviDrug.

0.28

Group-Conditional Conformal Risk Control: Exact Guarantees, Limits, and Overlapping Extensions

cs.LG stat.MLmath.ST ·2026-11-07 ·Tareq Aldirawi, Wenge Guo
Abstract · 68 pages, including 40 pages of appendices

We study group-conditional conformal risk control (CRC) for disjoint and overlapping groups under bounded monotone losses. For disjoint groups, we establish exact finite-sample conditional risk control, two-sided bounds on the group-restricted risk, and fundamental limitations arising from local support and calibration-rank granularity. For overlapping groups, we study two exactly valid approaches: group-restricted calibration followed by union deployment, and direct calibration of the deployed prediction-set family. We characterize the union's limiting reconciliation gap, identify conditions for its disappearance, and establish sharp lower bounds for structured losses in specified product-form models. We also demonstrate a strict size advantage for direct calibration in an ordinal model. Experiments show that direct calibration can reduce prediction-set size, but its efficiency depends on group structure, loss type, and shape fitting. Results on historical diabetic-retinopathy scores further illustrate these tradeoffs. Overall, local support, reconciliation slack, and threshold heterogeneity determine efficiency, with neither approach uniformly dominating.

0.28

Decentralised Drone Swarm Formation under Silent Failures with ACO Pheromone Claims and PSO Tuning

cs.AI ·2026-11-07 ·Md. Abu Horaira
theory
Abstract · 19 pages, 7 figures, 22 tables. Code, browser demo and raw results are included as ancillary files

We study decentralised drone formation, without a central planner, under the conditions that make it hard: nobody knows how many drones are alive, failures are silent, radio is short-ranged and lossy, and actuation is noisy. In our controller, ant colony optimisation (ACO) allocates drones to picture positions through claims that behave as reinforced, evaporating pheromone and are gossiped between neighbours, so evaporation doubles as failure detection. A particle swarm optimisation (PSO) law flies the drones, a noise-buffered Voronoi filter keeps them apart, and PSO also tunes the controller offline. We prove that claims of destroyed drones vanish within bounded time, that conflicts between drones that exchange tables are resolved consistently, that survivors cover the picture to within a factor of two, and that drones cannot collide while their safety cells are feasible. On held-out seeds and with two tuned configurations, the controller formed the picture in 95-99% of runs in the hardest regime, restored it after a silent loss of half the swarm in 96-100% of runs at normal difficulty, transferred unchanged to eight unseen shapes and to 300 drones, and overlapped in 5 of 21,088 runs. A CBBA-style baseline that shares the evaporating claims but uses argmax without swaps was at least as reliable as ours once tuned with the same budget, and repaired faster, whereas ours generalised better to unseen shapes. The benefit thus comes from the pheromone dynamics and from tuning, not from ACO's stochastic construction or PSO's social terms. A centralised oracle is faster.

0.28

What Do Collusion Probes Read? Three Pitfalls in Evaluating Multi-Agent Collusion Monitors

cs.LG cs.AIcs.CLcs.MA ·2026-11-07 ·Mainul Hasan Mohsin, Mohammad Shahidur Rahman
benchmarkcode/data
Abstract · 30 pages, 5 figures. Code: https://github.com/mainul3195/collusion-probe-pitfalls ; data: https://huggingface.co/datasets/mainul3195/collusi…

Activation probes look like near-perfect monitors for secret collusion among large language model (LLM) agents. We show that they mostly detect that an agent was told to collude, not that it decided to. Our benchmark (2,500 committee and 1,476 code-review episodes) adds agents who receive the collusion instruction and refuse it, separating instruction from decision. Published detectors barely tell colluders from these refusers, and our best probe separates them mostly by reading the instruction: without it, agent AUROC falls from 1.00 to 0.61. Two further pitfalls inflate reported results: scored after the vote, a rule that only reads the votes reaches AUROC 0.990 but finds nothing before it; and in 57% of attempts no colluder argues for the target in public. The decision is still there: a probe trained on agents' own choices reads it (AUROC 0.73 without the instruction), and probes find colluders whose goal was fine-tuned into the weights (AUROC 0.995; pre-registered). We turn the controls into a protocol for testing any collusion monitor, and adapt scan statistics into an online monitor for groups of any size.

0.27

From Task Affinities to Optimal Allocations: Probabilistic Graphical Modelling for Supervised Finetuning

cs.LG ·2026-11-07 ·Prateek Chanda, Saral Sureka, Krishnateja Killamsetty, Ganesh Ramakrishnan
benchmark
Abstract · 43 pages, 6 figures

Supervised fine-tuning over a large pool of heterogeneous tasks is often computationally expensive, leading practitioners to train language models on a fixed training budget of carefully chosen SFT tasks. Thus, achieving strong downstream performance hinges on selecting an effective subset of tasks, or equivalently, identifying an optimal task mixture under a fixed budget constraint. In practice, supervised finetuning mixtures are often determined using simple heuristics (e.g., uniform or size-proportional sampling) that ignore task interactions, which can hurt transfer and waste budget on redundant task sources. We introduce TASKMIXPGM, a framework for learning continuous task mixtures through energy-based modelling over SFT tasks. Each task forms a node in a Markov random field, with unary potentials capturing task utility and pairwise potentials modelling inter-task relationships via behavioral divergences between predictive distributions of single-task fine-tuned models. Optimizing this objective yields task mixtures that balance utility, coverage, and redundancy. We further show that the induced set function defined over the set of such tasks is weakly submodular under budget constraints, enabling approximation guarantees for discrete task selection. Empirically, we provide a fixed, benchmark-independent mixture that improves over heuristic mixing across multiple model families and different training budgets. TASKMIXPGM also uncovers interpretable structure in task interactions, revealing complementarities and redundancies that inform effective SFT mixtures.

0.27

Breaking Precision Walls in First-, Second-, and Zeroth-Order Optimizers: One Constant, Set in Advance

cs.LG ·2026-11-07 ·Lngel Kyo
Abstract · 23 pages, 6 figures, 9 tables

Under power-law signal decay across directions, an optimizer can fit strong directions and miss weak ones while average loss hides it. A precision wall measures the steepest decay where the weakest direction recovers within 5% in a fixed budget. On low-rank matrix sensing and quadratics one constant sets the wall: once it dominates the denominator, Adam or grafted Shampoo is momentum descent at step $η/c$, setting the wall; in Shampoo, matching grafting's constant to damping restores the wall after three hypotheses failed. In zeroth-order Adam, $\varepsilon$ acts against $\varepsilon^\ast$ from a mean-square spectral model; below $\varepsilon^\ast$ error hits a noise floor. Set from true curvature or spectrum on new shapes, the rule raised the wall above the same-step default by a forecast amount: on two sensing shapes by $1.96, 1.15$ (Adam) and $1.59, 1.74$ (grafted Shampoo), and on two Hadamard-rotated quadratics by $1.12, 0.73$ (zeroth-order Adam, longer budget), rising on all seeds; fixed-step momentum at that step reaches those walls. A cosine-scheduled default (small constant giving weak modes large steps) is higher on one sensing shape, matched on new seeds of the other, and lower with the rule's constant. A default step retuned per budget catches up on rotated quadratics and population logistic regression, fails at 1 query/step on two diagonal quadratics (catching up at 16), and exceeds the rule on digits logistic regression (full-batch and minibatches, where gradient noise sets the $η= 0.01, 0.1$ default walls). On logistic regression the rule did not raise the first-order wall: same-step default Adam was higher (its wall on 4 new shapes at the forecast stability limit), while in zeroth order on the population loss setting $\varepsilon$ raised the wall by $0.98, 0.88$.

0.27

AERO: Adaptive Ensemble-Disagreement Routing for Oracle Feedback in Sample-Efficient Online RLHF

cs.LG ·2026-11-07 ·Shrey Singhal, Nannan Huang, Haytham M. Fayek
top venue
Abstract · Accepted at NeurIPS 2026

Aligning Large Language Models (LLMs) with human preferences via Reinforcement Learning from Human Feedback (RLHF) has driven substantial improvements across a wide range of tasks, but the cost of acquiring preference labels remains the dominant bottleneck in the alignment pipeline. While many recent alignment methods improve sample efficiency through active exploration over candidate responses, they typically query the oracle uniformly across contexts, even for contexts where the learned reward estimates are already reliable enough to provide synthetic labels. We propose Adaptive Ensemble-disagreement Routing for Oracle feedback (AERO), a fully online RLHF algorithm built on a Dyna-style model-based RL architecture, in which an ensemble-based Epistemic Reward Model (ERM) serves both as a reward posterior for exploration and as a learned preference-feedback model for synthetic labeling. AERO uses ensemble vote entropy, a Query-by-Committee disagreement measure, combined with an adaptive thresholding mechanism to decide which contexts require oracle feedback and which can instead receive ERM-based synthetic preference labels, concentrating expensive labels where they are most informative. Empirical results across different model families and scales show that AERO achieves highly competitive alignment performance while reducing the required oracle-query budget compared to online DPO and strong active-exploration baselines.

0.27

Toward a theory of Evaluability

cs.LG stat.ML ·2026-11-07 ·Kunal Talwar
published
Abstract · 19 pages. NeurIPS 2026

Machine learning models, particularly large foundation models, are increasingly evaluated along multiple axes that are relevant for their usage. In addition to accuracy measures appropriate for the specific task of interest, measures capturing safety, calibration and fairness properties are often of interest. The classical approach to these problems assumes that the test set has been held out from the training process, an assumption that is increasingly harder to justify for large foundation models. In this work, we formalize a notion of evaluability without this assumption, and ask if it is feasible to evaluate black-box models in settings where our test samples may have been used during training. We study this problem under different settings, capturing different levels of access to the model. We first study the natural setting, where the evaluator only looks at the output of the model on the test set. We fully characterize the class of distributions where accurate evaluation is possible, and show tight upper and lower bounds on the sample complexity of evaluation. Our results show that black-box evaluation in this setup is feasible if and only if the distribution is close to being small support. We then consider evaluation algorithms that can query the model on additional inputs and demonstrate connections to self-correctors studied in program testing. Using these, we show that a small amount of query access can increase the power of the evaluator for natural distributions and concept classes.

0.27

Sampling Is Not Curiosity: Why LLM Agents Should Investigate

cs.AI ·2026-11-07 ·Alfonso Amayuelas, Piotr Piękos, Xin Eric Wang, William Yang Wang, Jürgen Schmidhuber +1 more
top venueposition
Abstract · Accepted to the 40th Conference on Neural Information Processing Systems (NeurIPS 2026). Position Paper Track

LLM agents are increasingly used for open-ended tasks, very common in research scenarios where they are required to explore, investigate, or act curiously. At inference time, an LLM is a fixed map from context to next-token distribution, the pretrained weights determine the function, and only its conditioning changes as context accumulates. Mechanisms like temperature, best-of-N, tree search, or RAG vary how the distribution is sampled or what it is conditioned on, but do not update the agent's model of the environment. We argue that current LLM agents do not explore curiously; they explore stochastically. From how curiosity is understood in the RL literature, it requires a predictive structure over how the environment responds to the agent's actions, which improves online from current interaction, drives action selection through its predicted improvement, and persists beyond the context window. Additionally, in its open-ended form, the agent should intrinsically originate what to investigate. Thus, we propose a framework defining curiosity in LLM agents, reserving the term for systems that satisfy it. In doing so, we identify the research directions its implementation opens up.

0.26

The Fast Tree Transform

cs.LG math.DGmath.SP ·2026-11-07 ·Paul Burchard
unusual
Abstract

The Fast Tree Transform is a locally-aggregated transform defined for arbitrary graph geometries that generalizes both the Hadamard frequency transform and the Haar wavelet transform on the circle. The pure frequency part of the transform approximately diagonalizes the graph Laplacian; this fact makes it useful as a preconditioner in geometric optimization problems. The full transform provides an only slightly-redundant representation of signals on the graph that allows sparse representation of signals that include both edges and textures.

0.26

Towards Identity-Preserving LLM Agents under Model Transition

cs.AI cs.CYcs.HC ·2026-11-07 ·Yihuang Kang, Pei-Cing Huang, Chan-Tung Ku, Chan Hsu, Pei-Ju Lee +1 more
Abstract

When an LLM agent's backend model is upgraded or replaced, operators typically reload the same persona, memory, and rules onto the new model and assume the agent will behave as before. We show this assumption requires empirical validation-and that whether identity is preserved depends on the specific model pairing. Across 5,760 compliance-testing runs in 12 regulated domains with three open-weight models, we find that some transitions produce significant behavioral drift while others do not, even when the agent's external state remains identical. Crucially, this drift is invisible to verbal verification: every model claims full policy compliance, yet actual decisions under pressure differ across models. The same external state can reorder which model is most compliant-a large positive effect on one substrate, negligible on the other two. Practitioners should treat model transitions as behavioral changes requiring action-level empirical testing, not as transparent infrastructure operations.

0.26

TurnFlow: Benchmarking User-State Reasoning in Multi-Turn LLM Agents via Scene-Guided User Simulation

cs.AI ·2026-11-07 ·Zhixun Chen, Ming Li, Anjie Liu, Xiaoyuan Zhang, Quanyu Dai +1 more
top venuebenchmark
Abstract · EMNLP 2026 Findings

Large Language Models (LLMs) are increasingly deployed through conversational interfaces, yet strong single-turn performance does not imply reliability in multi-turn settings. Multi-turn assistance requires a model to maintain an evolving representation of user requirements and to reason over it before complying with later requests. Failures can therefore be subtle: models may fail to recognize a changed constraint or hidden conflict, or recognize it without enforcing it in the final response. Existing benchmarks primarily test fully specified prompts, long-context recall, or controlled domain environments, leaving open-domain requirement reconciliation undermeasured. We introduce TurnFlow, an interactive benchmark for evaluating whether models preserve, revise, and act on user requirements across turns. TurnFlow combines adaptive user simulation with verifiable judging, allowing conversations to unfold naturally while keeping hidden requirements and pass/fail criteria controlled. In our evaluation, nine LLMs achieve overall pass rates from 42.86% to 59.59% on 490 scenarios. Performance varies substantially across challenge types: models perform well on Stateful Multi-Constraint Following but remain considerably less reliable on Scope Creep and Goal Shift. These results reveal persistent limitations in reasoning over evolving user requirements across turns.

0.26

Learning When to Think: Adaptive Internal Computation for Reinforcement Learning

cs.LG cs.RO ·2026-11-07 ·Rohit Kumar Salla, Simon Stepputtis
top venuebenchmark
Abstract · Accepted at NeurIPS 2026

Iterative neural policies usually spend the same computation on every decision, although the benefit of additional processing varies across states. We introduce Adaptive Internal Computation (AIC), an iterative policy that learns when to stop refining its latent representation and act. In AIC, we train action selection and stopping jointly, an expected-depth penalty values computation and intermediate value predictions provide auxiliary supervision. We show that when only some states benefit from further refinement, an ideal state-dependent allocator outperforms every state-independent stopping schedule at the same expected depth. With at most eight refinement steps per decision, AIC executes 3.3 steps on average across twelve benchmark tasks and lowers per-decision latency by 31--44% relative to a fixed eight-step policy on four control and robotics tasks. Its mean reward exceeds the best fixed-depth policy on seven tasks and trails it on four. Ablating adaptive stopping and value supervision separately shows that stopping accounts for most of the reward gain on PickCube and AntMaze, where it also executes more refinement steps and that supervision accounts for most of it on RareEvent.

0.26

Rule-table structure of life-like cellular automata: activity, asymmetry, sensitivity, and the propagation storage factorisation

cs.LG q-bio.PE ·2026-11-07 ·Yair Cohen
unusual
Abstract

What structure on the rule space of a cellular-automaton family exposes which rules can support which behaviours? For the 262,144 life-like rules -- the two-dimensional outer-totalistic family containing Conway's Game of Life -- the standard coordinate, Langton's activity parameter, is long known to be too coarse. I organise the family by three symmetry-derived rule-table coordinates: an unweighted activity, a birth-survival asymmetry (odd under the swap of birth and survival), and the Vichniac-Binder Boolean-derivative sensitivity in closed form, whose centre-bit component vanishes exactly on the centre-blind (B=S) submanifold. The cardinality coordinates lambda, delta are necessary but never sufficient to certify behaviour; the centre-bit sensitivity labels the B=S submanifold on which the paper's one exact exclusion holds. Combined with an exhaustive enumeration of every rule's compact gliders, still lifes and oscillators (three-by-three seed, period at most sixteen), they expose a hierarchical factorisation: the birth condition is a narrow gate deciding whether anything can propagate, the survival condition tunes storage, and every compact-glider-supporting rule is birth-light and survival-rich. One exclusion is exact: no B=S rule supports a compact glider at that footprint, proven by exhaustive enumeration and closing only for compact-footprint patterns -- 36 B=S rules ship at four-by-four seed and 40 at five-by-five. Two dynamical results frame the whole: glider-capable rules that random sampling never expresses show that capability is certifiable from the table while its expression is only estimable from dynamics; and every uncatalogued rule producing a richly Life-like ecology turns out to be a Conway or HighLife variant on the neighbour counts its trajectories actually visit.

0.26

A Computational Map of the Disaster Mental Health Research Landscape

cs.AI cs.CYcs.DL ·2026-11-07 ·Soheil Shapouri, Saber Soleymani
surveyunusual
Abstract

When disasters strike, they not only damage infrastructure, buildings, and the economy, but also affect the mental health of affected populations. Research on the mental health effects of disasters has expanded rapidly and has become an important, well-funded area of psychology, public health, and related social sciences. However, the overall structure of this research landscape has not been systematically characterized. Using large language models, we conducted a review of all articles related to terms 'disaster' and 'mental health' indexed in PubMed. 7,859 articles' metadata were retrieved using the NCBI API, and then Claude Sonnet was used to extract structured variables from the abstracts. The results indicated that earthquakes and epidemics were the most commonly studied disasters while many other disasters such as heat and cold waves have received less attention. PTSD, depression, and anxiety were well investigated but many other mental health issues (e.g., aggression, suicide) are relatively ignored. Despite enjoying relatively large sample sizes (median = 452, IQR = 142 - 1,568), the field rarely utilizes machine learning algorithms. Hypothesis significance testing, reviews, qualitative methods, and structural equation modeling are respectively most common methods while agent-based modeling, machine learning, casual inference, and NLP each appear in less than 1% of articles. Studies are limited to a small number of countries, particularly the United States, Japan, and China, while many disaster-prone regions remain substantially underrepresented. This LLM-assisted computational map provides a broad overview of the disaster mental health literature while highlighting important methodological gaps, with particular emphasis on the underutilization of computational methods that could help advance the field.

0.26

Beyond the Mean Gap: What Transferred-Angle QAOA Retains on a Superconducting Processor

cs.AI quant-ph ·2026-11-07 ·Sanim Yousuf Fahim, Mohammad Arif Hossain
Abstract

Hardware executions of the quantum approximate optimization algorithm (QAOA) are often validated by comparing the device's mean cost with a simulation, a check that can miss what matters for optimization. Under global depolarization, the mean normalized gap and the floor-corrected probability of every set of bitstrings, the optimum included, retain the same fraction of their noiseless improvement over uniform sampling. We test this prediction with depth-2, transferred-angle QAOA circuits for binary subproblems of an alternating-direction decomposition on the IBM Heron processor \texttt{ibm\_marrakesh}, with and without dynamical decoupling and twirling. At 12 qubits, the main-family circuits retained 29\% to 46\% of their mean-gap improvement but only 5\% to 15\% of their probability of the optimum, lower on at least nine of ten instances per run. Near-optimal bitstrings were lost about as much as the optimum, whereas the avoidance of poor bitstrings was retained best. In the first submission, a device-matched noise model reproduced the pattern but overestimated the retention of the optimum five- to eightfold. On a rugged planted family, the device found the optimum in 62\% of 768-shot blocks (noiseless: 98\%), within the range of local heuristics. Validations of optimization circuits should report retention of the optimum and of near-optimal sets.

0.26

Evaluating classification and explainability in multimodal petrographic image analysis

cs.AI cs.CV ·2026-11-07 ·Thomas James Wood, Oktay Karakuş, Can Eyüpoğlu
Abstract

Automated thin-section classifiers increasingly combine plane-polarised and cross-polarised light (PPL and XPL), two complementary views central to petrographic identification. Classification accuracy alone does not reveal whether model errors reflect mineralogical relationships, how predictions depend on each optical view, or whether attribution maps identify regions that influence model decisions. Here, we examine these complementary aspects of the released LITHOS dual-stream vision transformer using outputs from an earlier evaluation of 105,802 expert-annotated grains across 25 categories. Classification errors frequently followed the label taxonomy: 72.6\% of misclassified plagioclase, microcline and sanidine grains were assigned the generic Feldspar label, while 87.1\% of quartz-family grains received a label within that family, compared with 66.6\% assigned their exact label. Replacing the XPL input with its per-channel spatial means produced a greater reduction in true-class probability than equivalent PPL replacement in 20 of 25 categories, but not in the three quartz categories. Five explainable artificial intelligence (XAI) attribution procedures generated visually distinct maps for identical inputs. In a superpixel-deletion analysis involving 30 grains from 29 thin sections, method-minus-random contrasts ranged from $-0.043$ to $0.057$, with all 41 simultaneous 95\% confidence intervals including zero. Together, these findings show that error structure, modality sensitivity and attribution behaviour capture distinct aspects of model performance that cannot be established from classification accuracy or visual inspection of attribution maps alone.

0.26

Physics2Weight: Direct Neural Network Architecture and Weight Generation from Physical Laws Using Large Language Models

cs.AI ·2026-11-07 ·Shogo Sato, Tomoyasu Shimada, Kazuhiko Murasaki, Takashi Sano
Abstract

Recent advances in large language models (LLMs) have enabled weight-space programming (WSP), which directly generates neural network weights without gradient-based optimization. However, extending WSP to physical problems remains challenging, particularly in achieving both solution accuracy and physical consistency. In this paper, we propose Physics2Weight (PtW), a framework that extends WSP to solving partial differential equations by directly generating neural network weights and editing architectures based on governing equations and initial and boundary conditions. PtW introduces physics-constrained prompting and physics-guided architecture editing, combined with iterative weight refinement from WSP, to improve solution accuracy and physical consistency without gradient-based optimization. Experiments on four PDEs, namely the heat, advection, Burgers', and Allen--Cahn equations, demonstrate lower solution errors than conventional physics-informed neural networks, while revealing limitations in satisfying the governing equations for nonlinear PDEs.

0.26

Deep backward multistep high-order regression methods for high dimensional nonlinear parabolic partial differential equations

cs.LG stat.MLmath.AP ·2026-11-07 ·Qiang Han, Shaolin Ji, Zihao Xv
theory
Abstract · 41 pages, 16 figures, 6 tables

We propose a deep backward multistep high-order regression (MHDBR) method for semilinear parabolic partial differential equations through their backward stochastic differential equation representations. The method combines multistep time discretization with regression targets formed by conditional expectations. Under explicit regularity, starting-value, forward-approximation and network-moment assumptions, we prove an error bound separating time discretization, conditional Monte Carlo sampling and neural-network approximation. The analysis assumes exact population minimization; finite-training-sample generalization and optimization errors are not included. For generators independent of the gradient variable, suitably constrained GroupSort networks and sufficiently large conditional sample sizes give order $J$ in the stated nodal $L^2$ error norm. The required network sizes depend on dimension, so this result does not establish dimension-independent complexity. We present a reproducible time-discretization test satisfying the analytical assumptions and three broader neural-network benchmarks, whose relation to the assumptions is discussed explicitly.

0.25

DiSARM: Diverse and Surprising Association Rule Mining with Large Language Models

cs.AI cs.LG ·2026-11-07 ·Lijia Liu, Hisashi Kashima
Abstract

Exploratory Data Analysis (EDA) aims to discover informative patterns in large datasets. For transactional data, Association Rule Mining (ARM) is an established approach that facilitates EDA by efficiently generating candidate patterns. Beyond efficient mining, identifying informative rules has also been extensively studied through both objective and subjective interestingness measures. Recent advances in Large Language Models (LLMs) offer new opportunities for assessing subjective interestingness using world knowledge. Nevertheless, relying solely on LLMs raises concerns about scalability and statistical reliability. Building on prior ARM research, we propose a hybrid framework, DiSARM, that leverages the world knowledge of LLMs to assess the subjective interestingness of rules, while maintaining statistical reliability and promoting diversity among the selected rules. Across three transactional datasets, experiments demonstrate that DiSARM effectively combines statistical reliability with surprise and diversity, offering a promising approach to ADE from transactional data.

0.25

TripletAD: Weakly Supervised Anomaly Detection in Embedding Space with Rare Anomalies and Scarce Labels

cs.LG ·2026-11-07 ·Sumit Wadhwa, Souvik Nath
Abstract · 15 pages, 2 figures. Under review at TMLR

Many practical anomaly detection problems come with only a handful of confirmed anomalies, and the unreviewed remainder of the data cannot be trusted to be normal. We study this setting under a curated-label protocol, in which unconfirmed anomalies are removed from training rather than relabelled as normal, and propose TripletAD, a weakly supervised detector that learns a triplet-margin embedding and calibrates a linear scoring head with two soft penalties: a partial cross-entropy on labelled anomalies and a relative-contrast hinge that separates the mean scores of anomalous and normal samples. Because every mined triplet contains a labelled anomaly, the few labels are reused across the entire training set instead of being diluted by it. On 24 ADBench datasets whose inputs are pre-trained embeddings (text, image and speech vectors), TripletAD outperforms established state-of-the-art weakly supervised detectors (DevNet, PReNet and DeepSAD) and gradient-boosted trees when anomalies are rare. At a 1% anomaly rate it has the highest mean AUCPR of six methods, 20% above DevNet; on the naturally rare Speech dataset it doubles the best baseline (0.394 vs. 0.196 AUCPR); and on MVTec-AD it is best when only a quarter or half of the anomalies are labelled. Pooled over 70 rare- or sparse-label settings, it has the best average rank (2.21 of 6) and outperforms each baseline under Wilcoxon signed-rank tests with Holm correction. When anomalies are more frequent it performs on par with these methods. The method needs no per-dataset tuning and is insensitive to its loss weights and training length; on raw tabular features, gradient-boosted trees remain the stronger choice.

0.25

RankShift: Evaluating Ranking Instability Across Benchmark Dimensions in Document Parsing

cs.AI ·2026-11-07 ·Halimat Afolabi, Daeyoung Lee
benchmark
Abstract

Document parsing benchmarks are often treated as stable measures of model quality, yet rankings may depend substantially on benchmark design choices. In this work, we present RankShift, a systematic study comparing OCR and vision-language models (VLMs) under controlled variation in dataset composition, matching strategy, and scoring methodology. Using OmniDocBench as a case study, we evaluate the ranking stability of 16 models across 3 dataset configurations, 6 matching strategies, and 3 scoring metrics. We quantify rank stability using pairwise rank correlation, top-k churn, and pairwise reversal rate. Our results show that the interaction between matching strategy and ground truth format is a substantial but understudied driver of instability: scoring against OmniDocBench's JSON annotations versus its Markdown rendering yields a mean Kendall's tau of approximately 0.33 under region-level matching strategies, with individual pairs as low as tau = 0.23, compared with tau = 0.77 when each page is scored as a single concatenated string. Dataset composition produces comparable instability across page-type strata (mean tau = 0.48) and layout-type strata (mean tau = 0.49), while the cross-dataset shift from OmniDocBench to Real5's Scanning reconstruction of the same pages is more stable (tau = 0.80). Metric choice is the most stable dimension, with mean tau of approximately 0.78 across NED, BLEU, and METEOR. These findings indicate that leaderboard position is not solely a property of model capability, motivating leaderboards that report not only headline scores but also the robustness of rankings across reasonable design choices.

0.25

Differentiable Agent-Based Modeling for Multi-Echelon Supply Chains: Gradient-Based Optimization of Heterogeneous Policies at Scale

cs.LG cs.AI ·2026-11-07 ·Bulent Soykan, Murtadha Alsayegh, Sean Mondesire, Ghaith Rabadi
Abstract · 13 pages, 3 figures, 2026 Annual Modeling and Simulation Conference

Global supply chains rely on the interactions of thousands of heterogeneous agents, posing significant challenges for optimization. While traditional optimization often assumes homogeneity to ensure tractability, Reinforcement Learning struggles with sample inefficiency when scaling to high-dimensional state spaces. We propose a Differentiable Agent-Based Model for multi-echelon inventory management. By implementing stochastic dynamics via differentiable tensor operations, we enable direct gradient propagation through the simulation. We validate the framework by recovering theoretical Newsvendor optima without supervision. Crucially, we demonstrate ``Mass Personalization'' by optimizing 10,000 distinct agent policies simultaneously in seconds, achieving a 53% cost reduction compared to scalar baselines. Furthermore, we utilize gradient sensitivity analysis to generate causal heatmaps, accurately recovering inherent lead-time delays from the computation graph. This approach bridges the gap between simulation and optimization, enabling the efficient training of self-correcting Digital Twins.

0.25

RetinaGuard: Risk-Controlled Agentic Triage for Multi-Disease Retinal Screening

cs.AI ·2026-11-07 ·Sohaib Sulman, Muhammad Atif Saeed, Akhtar Jamil, Alaa Ali Hameed, Roland Aydin
Abstract

Retinal screening programmes need to know which automated decisions can be accepted without review. This paper presents RetinaGuard, an agentic triage framework in which quality, diagnostic, uncertainty, verification and explanation agents gather evidence for each fundus photograph, a learned failure predictor estimates the probability that the decision is unsafe, auxiliary tools are invoked only when their expected value of information exceeds their cost, and the acceptance threshold is certified by conformal risk control. On the RFMiD test split, RetinaGuard was compared with confidence and calibrated-uncertainty referral at matched referral budgets and equal certified safety, using two RETFound primary models of different strength. With the weaker model (macro-AUROC 0.896), verifier disagreement added significant information, RetinaGuard captured 119 of 280 unsafe decisions at a 20% budget against 88 for lowest-confidence referral, and at a nominal 5% unsafe level it accepted 37.0% of cases against 10.3%. With the stronger model (0.925) the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points. In both settings its coverage at the nominal 5% level stayed near 37.7%, gating cut computation by about 30%, and a quality agent reduced unsafe acceptances under most image corruptions. The benefit of agentic evidence gathering thus depends on how much the primary model's own uncertainty already reveals about its errors.

0.25

Risk-Sensitive Reinforcement Learning for an Options Market Maker: An Empirical Study with Real S&P 500 Options

cs.LG q-fin.RMq-fin.CP ·2026-11-07 ·Jacky Chen, Ing-Haw Cheng, John Hull, Landis Li, Zissis Poulos +3 more
Abstract · 9 pages, 2 figures, 5 tables. Authors listed alphabetically by last name

Options market makers absorb client order flow and must hedge the resulting inventory risk under trading frictions. We study whether an option hedging reinforcement learning (RL) agent trained on historical market data outperforms three alternatives in tail-risk protection: (i) traditional delta hedging, (ii) simple hedge-ratio rules, and (iii) the same agent trained in simulated markets. Most RL hedging agents are trained in model-generated markets that do not capture the crashes and volatility regimes present in real data. We introduce an empirical replay environment using S&P 500 option data from 1996 to 2017 and compare against a matched geometric Brownian motion (GBM) environment, a stochastic-volatility (SABR) simulator, and a hybrid that values historical paths with Black-Scholes, all sharing the same portfolio mechanics, stochastic arrivals, costs, and test period. We train a distributional Twin Delayed Deep Deterministic Policy Gradient (TD3) agent and evaluate it out of sample over 2018-2023. The agent trained on empirical data reduces Conditional Value-at-Risk (CVaR) tail loss by 3.1% compared to delta hedging and outperforms every hedge-ratio rule chosen during the training period and every alternative training environment. The reduction reaches 8.9% in volatile and falling markets, but reverses to a cost of 7-10% in rising markets. Empirical training induces a state-dependent short net delta: the policy under-hedges negative-delta books, most strongly after prices have fallen. This tilt disappears with GBM training and is weaker with SABR or hybrid training. These findings suggest that exposure to historical market dynamics can materially change a tail-risk hedging policy, although the evidence rests on a single test path and the edge is regime-conditional.

0.24

Represented but Not Used: Value-Policy Dissociation in LLM Agents

cs.AI ·2026-11-07 ·Yuzhang Luo, Mohamed Aghzal, Liangming Pan, Ziyu Yao
Abstract · 19 pages, 5 figures

Large language model (LLM) agents need to choose actions whose consequences unfold over multiple future steps, yet poor decisions do not reveal whether the model failed to represent those consequences or failed to use them. Mechanistic interpretability has studied internal representations of environment state and plans, but sequential decision making also requires evaluating how candidate actions change future outcomes. Reinforcement learning formalizes this as action value. We study this computation in Qwen2.5-7B-Instruct, Qwen2.5-32B-Instruct, and GPT-OSS-20B across a partially observed Grid environment and WebShop, a simulated web shopping environment. We find that the relative values of candidate actions are linearly decodable from intermediate hidden states, including at states where the policy prefers a worse action. We call this gap between value representation and policy choice value-policy dissociation. To investigate how represented value relates to action selection, we compare direct interventions along the value readout direction with Natural Value Transport (NVT), a transformation learned from hidden state changes that naturally result from differences in action value. While readout-direction interventions have limited and inconsistent effects, NVT more consistently influences action preferences, particularly at later layers. We also test a closed-loop controller that uses the value readout as a sensor and NVT as an actuator, observing improvements in task outcomes. These results suggest that LLM agents can represent useful prospective value without reliably translating it into policy choice.

0.24

Spectral Bias Reduction in Local Gated ReLU Architectures

cs.LG math.NA ·2026-11-07 ·Victorita Dolean
Abstract · 24 pages, 4 figures, 5 tables. Preliminary version; comments welcome

Neural networks trained by gradient-based methods learn the low-frequency content of a target function faster than its high-frequency content. This spectral bias is well documented, but it is much less clear which architectural choices mitigate it and which cannot. We study the question for shallow ReLU models linear in their trainable parameters, where each mode is learned at a rate given by a Gram-matrix eigenvalue, and for a general symmetric bilinear form $a(\cdot,\cdot)$, so that least-squares regression ($a=$ the $L^2$ inner product) and variational or physics-informed losses ($a=$ the $H^1$ energy form) are two instances of one analysis. Within this framework we propose gated local ReLU dictionaries, obtained by multiplying each ReLU feature by a hat partition-of-unity weight, in the spirit of partition-of-unity finite elements and overlapping Schwarz methods. The resulting Gram matrices are block-banded for any local bilinear form, with a block-Toeplitz interior whose matrix-valued symbol governs the spectrum. The two forms then behave very differently. For $L^2$ regression, the symbol is bounded away from zero, the ratio is uniform in the number of windows, and the measured spread of frequency-dependent learning rates falls from $1.9\times10^{5}$ to $4.5$ at matched model size. For the $H^1$ stiffness matrix, the symbol vanishes at zero frequency: gating leaves the $Θ(N^2)$ conditioning intact and inverts the frequency ordering instead of removing the bias. A preconditioner built from the same interior stencil will fix the spectral bias, showing that architecture alone is not enough. Experiments on $1$D regression and Poisson problems confirm the dichotomy, and the regression effect persists when the hidden-layer parameters are trained as well.

0.24

Spiking Sensory Networks

cs.LG stat.ML ·2026-11-07 ·Lleyton Ariton
theory
Abstract · 46 pages, 18 figures

Effective sensory reception is critical for survival, yet the computational mechanisms behind how peripheral sensory neurons (SNs) transduce external signals into spiking representations for downstream brain circuits remain to be fully understood. Numerous theoretical frameworks have been proposed to model sensory processing, with Bayesian approaches among the most prominent. However, existing machine learning attempts to apply Bayesian theory to spiking neurons remain largely phenomenological, disconnected from empirical biology, and poorly validated on non-stationary temporal data. Conversely, neuroscientific models often rely on strong inductive experimental biases or explicit kinetic dynamics fit post hoc to match spiking data. Here, we introduce a class of recurrent Poisson VAE models, termed Spiking Sensory Networks (SSNs), that bridge these paradigms by reconciling variational inference with biologically inspired spike transduction for time-varying stimuli. Evaluated on real-world odor plume recordings from locust olfactory receptor neurons (ORNs), SSNs match or exceed the reconstruction fidelity of Spiking Neural Network (SNN) and GLM baselines while exhibiting superior representational characteristics. Furthermore, SSNs are well-suited for neuroscientific investigation, learning empirical in vivo ORN response motifs and exhibiting emergent spike-frequency adaptation under unsupervised reconstruction without manual tuning. Moreover, we theoretically link the recurrent decoding objective to mutual information maximization, and derive a formal thermodynamic lower bound connecting the Poisson Kullback-Leibler regularizer to the excess non-equilibrium metabolic power of biophysical homeostasis. Ultimately, our work synthesizes principles from machine learning theory and empirical neuroscience, providing a viable bridge that concretizes Bayesian models with biological mechanisms.

0.24

Reproducible LoRA Inference Through Deterministic Split-K Reduction

cs.LG cs.AIcs.DCcs.PF ·2026-11-07 ·Dhayaneshwar Balusamy, Ankur Mohan
Abstract

Reproducibility is important for LoRA-adapted language models used in repeated evaluation, auditing, and decision-making. However, Split-K kernels used to introduce parallelism in low-rank LoRA multiplications can have the unintended consequence of numerical nondeterminism when partial results are combined using floating-point atomic additions. We investigate this non-determinism in vLLM across GSM8K mathematical reasoning, MBPP code generation, and Jev-style structured toxicity decisions on Civil Comments and introduce a deterministic Triton Split-K implementation. Under identical-input replay, Atomic Split-K=8 produces different FP32 shrink outputs in 97.61% of comparisons, while our Deterministic Split-K=8 shows no observed variation. Repeated atomic inference changes generated outputs in 17.30% of GSM8K and 14.96% of MBPP comparisons, sometimes altering task correctness. Structured-decision probabilities vary in 81.47% of atomic comparisons, with decision flips. Controlled tracing demonstrates how shrink-output differences propagate into the KV cache and how accumulated cache differences can change token decisions and final answers. Our deterministic Triton Split-K=8 implementation preserves parallelism, achieving 16.8-21.1% latency-based speedup over unsplit reduction with only 1.02-2.62% additional latency relative to Atomic Split-K=8.

0.24

Self-contained Data Cleaning for Unsupervised Anomaly Detection

cs.LG cs.AI ·2026-11-07 ·Jordan F. Masakuna, Ahmed Yacine, Djeff K. Nkashama, Arian Soltani, Marc Frappier +2 more
Abstract · 13 pages, 10 figures and 4 tables

Training datasets for unsupervised anomaly detection are often assumed to be anomaly-free, but in practice, they frequently contain malicious or anomalous contamination. This contamination severely degrades the dependability and security of downstream detection models. While some contamination-robust methods exist, they rarely generalize across different model architectures and typically rely on a priori knowledge of the contamination ratio\textemdash information that is rarely available in real-world security operations. In this paper, we propose a novel, self-contained preprocessing framework for anomaly detection that identifies and removes training set contamination without relying on known contamination ratios. Our approach performs a dynamic analysis of the cosine similarity between each data point and its reconstruction during the early training stages of an auto-encoder, mapping points into a dynamic space where anomalous regions naturally exhibit lower density. By executing ``hard cleaning''\textemdash removing all identified anomalous candidates from the training pool\textemdash we systematically improve data quality. Because this cleaning occurs at the data level, the security and reliability improvements generalize seamlessly across various downstream anomaly detection architectures. Evaluating our framework across 8 distinct models and 10 security-relevant datasets, our approach achieves up to 100\% contamination reduction, yielding average improvements of 11.2\% in F1 score and 7.3\% in classification accuracy.

0.24

Prescriptive Learning of Linear Policies with Feasibility Guarantees for Binary Decision-Making

cs.LG stat.ML ·2026-11-07 ·Farzaneh Pourahmadi, Trine Krogh Boomsma, Pierre Pinson
Abstract

Binary decision-making under uncertainty arises across a wide range of applications, including energy systems scheduling, transportation, and workforce planning. While machine learning has been increasingly applied to accelerate or replace mixed-integer solvers in such settings, existing approaches rely on predictive learning, training models to reproduce optimal decisions from labeled data, which lacks knowledge of counterfactual outcomes and provides no guarantees that the resulting decisions will be feasible. This limitation is further compounded under distributional shift, where policies trained on historical data may produce systematically infeasible decisions. We propose a prescriptive learning framework for binary decision-making that addresses both limitations. Rather than predicting optimal labels, the framework jointly optimizes over linear decision policies and feasible assignments across all training scenarios, allowing feasible alternatives and their operational performance to be considered directly during training. We further extend the framework to a distributionally robust chance-constrained formulation using a Wasserstein ambiguity set, which enforces feasibility with a prescribed probability under the worst-case distribution in the ambiguity set and enables a tunable trade-off between robustness and solution quality. We demonstrate the framework on two case studies, namely a unit commitment problem and a staff allocation problem. Out-of-sample experiments under test distributions that either match or differ from the training distribution show that the robust model is more conservative when the two distributions coincide, while under distributional shift it substantially reduces unmet demand and lowers total cost by up to 76.5% relative to classical predictive learning.

0.24

Inference for Reasoning Models: Practical Lessons from Evaluation and Serving

cs.AI ·2026-11-07 ·Luv Verma
publishedbenchmark
Abstract · 217 pages. Book, first edition, version 1.0 (August 2026). Published on Zenodo

Running a language model involves more than its trained weights. The same model can receive very different benchmark scores, produce different answers, or cost more to serve depending on how it is evaluated and run. In one experiment in this book, changing only the maximum response length moved accuracy on a math benchmark from 0.08% to 93.18%. This book explains how inference settings shape the results we report and the systems we deploy. We begin with evaluation, including generation limits, answer extraction, scoring methods, and quantization. We then examine why outputs can change across batches, attention backends, and caching, even when sampling is disabled. The next chapters cover serving performance, including batching, memory capacity, latency, throughput, and cost. We also study structured outputs, tool calls, reasoning budgets, and long-context tasks, showing what these controls and benchmarks do and do not establish. The book concludes with practical guidance for recording inference runs, checking evaluations, estimating capacity, and comparing systems fairly. Written for researchers and engineers working with language models, it aims to make inference results easier to understand, reproduce, and trust.

0.24

Pulse waveform morphology encodes haemodynamic signatures of impending stroke

cs.AI ·2026-11-07 ·Cheng Ding, Jiaming Liu, Jian Wu, Hongxia Xu, Daoqiang Zhang
Abstract · 23 pages, 10 figures

Stroke is a disorder of the circulation, yet the haemodynamic state that immediately precedes it has rarely been observed, because continuous physiological recordings before a stroke are seldom available. Photoplethysmography (PPG), recorded by every pulse oximeter and by hundreds of millions of wearable devices, captures the peripheral pulse beat by beat, but its morphology is largely discarded. Here we ask whether pulse waveform morphology encodes haemodynamic signatures of impending stroke. Using in-hospital stroke as a natural experiment that provides continuous PPG before physician-adjudicated stroke recognition, we studied intensive-care (MIMIC-III, $n=176$) and emergency-department (MC-MED, $n=154$) cohorts and described each beat with interpretable descriptors of pulse timing, amplitude, contour and variability. At a nominal 360-min horizon, morphology separated pre-recognition from stable windows with $F_1$ scores of 0.941 and 0.989 (sensitivity 0.998 and 0.980), with models transferred unchanged to the second cohort, and outperformed heart rate alone on identical windows. The signature was dominated by systolic crest time and its change from each patient's own baseline, feature attribution was concordant across cohorts (Spearman $ρ=0.865$), and morphology recorded in the first two hours of care stratified the subsequent clinical course ($F_1$ 0.714 and 0.691). In high-risk patients without stroke, 29% and 14% of windows were positive, but five-window persistence rules limited alerts to 17% and 2% of patients. The results demonstrate that the routinely recorded pulse thus carries haemodynamic imprint of the period preceding stroke, positioning pulse morphology as a candidate continuous cerebrovascular biomarker.

0.24

Regularization Amplifies Subliminal Learning

cs.LG ·2026-11-07 ·Lawrence Feng, Jacob Mitchell Springer, Gaurav R. Ghosal, Ziqian Zhong, Aditi Raghunathan
Abstract

Recent work has shown that hidden biases can be transferred through data with no apparent semantic relationship to those biases. Can regularization limit this covert influence? By constraining adaptation, regularization might be expected to suppress behavioral changes on seemingly unrelated inputs. We systematically study this hypothesis in subliminal learning, as well as in related settings of covert influence, including phantom transfer and logit-linear selection. We find the opposite: regularization amplifies the transfer of hidden biases. In particular, subliminal learning is typically absent under full fine-tuning, but low-rank adaptation or explicit penalties can induce strong transfer. Phantom transfer and logit-linear selection show similar trends. Adding more data, an implicit form of regularization, also strengthens covert influence. Together, these findings suggest that the simplicity biases induced by regularization may facilitate covert influence. We further hypothesize that the level of regularization that maximizes covert transfer depends on task complexity.

0.24

Efficient LLM Adaptation with Forward-Only Passes

cs.AI ·2026-11-07 ·Baichuan Huang, Ananth Balashankar, Amir Aminifar
Abstract

Large language model (LLM) inference serving is undergoing rapid growth and large-scale deployment, motivating us to rethink how the inference process itself can be leveraged to enable efficient task-specific LLM adaptation. In this paper, we propose a forward-only approach for efficient LLM adaptation with forward-only passes in LLM inference serving. We exploit the angle concentration of activations induced by each singular value decomposition (SVD) component to measure its contribution-dispersed or concentrated-to the angle concentration of the hidden states. Based on this, our forward-only search then efficiently identifies the weight matrix with the highest dispersion that merits rank reduction. We then selectively remove the higher-order components and retain the lower-order components in SVD. Empirical results across diverse datasets demonstrate the competitive accuracy of our forward-only approach, while theoretical analysis shows lower peak memory and greater speedup than the gradient-based approach and greater speedup than the exhaustive search approach. The experiments further present the robustness and generalization of our forward-only approach to various LLMs with up to 57B parameters. The code is included in the supplementary material.

0.24

A Geometry-Guided Feature-Space Framework for Interpretable Unsupervised 4D-STEM Analysis of Complex Multiphase Materials

cs.AI cond-mat.mtrl-sci ·2026-11-07 ·Arda Genc, Daniel A. Bolden, Albert Nguyen, Daniel S. Gianola, Ravit Silverstein
benchmark
Abstract

Rapid, high-fidelity analysis of four-dimensional scanning transmission electron microscopy (4D-STEM) datasets from heterogeneous materials remains challenging because coexisting phases, orientation variants, local strain gradients, and intrinsic TEM thin-foil effects produce complex, overlapping diffraction signatures. Here, we present a machine-learning (ML) framework that combines physics-informed geometric descriptors with weighted high-dimensional embeddings to identify latent structural patterns with minimal prior knowledge. By constructing a regularized descriptor space before embedding and density-based clustering, the method improves cluster stability, suppresses noise-driven fragmentation, and reduces sensitivity to embedding hyperparameters. Spatial weighting produced the largest stability gain, increasing median cluster persistence from 0.04 to 0.55 while reducing the noise fraction from approximately 22% to 0%. Rotation weighting increased DBCV from 0.80 to 0.92, although the noise fraction rose from 0.2% to 1.0%, highlighting a trade-off between improved density separation and noise rejection. We validate the framework on two benchmark systems: Si/Si$_x$Ge$_{1-x}$ epitaxial layers with continuous lattice variation and deformation, and multiphase Ti-Nb alloys containing distinct phases and orientation variants. Comparisons with a pretrained vanilla autoencoder and model-free pattern vectorization show that the physics-informed geometric descriptor provides higher cluster fidelity, greater interpretability, and substantially lower dimensionality.

0.24

Compact Quantum Channel Attention for Static ASL Recognition: A Controlled Study with Parameter-Matched Baselines

cs.AI quant-phcs.CVcs.LG ·2026-11-07 ·Ronak Bhalgami, Deep Kothadiya, Nikita Bhatt
Abstract

Hybrid quantum-classical models often report good accuracy on image tasks, but it is rarely clear whether the quantum part is responsible. This paper proposes QCA-DenseNet, a static American Sign Language (ASL) recognition pipeline in which an eight-qubit variational circuit acts as a channel-attention gate on DenseNet121 features. The quantum channel attention (QCA) gate has 17,464 parameters, 7.51 times fewer than a squeeze-and-excitation (SE) block. The pipeline is tested on 1,815 MU HandImages images by holding out one of five signers at a time, and a frozen-backbone ablation uses small parts of the Kaggle ASL Alphabet data. On unseen signers, QCA-DenseNet reaches 94.50% mean accuracy, above SE (93.72%) and within 0.28 points of the best model. A matched classical MLP and a model without attention perform about the same, and no difference is significant. QCA gives a much smaller attention gate without a detectable loss in accuracy, but no quantum advantage.

0.24

Horizon-Scalable Q-Learning

cs.LG cs.AI ·2026-11-07 ·Haoran Xu, Pragya Srivastava, Xinjie Liu, Siddhant Agarwal, Amy Zhang
Abstract

Learning accurate off-policy state-action value functions (Q-functions) from offline datasets becomes harder as the task horizon grows. One-step temporal-difference (TD) learning bootstraps after every transition, so small per-step errors accumulate along the horizon, and early states can be badly misvalued even when the TD loss is small. Multi-step TD shortens this chain by following the dataset trajectory for several steps; however, since the behavior policy is often suboptimal, every followed action adds a bias, and no fixed number of steps suits a whole trajectory. In this paper, we propose Horizon-Scalable Q-Learning (HQL), a simple method that enables accurate long-horizon value learning by automatically balancing between one-step and multi-step returns. HQL extends a target through a dataset action only when the critic values it at least as highly as the best-of-N actions sampled from the BC policy, and otherwise falls back to a one-step target. We show that this rule is a principled off-policy correction that needs no action densities, which makes HQL compatible with expressive flow and diffusion policies. Experiments show that HQL avoids the failures of both one-step and multi-step TD on a combination-lock task across horizons, and outperforms prior horizon-reduction methods on long-horizon OGBench and OCBench tasks.

0.23

Dual-Granularity Learning for Regression with Continuous Noisy Labels

cs.LG ·2026-11-07 ·Hui Guo, Boyu Wang, Grace Y. Yi
Abstract

Learning with noisy labels has been extensively studied for classification, yet its regression counterpart remains relatively underexplored. Unlike classification, regression involves continuous-valued targets where label noise exhibits heterogeneous magnitudes across instances, which poses conflicting requirements on the model: correcting small-to-moderate deviations requires sensitivity to label values for instance-level refinement, whereas large deviations caused by corrupted or outlying annotations render exact label values unreliable, necessitating reduced reliance on exact labels and instead leveraging ordinal structure among samples. These requirements call for methods that capture both global structure and sample-specific deviations in a unified manner. To this end, we propose GRAIN, a dual-granularity framework for robust noisy regression. At the fine-grained level, we derive an explicit residual learning mechanism from a robust loss formulation, introducing instance-specific auxiliary variables to decouple label noise from predictive signals. At the coarse-grained level, we introduce a binning-based contrastive regularization that enforces an order-preserving feature structure aligned with label proximity, addressing noise beyond the capacity of local residual refinement. The two levels are coupled via a histogram-based bridging mechanism that enables bidirectional information flow: coarse-grained structural guidance establishes a well-ordered feature space that anchors residual learning, while fine-grained residual refinement corrects instance-level deviations that structural regularization alone cannot resolve. Experimental results show that GRAIN consistently improves robustness and generalization under diverse noise conditions.

0.23

BuySideFlow: A Point-in-Time Benchmark for LLM Agents in Investment Research

cs.AI ·2026-11-07 ·Dong Bo, Fanjin Zhang, Ke Cheng, Xinxin Zhao, Xiaowei Su +5 more
benchmark
Abstract

Large language models (LLMs) can generate SQL and SQL--Python workflows for investment research, but ensuring point-in-time (PIT) validity remains challenging. Agents must enforce disclosure cutoffs and select the appropriate historical states and record versions throughout data retrieval and financial analysis. We introduce \textsc{BuySideFlow}, a benchmark of 408 practitioner-derived Chinese-language tasks spanning equities, funds, fixed income, and macroeconomics over a fixed snapshot from a licensed production financial database. We provide executable references and a taxonomy of temporal errors, pairing result checks with expert audits of executed workflows. We evaluate nine agent configurations on all tasks, with result accuracy ranging from 16.2\% to 36.8\%. An audit of 900 runs on 100 shared tasks confirms PIT violations in 193 of the 380 runs that pass result checks (50.8\%), showing that output agreement alone does not establish temporal validity.

0.23

Approximate Recall, Approximate Forecasts: Recall as a Diagnostic for LLM Forecasting Errors

cs.AI cs.CEcs.LG ·2026-11-07 ·Prashanth Bhaskara, Shubhaankar Gupta, Seojoon Yeon
published
Abstract · 16 pages, 7 figures (including appendix). Accepted as a Spotlight at the ICML 2026 Workshop on Forecasting as a New Frontier of Intelligence

Large Language Models (LLMs) exhibit approximate recall: they retain coarse associations about past events while losing fidelity on specific details. Past research on LLMs has quantified different models' forecasting ability in real-world events using prediction markets, but has not studied how losing said fidelity impacts their forecasting abilities. For example: while an LLM correctly recalls that the answer to: ''Did Anna Sawai win the Best Drama Actress in the 76th Emmy Awards?'' is ''Yes,'' it incorrectly states the event date as September 22 rather than September 15. In this paper, we introduce a novel metric called the Recall Fidelity Score to study how imprecision in internalized knowledge propagates to forecasting capability on future events. On resolved historical markets, we find that higher local recall fidelity around semantically similar past events predicts lower forecasting error on resolved markets, and the same signal predicts smaller distance from open markets prices.

0.23

WARP-JEPA: World-Model-in-the-Loop Amortized Rollout Planning for Frozen Hierarchical Joint-Embedding Predictive Architectures

cs.AI ·2026-11-07 ·Junjie Li, Zhaojiacheng Zhou, Kaixiang Wang, Zihan Wang, QingSong Zhao +2 more
code/data
Abstract

Hierarchical Joint-Embedding Predictive Architectures (H-JEPA) enable long-horizon visual control but rely on costly test-time gradient descent for planning. Our analysis reveals that iterative optimization consumes up to 97.7% of episode runtime and that low predicted rollout costs do not necessarily translate into successful control. We further identify a planning-horizon asymmetry between long-horizon top-level composition and short two-step subordinate planning. To address these challenges, we propose WARP-JEPA, a hierarchical amortized planning framework that replaces online gradient optimization with learned multi-candidate trajectory proposers while keeping the world models frozen. WARP-JEPA combines a lightweight world-model-in-the-loop causal Transformer at the top level with parallel residual MLPs at subordinate levels. The proposers are trained using model-based rollout costs and best-of-K behavior cloning, enabling efficient candidate selection without test-time gradient optimization. Across 30 frozen H-JEPA models on four benchmarks, WARP-JEPA achieves 9.6-18.8x faster episode inference on long-horizon tasks (2.9-3.0x on Cube) and improves task success by up to 33.3 percentage points. These results demonstrate that amortized planning can substantially improve both inference efficiency and control performance without retraining the world models. Code is available at https://github.com/hideinsewer/WARP-JEPA.

0.22

Phase-Conditioned Reward Shaping for Token-Efficient LLM Reasoning

cs.LG cs.AI ·2026-11-07 ·Zhe Chen, Xiaolin Zhou, Xiyang Hu
Abstract · 24 pages, 17 figures, 15 tables

Extended reasoning can help language models solve complex problems, but long responses often contain both useful computations and redundant checks. Conventional length penalties encourage shorter responses without distinguishing the functions of intermediate steps. We propose Phase-Conditioned Reward Shaping (PCRS), a framework that combines phase identification with training-time reward allocation. We investigate two classification routes: PCRS-H with a hidden-state classifier and PCRS-T with a frozen text classifier. Our main text classifier implementation assigns overlapping reasoning, execution, and revision labels to native chunks, then uses these labels and exact repetition to adjust existing token penalties. We compare the completed configurations on six mathematical and scientific reasoning benchmarks and examine four reward-weight settings with the text classifier. Compared with a baseline using token-level length penalties, the default text classifier configuration reduces mean response length by 3.13%; sampled Pass@1 is approximately 62.07% for both methods, while oracle Pass@16 rises from 86.08% to 87.67%. These results show how phase-informed penalty allocation can reduce output volume and improve multi-attempt coverage without inference-time classification.

0.22

Factor Rank: A Canonical Algebraic Generalization of Tree Decision Diagrams

cs.AI cs.LO ·2026-11-07 ·Alexis de Colnet, Stefan Szeider
Abstract

Tree decision diagrams (TDDs), recently introduced by Capelli et al. (SAT 2026), are representations of Boolean functions. They generalize OBDDs by replacing variable orders with vtrees. The width of a TDD counts the distinct subfunctions across each vtree cut. Counting subfunctions is blind to linear structure. We introduce algebraic tree decision diagrams (ATDDs) by replacing this count with the rank of the cut matrix. The maximum of these ranks over all cuts, the factor rank, is exactly the width of the ATDD. Every TDD can be converted into an ATDD of no greater width. An ATDD is a tree tensor network in canonical form. ATDDs are exponentially more succinct than TDDs, and quasi-polynomially more succinct than structured deterministic DNNFs. Our main result is that this gain in succinctness does not cost canonicity. ATDDs are the first strict canonical form for a rank-based representation over arbitrary vtrees. For each function and vtree there is exactly one ATDD, over the rationals and without any genericity assumption. The canonicity admits a local characterization. Whether a tree tensor network is the ATDD of the Boolean function it computes can be checked in polynomial time from its tensors alone. Equivalence checking becomes a test of syntactic equality. The operation set includes polynomial-time model counting, weighted model counting, and a test whether a tree tensor network represents a Boolean function. Capelli et al. asked for the complexity of changing the vtree of a TDD and suggested local vtree changes. We answer the question. After a rotation of the vtree at one node, the new ATDD differs from the old one in two tensors only. Vtree change runs in output-polynomial time both for TDDs and for ATDDs. We formalize and verify our core results in Lean. We provide a prototype implementation of TDD and ATDD compilers as a first empirical validation.

0.21

Lightfoot: Graph Retrieval-Augmented Generation with Exact, Model-Free Document Deletion

cs.AI cs.CLcs.IR ·2026-11-07 ·Uday Allu
code/data
Abstract · 9 pages, 1 figure. Code: https://github.com/udayallu-ml/Lightfoot-GraphRAG

Graph-based retrieval-augmented generation (GraphRAG) improves multi-hop question answering, but deployed systems face two costs that research comparisons rarely measure: building the graph, and removing documents from it. We present Lightfoot, a GraphRAG system in which every entity, link, fact and synonym edge is a deterministic function of per-passage extractions carrying chunk-level provenance. Indexing makes one language-model call per passage, retrieval makes none, and deleting a document is a database operation with no model calls. On the full 11,402-passage MuSiQue corpus we verify that, after deleting 10% of the documents, the graph is identical to a fresh index of the surviving documents. We compare Lightfoot with HippoRAG 2, LightRAG and fast-graphrag under identical extraction, answer and embedding models on MuSiQue, 2WikiMultiHopQA and HotpotQA. Lightfoot builds the cheapest index on all three datasets, 26-31% below HippoRAG 2 and 3-6x below LightRAG and fast-graphrag. Against HippoRAG 2 it is significantly better on MuSiQue retrieval (+2.8 points recall@5 and +4.3 recall@10 on a held-out split) and on HotpotQA answers (+7.7 points exact match), and significantly worse on 2WikiMultiHopQA (-3.9 recall@2, -8.4 F1). Deleting the same documents took LightRAG 2,187 model calls and 3.7 hours; fast-graphrag has no deletion path. Code and per-question results: https://github.com/udayallu-ml/Lightfoot-GraphRAG

0.21

Information-Theoretic Generalization for Set-Input Optimization-Valued Objectives

cs.LG stat.ML ·2026-11-07 ·Futoshi Futami, Masahiro Fujisawa
Abstract

Modern machine learning is increasingly evaluated using criteria such as multicalibration, CVaR, and Cressie-Read DRO, which target subgroup reliability, tail risk, and distributional robustness rather than averages of fixed per-example losses. Existing information-theoretic generalization bounds for pointwise losses do not directly apply, because these empirical criteria depend on the entire batch and involve a post-hoc auxiliary optimization over witnesses, thresholds, or reweightings. We formalize this structure as set-input optimization-valued objectives. In the balanced supersample setting, we bound the expected optimized validation-training gap in terms of an objective-dependent sub-Gaussian parameter and the square root of the prediction-based conditional mutual information (fCMI) between the learned supersample prediction table and the selector vector. Combining this with fixed-predictor population-validation comparisons yields unified algorithm-dependent generalization bounds. We verify sub-Gaussianity of the selector gap using stability and convex L2-Lipschitz certificates, obtaining concrete bounds for multicalibration, CVaR, and Cressie-Read DRO, with a stability-based single-index refinement.

0.21

IdealCache: Rethinking Cache Scheduling in Diffusion Transformers via Ideal Trajectories

cs.AI ·2026-11-07 ·Shang Xue, Li Zhu, Ping Chen
top venueposition
Abstract · accepted 40th Conference on Neural Information Processing Systems (NeurIPS 2026)

Feature caching accelerates Diffusion Transformers by reusing intermediate computations across denoising timesteps, yet existing methods share a fundamental blind spot: they treat the standard reference denoising trajectory as the optimisation target, approximating its per-step states and thus inheriting its output as an implicit quality ceiling. We argue that this target is arbitrary -- what truly matters is the ideal result, not the intermediate path taken to reach it. We propose IdealCache, a training-free framework that abandons step-wise imitation of the reference trajectory. Instead, it anchors caching on a per-input ensemble of ideal trajectories constructed from candidate target results, aggregates their geometry into a single graph, and discovers optimal non-uniform schedules via a constrained shortest path on this graph. By decoupling from the reference process, IdealCache breaks the performance bottleneck of trajectory-approximation; our schedules can attain near-lossless fidelity in generation and even surpass on task-specific metrics (e.g., maintaining higher semantic consistency in image editing). Without retraining, IdealCache achieves 6.35x speedup on Qwen-Image, 5.84x on Qwen-Image-Edit, and 4.32x on HunyuanVideo, with competitive or superior quality across tasks.

0.21

Representing Tensors as Compositions for Efficient Computation on Sparse and Structured Data

cs.LG cs.MScs.PL ·2026-11-07 ·Valérian Rey, Praneth Paruchuri
code/data
Abstract · 15 pages, 6 figures

We propose a new mathematical formalism in which all kinds of tensors and of operations on those tensors are represented as simple function compositions. Tensors are functions from indices to values. An element-wise operation (e.g. an activation function) is a composition with a function applied to values; re-indexing (e.g. reshape, transposition) is a composition with a function applied to indices; and a reduction is a composition with both an index function, mapping each index to several indices, and a reduction operation (e.g. product, sum, max). Combined with tensor stacking, these compositions can represent most interesting kinds of tensors (e.g. dense, jagged, all kinds of sparsity patterns, matrix factorizations), as well as lazy versions of almost every operation available in modern tensor computation frameworks, including einsums between tensors of any kind. Identities between compositions can then be used to rewrite such expressions into equivalent ones, without evaluating them. For instance, the product of two sparse matrices can be rewritten so that only products between non-zero values are computed, turning a sparse problem into a dense one. Using this algebra as the intermediate representation of a tensor computation engine would shift the focus from defining operations on tensors (e.g. einsum between sparse tensors) to defining an algebra on index functions. This should reduce the codebase complexity from $\mathcal{O}(mn^2)$, with $m$ the number of binary operations between tensors and $n$ the number of tensor types, to $\mathcal{O}(m + k^2)$, with $k$ the number of types of index functions. A draft implementation of a tensor computation framework based on this algebra is publicly available.

0.21

ConwayNet: Interpretability, Activation Coherence, and Noise Resilience via a Game-of-Life-Inspired Lateral Inhibition in Artificial Neural Networks

cs.LG ·2026-11-07 ·Yair Cohen, Scott M. Gilmore
Abstract

Deep neural networks achieve universal function approximation by stacking interconnected nonlinearities, but as a consequence they are sensitive to noise and resist mechanistic interpretation. In contrast, the local update rules of cellular automata, such as Conway's Game of Life, give rise to not merely interpretable but programmable dynamics. We explore a bridge between these worlds: a lateral inhibition in neural networks based on continuous Game-of-Life rules that derives inhibitory and excitatory responses from the layer input and blends them into the activation map in a state-conditioned way. The mechanism is differentiable and architecture-agnostic, and its regulation is imposed rather than learned: the rule's parameters are fixed and are not optimised, while the rule's values and derivatives shape training throughout the network. Wired into fully connected, convolutional, wide-residual and graph convolutional networks, it gives rise to several properties. Activation maps become coherent across space, depth and seeds, with coarse structure persisting through depth while fine detail is renewed at every layer. The co-activation graph of the hidden units departs from an Erdős--Rényi random graph: it is laid out on the image and its giant component grows in steps, as in a spatial graph. Networks started from nearly identical initialisations, under label noise, disagree on thirty times fewer test images. Saliency maps rest on the object rather than the background, and resilience to pixel noise not seen in training is substantially improved.

0.21

BunshinRL: Whole-Workspace Forking for Agentic Reinforcement Learning Rollouts

cs.AI cs.OS ·2026-11-07 ·Yutong Huang, Vikranth Srivatsa, Alex Asch, Yiying Zhang
Abstract

Agentic reinforcement learning (RL) rollouts interleave model reasoning with tool calls in persistent workspaces. Scaling these rollouts incurs substantial memory costs for concurrent environments, while slow tool calls delay entire rollout batches. We present BunshinRL, an agentic RL rollout system that exploits shared environment state and recurring tool calls through three techniques: memory sharing across environments, tool-prefix caching, and speculative tool execution. These techniques build on BunshinFork, a novel live whole-workspace forking mechanism for containers. BunshinFork reconstructs process trees, shares memory copy-on-write, and preserves file, communication, and GUI state, enabling cached-state reuse and rollback after incorrect speculation. We implement BunshinRL and BunshinFork through Linux kernel module and CRIU modifications. BunshinRL reduces peak memory by up to 43%, total memory consumption by up to 51%, and end-to-end rollout latency by 23%.

0.20

ProtLink: Sequence and Multispecies Graph Context for Protein Interaction Prediction

cs.LG q-bio.MN ·2026-11-07 ·Xingyuan Zhang
Abstract · Submitted to AI for Scientific Research at AAAI 2027

Predicting protein-protein interactions (PPIs) from sequence alone is difficult, while known interaction networks may provide informative biological context. We introduce ProtLink, a sequence-conditioned graph-learning framework that combines frozen ESM-2 representations with protein interaction topology and orthogroup-mediated connections between species. We compare sequence-only models, GraphSAGE, a homogeneous graph transformer, and a heterogeneous graph transformer (HGT) for human PPI prediction. In an initial edge-level comparison, three-layer multispecies HGT with ESM-2 650M features achieves an average precision (AP) of 0.980 under random negative sampling, compared with 0.951 for the homogeneous transformer; its AP decreases to 0.878 with two-hop-preferred negatives. In controlled three-seed experiments, three-layer full-graph HGT obtains an AP of $0.9701\pm0.0022$, exceeding an orthogroup-only model ($0.9435\pm0.0094$). The corresponding two-layer comparison shows little additional benefit from non-human PPI edges, consistent with the greater message-passing depth needed to transmit their information. For human proteins excluded from training interactions, multispecies HGT achieves $0.8335\pm0.0094$ AP versus $0.6482\pm0.0052$ for a sequence-only pair model. These results identify the value, depth dependence, and limits of evolutionary graph context while distinguishing prediction of held-out edges from generalization to held-out proteins.

0.20

In Search of Lost Time: Temporal Competence in Large Language Models

cs.AI cs.CLcs.LG ·2026-11-07 ·Ryan Liu, David Lee, Icey Siyi Ai, Marwa Abdulhai, Rachit Dubey +1 more
Abstract

Today's large language models (LLMs) increasingly interact with time: AI agents schedule tasks, coordinate with people and other AI systems, and even decide how long to think. Yet LLMs only ever encounter time secondhand---in descriptions rather than experience. Does this hinder their reasoning and decisions that involve time? Our paper traces this question from a model's knowledge to its actions through four successive evaluations. We find that models know the rules of time, which can be read, but struggle to predict durations, which must be lived; while they are highly accurate (92.0-97.5\%) on a collection of questions built from the theorems of various time-related literatures, they underestimate their own completion times by up to 99.5\% and fail to identify which tasks will take them longer to run. And while they are better at estimating durations of human tasks, their duration judgments change based on how an event is worded, magnifying similar human biases by up to 1.3$\times$. Moving from estimation to action, we find that some LLMs disregard time costs completely when answering user queries, ignoring a 1,200-fold increase in wait time entirely. Together, our experiments reveal new challenges posed to LLMs as they navigate a dimension they cannot directly perceive, and highlight how their blind spots can even be simple cognitive abilities that humans take for granted.

0.20

CALFA-KV: Joint Calibration and Fisher-Guided Anchors for 1-Bit KV Cache Compression

cs.LG ·2026-11-07 ·Xiancai Tian, Leichao Zhang, Shenhui Zhang, Yi Shan, Hailong Li +3 more
Abstract

1-bit key-value (KV) cache compression can combine quantizer calibration with selective high-precision storage to reduce reconstruction error. We propose Calibration and Fisher-guided Anchors for KV cache compression (CALFA-KV). Offline, we jointly fit a key center, radial normalization parameters, and K/V vector codebooks to match the frozen model's projected attention output. We also estimate channel-wise Fisher-type sensitivities from squared gradients of a reference next-token cross-entropy loss. Online, we use these sensitivities to weight the squared reconstruction errors of each KV state. Here, a KV state is one token's key-value pair within one layer and KV head. We retain the highest-scoring states at high precision under a per-layer, per-KV-head budget. Under exact restoration, we bound the excess risk of score-based anchor selection relative to the optimum at the same anchor budget, where risk is the conditional mean square of the first-order loss change. With a budget of 56 additional FP16 anchors per layer and KV head, CALFA-KV (d8b8) outperforms all evaluated nominal 1-bit baselines on the mean score across seven benchmarks for each of the three models. At the same anchor budget, CALFA-KV (d8b10) uses a nominal index rate of 1.25 bits per channel and narrows the gap to BF16 to 0.90--2.01 points on the same metric.

0.20

Zeroth-Order Riemannian Optimization on Fixed-Rank Manifolds for LLM Fine-Tuning

cs.LG math.OC ·2026-11-07 ·Xinjue Wang, Xiuheng Wang, Yejun Zhang, Sergiy A. Vorobyov, Esa Ollila
Abstract · 32 pages, 2 figures, 23 tables

Zeroth-order (ZO) optimization estimates descent directions from function values alone, which removes the activation storage required by backpropagation and makes it attractive for memory-efficient fine-tuning of large language models (LLMs). Existing low-rank ZO methods either parameterize the additive update through Euclidean factors with a non-unique representation, or restrict each finite-difference direction to be low-rank. In this work, we propose to constrain each additive update block to a fixed-rank matrix manifold while keeping the pretrained weights frozen, so the learned update remains rank-constrained throughout fine-tuning. On this geometry, we build a Riemannian ZO gradient estimator that projects a random matrix onto the tangent space, normalizes it to correct the scale mismatch induced by projection, and further propose Zeroth-order Riemannian Gradient Descent (ZO-RGD) and its smoothed variant, ZO-RGD-S. We show that the ambient finite difference matches its retracted counterpart, the normalized estimator recovers the Riemannian gradient in expectation, and ZO-RGD converges under trajectory-regularity conditions. Experiments on LLM fine-tuning tasks show that the proposed algorithms are competitive with strong forward-only baselines across model families and tasks, with gains in several settings.

0.20

Meta-Reinforcement Learning with Zero-Shot Reinforcement Learning

cs.LG ·2026-11-07 ·Jake Grigsby, Siddhant Agarwal, Yu Lei, Leonidas Varveropoulos, Yuke Zhu
published
Abstract · NeurIPS 2026

Meta-reinforcement learning (meta-RL) trains agents to adapt to novel tasks from test-time experience, but requires diverse training environments and reward functions. While environments are becoming cheaper to generate, it remains difficult to anticipate every objective an agent may be asked to pursue at test time. Behavior Foundation Models (BFMs) learn policies from reward-free data and support zero-shot RL (ZSRL): adaptation to new objectives at test time using a large reward-labeled dataset. Recent work has shown that BFM task inference can instead be performed online without prior task-specific data. We note that this capability positions BFMs and meta-RL as alternative approaches for online adaptation to new reward functions in a fixed environment. We then ask whether we can extend BFMs to compete with meta-RL more broadly, including in novel environments, and identify four challenges: environment identification, online data collection, test-time exploration, and learning from mixed reward supervision. We study each individually, then propose MetaBFM, a hybrid method that addresses all four. During training, it combines supervised reward-following with unsupervised learning; at test time, it can interpolate between exploration, exploitation, and ZSRL-style reward inference. We demonstrate that MetaBFM can recover the in-distribution performance of supervised meta-RL while generalizing to reward functions outside the training set. Our work suggests that BFMs and meta-RL are complementary research directions whose combined strengths allow adaptive agents to learn more diverse behaviors from a fixed set of supervised objectives.

0.20

Beyond Query Similarity: Safety-Conditioned Cache Routing for LLM Response Reuse

cs.AI ·2026-11-07 ·Shilong Pan, Zhiliang Tian, Weiguo Chen, Zihan Chen, DongSheng Li
position
Abstract

Semantic response caching reuses a cached response for a semantically similar query to improve LLM inference efficiency. However, query similarity alone can lead to unsafe reuse when safety conditions change. Existing approaches improve cache safety through stricter similarity checks, context-aware validation, or response verification, yet they may still miss safety conflicts between the cached response and the new request or introduce unnecessary verification overhead. We argue that safe and efficient semantic caching requires a reuse policy that makes the cached response's safety conditions explicit and uses them to guide the reuse decision with a balance between safety and utility. In this paper, we propose SafeCache, a safety-aware framework that separates candidate retrieval from safety approval and formulates semantic response reuse as a safety-conditioned routing problem. SafeCache matches each cached-response condition against the new request and aggregates the per-condition mismatches into an overall risk score. A separate contradiction guard prevents explicit conflicts from being diluted by strong matches. It further estimates matching uncertainty and calibrates the routing policy under a required unsafe-reuse target and a verifier-call budget to choose among direct reuse, selective verification, or regeneration. Experiments show that SafeCache achieves the better safety-utility trade-off among the evaluated baselines, enabling safer response reuse at comparable cache utility.

0.19

RBLA+: Efficient Rank-Specific LoRA Adapter Extraction for Federated Fine-Tuning with Heterogeneous Clients

cs.AI cs.DC ·2026-11-07 ·Shuaijun Chen, Omid Tavallaie, Xin Chen, Wentao Gao, Nguyen Tran +1 more
hot direction
Abstract

Federated fine-tuning with low-rank adaptation (LoRA) reduces communication and computational overhead by training and exchanging compact trainable LoRA factors. Clients may use different LoRA ranks to meet adapter communication budgets and accommodate different local adaptation needs. The server must process these differences during aggregation and construct adapters at each client's rank for local training. We distinguish pre-aggregation alignment, which handles factor dimensions and scaling conventions before aggregation and rank-specific LoRA adapter extraction, which constructs client-sized pairs from the aggregated global model. Two common approaches to rank-specific extraction are direct rank truncation and Singular Value Decomposition (SVD) of the dense update. Direct prefix truncation keeps the leading columns of one aggregated factor and the corresponding rows of the other. It is cheap but can increase reconstruction error. The dense SVD approach first forms the product of aggregated LoRA factors. It then decomposes this matrix and retains the leading singular components to construct adapters at the target rank. This gives the optimal rank-constrained approximation in Frobenius norm but incurs significant computational overhead. To address this trade-off, we propose RBLA+ as an extension of RBLA. It uses established compact QR/core-SVD decomposition to minimize Frobenius error for the extracted LoRA update without forming the dense update. It also accounts for client and server scaling conventions during upload and broadcast. Across vision and language tasks, RBLA+ improves accuracy over RBLA in multiple heterogeneous settings. In CPU benchmarks with synthetic adapters, RBLA+ has lower extraction time than the dense partial-SVD baseline. RBLA+ preserves each client's prescribed adapter size and per-round communication payload.

0.19

SLA-Aware Task Offloading in the IoT Edge-Cloud Continuum: Class-Weighted Deep Reinforcement Learning Against an Exact Optimum

cs.AI cs.LGcs.NI ·2026-11-07 ·Shilpa Golla, Tharun Kumar Korine Palli, Urmila Bandaru
benchmark
Abstract

Deep reinforcement learning (RL) is often used to decide where Internet of Things (IoT) tasks with different latency limits run in the edge-cloud continuum, yet some methods give every missed task the same penalty and few compare their agents with an optimal policy. This benchmark study builds four offloading systems small enough to solve exactly by dynamic programming and measures reward and learner against the optimum. A bonus for a met limit makes each miss cost its penalty plus the bonus, which shifts the class weights of published per-class rewards. A class-blind penalty stays 65% to 118% above the optimal cost even with a perfect learner, while a corrected per-class penalty comes within 0.13% of the optimum. With a per-class reward, tuned proximal policy optimization (PPO) agents come within 9% to 11% of the optimum in the smallest system, and the corrected one beats a tuned reservation rule in 9 of 10 seeds. In larger systems the agents stay 41% to 76% above the optimum, and a tuned rule beats their mean. Training ten times longer does not close this gap, and a PPO-Lagrangian agent with reachable targets does no better than the per-class reward.

0.19

Learning Actionable Information Landscapes for Multimodal Active Sensing in Hawkmoths

cs.LG q-bio.NC ·2026-11-07 ·Abdelrahman Sharafeldin, Yaqing Wang, Simon Sponberg, Hannah Choi
Abstract

Animals integrate information across sensory modalities to guide exploration, suggesting that they maintain internal estimates of where actions are expected to reduce uncertainty. However, how such action-conditioned information maps are learned from multimodal sensory experience remains unclear, as previous studies have been largely confined to unimodal sensing frameworks. We introduce a multimodal predictive-coding framework for learning uncertainty-reduction maps from visual and mechanosensory observations in hawkmoth flower interaction tasks. A variational generative model trained on simulated environments recovers modality-specific information landscapes over the action space, capturing how visual patterns, surface geometry, and mechanosensory cues shape the value of exploratory actions. These learned maps are consistent with several features of hawkmoth behavior, including nonzero tracking offsets, probing along visual patterns, and geometry-dependent exploration. Applying the model to behavioral data from flower tracking and nectary search, we find that natural trajectories tend to occupy regions of high uncertainty reduction in these learned maps. Building on these observations, we develop an uncertainty-gated lexicographic algorithm that switches between information gathering and reward exploitation based on perceptual uncertainty. In nectary search, uncertainty gating improves learning relative to exploit-only, $ε$-greedy, fixed-time, and uncertainty-calibrated non-lexicographic controls, with better generalization and robustness to sensory and curvature shifts. These results link information landscapes learned through generative modeling to biological sensing and embodied exploration.

0.19

ADIndex: Layer-Adaptive Sparse Attention Indexing via Group-Level Distillation

cs.LG cs.AI ·2026-11-07 ·Jin-Fei Qi, Yiduo Li, Jiang-Xin Shi, Kan Liu, Tao Lan +2 more
Abstract · 27 pages,13 figures

DeepSeek Sparse Attention (DSA) reduces the computational cost of attention through token-level selection. However, its indexer still scores the entire causal prefix for each query, making indexing a dominant source of latency in long-context inference. Notably, the indexer alone can account for more than 50% of total prefill GPU time at long context lengths. To address this bottleneck, we propose ADIndex, an Adaptive and Distillation-driven Indexing method for token-level sparse attention. ADIndex integrates three key components: (i) a layer-wise calibration strategy that adaptively accommodates cross-query retrieval overlap, motivated by our observation of pronounced inter-layer heterogeneity; (ii) a group-level distillation scheme that leverages this heterogeneity to efficiently train adaptive layer-wise group indexers, capturing salient retrieval signals from individual queries; and (iii) a low-latency inference strategy that reduces repeated full-prefix scans through token bank sharing within query groups while retaining query-specific refinement. Experiments on DeepSeek-V3.2 and GLM-5.1 demonstrate that ADIndex preserves near-lossless task performance on LongBench, LongBench v2, and RULER while delivering substantial efficiency gains over standard DSA, including prefill and decoding kernel speedups of up to 6.22x and 3.40x, respectively, at a context length of 256K across the layer-wise reuse configurations, as well as an end-to-end serving speedup of 1.54x at a context length of 128K.

0.19

HexIt: Learning Hexahedral Block Decomposition with Expert Iteration on Combinatorial Maps

cs.LG cs.AIcs.CG ·2026-11-07 ·Arjun Narayanan, Per-Olof Persson
Abstract

Decomposing a polyhedral solid into a few well-shaped hexahedral blocks is a long-standing open problem in mesh generation. We cast it as a sequential decision problem on a 3-dimensional generalised map: an agent splits edges, faces and volumes with three insert operations until every cell is a hexahedron and the mesh passes a quality threshold, and a policy that convolves over the map's involutions chooses each insert. A successful decomposition takes over a hundred decisions, and a uniformly random policy never finds one. We clone a policy from constructed decompositions and improve it by expert iteration. On held-out chamfered solids that share no boundary topology with the cloning data, the final policy solves an expected $94.4\%$ in $20$ samples, and expert iteration raises the best-of-20 solve rate on large chamfered solids from $0.203$ to $0.355$. At matched compute, PPO, GRPO and VinePPO updates all make the cloned policy worse.

0.18

ATHENA: Hypergraph-Based Learning for Toddler Autism Classification from Eye-Tracking

cs.AI ·2026-11-07 ·Quang-Anh N. D., Duc Pham Minh, Thao Pham Phuong, Minh-Anh Nguyen, Thuy Linh Nguyen Thi +1 more
Abstract

In this work we proposed a architecture designed for the classification of ASD in toddlers using eye-tracking data, addressing the critical need for early and precise detection. The proposed model incorporates FMMix, a feature-level augmentation technique that bolsters the model's learning capacity, and in this work we proposed the ATHENA, which captures intricate feature relationships via an adaptive hypergraph framework. Evaluated on two datasets dataset of gaze coordinates-the model achieves state-of-the-art performance, with AUC scores of 99.80% and 99.86%, respectively. It also exhibits high accuracy of 98.45% and 98.56%, sensitivity 98.01% and 98.10%, and specificity 98.74% and 99.02%, respectively. Demonstrating its superior ability to identify subtle gaze patterns associated with ASD, these results significantly surpass existing methods, highlighting the model's potential for clinical ASD screening. Future efforts will focus on validation across broader populations and integration with multi-modal data to further refine diagnostic accuracy.

0.18

Embodied Science Agents: Making Embodied Experimental Design Learnable

cs.AI ·2026-11-07 ·Yihang Che, Zhongwei Yu, Yang Li, Xingyu Lu, Hongxia Yu +1 more
publishedposition
Abstract · Accepted by DAI 2026

Recent advances in large models have accelerated AI for Science toward agentic science, where AI agents autonomously conduct scientific discovery. Realising this vision requires closing the loop between computational reasoning and physical experimentation, positioning self-driving laboratories (SDLs) as critical enablers. However, existing SDLs exhibit a fundamental paradox: foundation model integration has increased linguistic sophistication while severing physical grounding. By treating models as disembodied planners that output symbolic workflows for separate robotic systems, current architectures inherently block the development of embodied experimental intuition. We argue that overcoming this paradox requires embedding \emph{learnability} along three interdependent axes: cognitive learnability (learning what to investigate), experimental learnability (learning where to search), and embodied learnability (learning how to act). Building on this view, we introduce the Embodied Science Agent (ESA), which unifies digital intelligence with physical embodiment through three synergistic components: a Digital Agent that provides cognitive and experimental learnability via memory-based reasoning and Bayesian optimisation, an Embodied Agent that provides embodied learnability via generalisable instrument control, and an Embodied Unified Interface for seamless coordination. We validated ESA in autonomous wastewater treatment experiments, navigating over 200 million possible conditions. Through 15 continuous rounds, ESA raised the final solution transmittance from 82\% to 99.7\%, corresponding to a 98.5\% reduction in residual dye concentration and near-complete decolourisation, while its learned visuomotor policies generalised across varying instrument placements, a promising step toward next-generation SDLs for autonomous scientific discovery.

0.18

R-BED: A Consequence-Aware Framework for System-Level AI Evaluation and Continuous Assurance

cs.AI ·2026-11-07 ·Vishal Srivastava
Abstract · 22 pages (18 pages main text plus appendices), 8 figures, 11 tables. Specification-level methodology with a synthetic walkthrough; no empiri…

Evaluations of AI systems usually report properties of a model, while decisions to deploy concern a configured system: a model together with prompts, tools, permissions, data, safeguards and an operating context. We ask how evidence about technical failures, deployment-control effectiveness and real-world consequences can be connected to defensible authorization decisions throughout the lifecycle of a deployed system. We present R-BED (Risk-Based Evaluability Design), a specification-level methodology with a fixed 6+4+3 architecture of thirteen diagnostic dimensions (technical failure, deployment controls, consequence and impact); a failure-mediated expected-loss identity that is kept separate from intended-behavior and misuse pathways; bounded evidence statuses with independence levels; a control assurance ladder; five authorization dispositions governed by eight rules; and a Continuous System Evaluation (CSE) layer in which evidence carries a configuration pin, an expiry and declared dependencies, so that a change invalidates evidence in proportion rather than wholesale. We work through a synthetic refund-agent walkthrough, with a deterministic script that reproduces every number, to show how a task-success view can miss a consequential control weakness that the procedure identifies as missing evidence. The contribution is a reproducible methodological specification. Most components are standard practice integrated into one evidence-to-decision record; the Evidence Validity Register, the authority envelope and the tier rules are proposals. None has been empirically validated, and we state seven hypotheses for independent testing.

0.17

Finite Time Analysis of Risk-Sensitive RL via Noisy Power Iteration

cs.LG ·2026-11-07 ·Waqar Mirza, Yashaswini Murthy, Laixi Shi, Eric Mazumdar, Adam Wierman
theory
Abstract

Estimating the principal eigenpair of a positive operator from stochastic evaluations gives rise to a noisy form of normalized power iteration: each step applies a sampled operator estimate and then renormalizes. We establish finite-time stochastic-approximation guarantees for this procedure, working in Hilbert's projective metric, which is well-suited to the multiplicative geometry of Perron--Frobenius eigenproblems. Our main application is risk-sensitive average-cost reinforcement learning (RL) with exponential utility. In this setting the Bellman equations are multiplicative, and value functions are characterized by nonlinear eigenvalue problems rather than additive fixed points. We cast both policy evaluation and control in this framework, obtaining risk-sensitive TD- and Q-learning algorithms that learn from a single Markovian trajectory. Under explicit positivity, mixing, and linear function approximation assumptions, we prove finite-time bounds on the recovered eigenvector and eigenvalue with $\tilde{O}(ε^{-2})$ sample complexity up to problem-dependent constants. The results cover both tabular and linear function approximation regimes, and through the duality between exponential utility and KL-robust control yield finite-time guarantees for KL-robust average-cost RL as well.

0.17

TrustMod-SM: A Multi-Axis Benchmark for Evaluating Trustworthiness of LLMs in Social Media Content Moderation

cs.AI ·2026-11-07 ·Aizan Zafar, Sreekantam Sai Venkata, Gathik Jindal, Ritu Raj Sharma, Zishan Ahmad +2 more
benchmark
Abstract

Large Language Models (LLMs) show strong potential for social media content moderation, but their trustworthiness beyond overall accuracy remains underexplored. We introduce TrustMod-SM, a benchmark of roughly 29,000 instances across eight datasets, to evaluate LLM moderators on five dimensions: trustfulness, fairness, safety, robustness, and context integrity. Evaluating 16 open-weight models (0.5B-284B parameters, spanning text-only and vision-language architectures) and one closed frontier model, we find persistent trustworthiness gaps. Higher detection accuracy often coincides with greater demographic disparity, and models frequently comply with concealment or exaggeration prompts, assign high confidence to wrong predictions, rarely escalate ambiguous cases, and struggle to separate hate speech from counter-speech or reclaimed language (especially below 8B parameters). Scaling does not uniformly resolve these issues, and no single model excels across all five dimensions.

0.17

Trained, Frozen, or Absent: Identifying What a Quantum Layer Contributes to Cross-Sectional Equity Ranking

cs.AI quant-phq-fin.CP ·2026-11-07 ·Chi-Sheng Chen, Xinyu Zhang, En-Jui Kuo, Rong Fu, Qiuzhe Xie +1 more
Abstract

Hybrid quantumclassical models are increasingly proposed for nancial prediction, but reported gains are rarely separated from the classical components around them. We study a hybrid quantum temporal convolutional network for daily cross-sectional ranking of 2000 Tokyo Stock Exchange equities under a protocol built to make the quantum contribution identiable: six conditions (circuit trained, frozen or absent, a parameter-matched classical surrogate, and the circuit alone, trained or frozen) through one code path; four walk-forward folds with ve seeds; bootstrap intervals, Deated Sharpe Ratios, backtest-overtting probabilities and family-wise error control; ve classical baselines including a gradient-boosted tree; and a locked holdout. On 500 out-of-sample days two one-sided tests place the trained-versus-frozen dierence inside 0.03 annualised Sharpe (p = 0.0055, minimum detectable eect 0.02): the circuit is equivalent to a frozen one, not merely indistinguishable from it. With the circuit as the head's only input the two are equivalent inside 0.25; against the network with no circuit the design resolves only 0.50. No condition is distinguishable from zero (best De-ated Sharpe Ratio 0.17). The holdout reproduces the ablation on a protable window, and 2000 US equities reproduce it where every condition loses money.

0.17

Pixels over Symbols: Sensory Realism Improves Behavioral Alignment in Models of Cognition

cs.AI q-bio.NC ·2026-11-07 ·Hao Yuan Bai, Xiaoxuan Lei, Zihan Weng, Motahareh Pourrahimi, Lucas Gomez +1 more
Abstract

Computational models of cognition have traditionally relied on abstract symbolic inputs that approximate sensory experiences with simplified vector representations. This simplification reflects historical computational constraints and a long-held assumption that the details of sensory realism are incidental to the study of cognitive processes. Recent advances have enabled the development of sensory-realistic models capable of performing diverse cognitive tasks, yet whether sensory realism meaningfully shapes cognition has not been formally established. We address this gap through a large-scale empirical comparison of abstract and sensory-realistic neural network models trained on a battery of memory-dependent decision-making tasks. Our results show that sensory-realistic models mirror human behavioral response patterns significantly more than their abstract counterparts, though a substantial gap to human internal behavioral consistency remains even for the best tested models. By analyzing the internal dynamics of these systems, we find that the learned recurrent weights are not the main source of this divergence. In fact, frozen abstract-trained weights driven by sensory-realistic inputs recover the human consistency of sensory-realistic models. The resulting dynamics are also within the seed-to-seed variability of sensory-realistic models. Together, our findings argue that across a broad set of visual working-memory tasks, sensory realism accounts for a substantial portion of the model-human behavioral agreement, more than the choice of sequence architecture and comparable to the choice of vision encoder. These results suggest that findings obtained from abstract task models in neuroscience should be more carefully interpreted in light of their potential behavioral limitations.

0.17

What Broke & Why: Practical Lessons from Reinforcement Learning Post-Training

cs.AI ·2026-11-07 ·Luv Verma
published
Abstract · 648 pages. Book, first edition, version 1.0 (2026)

Reinforcement learning with verifiable rewards (RLVR) has become an important way to improve how language models reason, solve problems, and use tools. But getting it to work in practice is not always straightforward. Training can fail because a model receives no useful reward, learns to exploit the reward system, stops exploring, or appears to improve when the evaluation itself is misleading. In this book, we give a practical introduction to reinforcement learning post-training for language models, focusing on what works, what fails, and why. We begin with the foundations of post-training, explain GRPO and the training loop step by step, and then follow five training programs covering math reasoning, search agents, training stability, mixture-of-experts models, and software engineering tasks. Each program shows the problems encountered, the changes made, and the lessons learned. The book also examines what reinforcement learning actually changes inside a model. The remaining chapters provide a practical reference for reward design, evaluation, training tool-using agents, debugging common failures, reproducibility, rollout engines, and hardware. Written for researchers and practitioners working with open models, the book aims to help readers understand how to train models, recognize when something goes wrong, and decide what to try next.

0.17

Universal Approximation by Equivariant Message-Passing Graph Neural Networks with a Bilinear Readout

cs.LG physics.chem-phphysics.comp-ph ·2026-11-07 ·Pingbing Ming, Han Wang
theory
Abstract

We prove universal approximation for the bilinear-readout equivariant message-passing network (BREMP), a reference architecture for equivariant graph neural network interatomic potentials. Its layers combine equivariant convolutions with affine maps, and its readout forms invariants from products of the equivariant features. We construct a one-layer feature that determines the local environment up to permutations of atoms of the same type. The construction uses spherical harmonic filters of degree no larger than the largest number of neighbors at a common distance from the center. Iterating the construction recovers the environment reachable in as many hops as there are layers, using the same bound on filter degrees at the layer cutoff. The readout approximates every continuous invariant function of the features uniformly on compact sets. Together these give universal approximation of atomic energies at a prescribed physical range using a shorter layer cutoff. We assume that, throughout the admissible class, message passing reaches all neighbors in each physical environment within a common number of layers. Finally, DPA4, an interatomic potential architecture in production use, inherits this result by approximating the BREMP layers constructed in the proof and realizing the bilinear readout.

0.16

InfoOrgBench: Can Agents Turn Scattered Information into Coherent Hierarchies?

cs.AI ·2026-11-07 ·Justice Ou, Lanming Ma, Zijian Zhang, Yilun Zhao
benchmark
Abstract

When agents organize records into folders, they may correctly group the excerpts of a design proposal yet place the proposal under the wrong owning group. Existing agent benchmarks do not directly measure such higher-level errors. We introduce InfoOrgBench, a benchmark of 150 tasks in which agents organize scattered records from meeting recaps, issue threads and design proposals into hierarchies. Integrity checks verify that every record is preserved, and a Structure Score measures agreement with the source hierarchy at every level, independent of folder names and order. We evaluate 45 systems that pair nine models with five agent harnesses. The best system scores 83.2 out of 100, but no system exceeds 78 on meeting recaps or issue threads. Model choice affects scores far more than harness choice: model means span about 27 points, whereas the average effect of each harness is below one point. More importantly, parent-level relations have higher error rates than leaf-level relations in all three domains. In design proposals, nearly half of the outputs analyzed by level make exactly this kind of error. These results show that evaluating information organization requires measuring higher-level relations separately from record preservation and local grouping.

0.16

TimeHF 2.0: Multimodal, Multivariate, and Multi-Region Collaborative Forecasting for Large-Scale Retail Supply Chains

cs.AI cs.LG ·2026-11-07 ·Zhongxuan Han, Zhengxin Shi, Kunxiu Chen, Xiaolin Huang, Yulin Huang +28 more
Abstract · 23 pages, 10 figures, 4 tables. Technical report

Large-scale retail supply chains need sales forecasts for specified products, future dates or planning windows, and regions under the conditions available at the forecast origin. Regional sparsity, changing environments, volatile sales, and unfamiliar business situations make these dimensions difficult to model together. We introduce TimeHF 2.0, a multimodal, multivariate, and multi-region collaborative forecasting framework that makes flexibly expressed business knowledge usable in numerical prediction. A coverage-aware directed encoder integrates regional sales and applicable covariates. A semantic alignment module transfers language-model hidden representations of natural-language plans and domain experience into target-specific forecasting context. Learning against sales observations connects these meanings to numerical evidence, allowing business users to supply experience, future plans, and changing circumstances without defining a new numerical feature for each description. Text-conditioned future queries and shared/routed experts adapt prediction to the supplied context. An ordered-distribution head allocates fine resolution to routine fluctuations and coarse resolution to extreme sales movements, while daily/window learning connects forecasts to cumulative planning quantities. Online parallel forecasting at JD yields relative accuracy improvements of 18.25% and 17.94% for national daily and planning-window sales, and 19.44% and 18.99% for DC daily and planning-window sales, over the original production deep learning baseline. Component comparisons, loss-design results, and product-type analyses examine the roles of regional information, covariates, business knowledge, context-dependent processing, and robust planning-oriented learning.

0.16

Improving Analytic Constants via Interpretable AI-Guided Lean Proof Refinement

cs.AI cs.LGmath.NA ·2026-11-07 ·Rahul Saha, Alan Li, Anton Xue, Swarat Chaudhuri, Adam Klivans +2 more
theory
Abstract

AI agents are increasingly capable of discovering new mathematical results, yet their reasoning and generated proofs can be difficult to understand and verify. We introduce an interpretable framework for AI-guided mathematical discovery, focused on improving quantitative bounds on analytic constants. Our key insight is that existing proofs often contain slack in intermediate lemmas, which can be tightened without reconstructing the entire argument. To exploit this structure, we represent proofs as partially-formalized Lean blueprints: human-interpretable dependency graphs of definitions, lemmas, and theorems. An AI agent then iteratively refines the Lean code while preserving the blueprint's dependency structure, allowing local improvements to automatically propagate to a verified final bound. This interpretable structure allows humans to inspect the agent's refinements and understand how local improvements propagate to the final theorem. For Grothendieck's constant ($\mathcal{K}_G$), we obtain $$ \mathcal{K}_G \leq \fracπ{2\log(1+\sqrt{2})} - 10^{-17}, $$ an explicit numerical improvement over Krivine's 1977 bound. For Korenblum's constant ($\mathcal{K}_K$), we obtain $$ \mathcal{K}_K \geq 0.356565, $$ improving Wang's 2025 bound of $0.3554$. Our experiments demonstrate the potential of both partial formalization and iterative agentic refinement for interpretable mathematical discovery.

0.16

M3FT: Micro-Macro-Micro Reasoning for Multi-Agent Cryptocurrency Perpetual Futures Trading

cs.MA ·2026-11-07 ·Jiashun Wang, Beile Wang, Guangyan Gan, Wentao Zhang, Siguang Li +3 more
Abstract · 9 pages, 3 figures, 2 tables

Cryptocurrency perpetual futures form a large-scale online financial market in which automated agents manage interacting asset positions under changing conditions. Policies relying solely on micro-level numerical states may overlook cross-asset relationships and become vulnerable to regime shifts, while macro-level linguistic judgments alone cannot ensure precise measurement or optimize positions under trading costs and risk constraints. Connecting market interpretation to coordinated multi-asset decisions remains challenging. We propose M3FT, a micro-macro-micro framework for multi-agent perpetual futures trading. The first micro stage constructs time-aligned numerical states from exchange and account observations through deterministic processing. The macro stage transforms these measured states into explicit market interpretations without adding external market information. Four asset-specific teams, each comprising a Market Interpreter, an Asset Communicator, and a Risk Reviewer, refine these interpretations through experience retrieval, cross-asset communication, and review of joint-position risks. The final micro stage combines encoded linguistic representations with the original numerical states to learn continuous target positions. We implement this framework with a language-conditioned multi-agent policy: asset actors use global and private numerical states together with linguistic context, while centralized critics evaluate joint actions using portfolio-wide information and shared-account rewards. Experiments on real cryptocurrency perpetual futures data, including comparisons across language models and policy-learning algorithms and component ablations, show that the best-performing M3FT configuration achieves a higher mean test return than the evaluated direct LLM, numerical RL, and rule-based baselines.

0.16

Polynomial Denoisers on the Image Manifold: Volterra Networks for Self-Contained Pixel Diffusion

cs.AI ·2026-11-07 ·Siddharth Roheda, Rohit Chowdhury, Amit Satish Unde, Aniruddha Bala
theory
Abstract

Natural images are often conjectured to lie on a low-dimensional manifold within the high-dimensional pixel space. Recent work leverages this to build self-contained pixel-space diffusion models without specialized machinery or pretrained components. A plain Transformer on raw image patches suffices, provided it predicts the clean image rather than the noise and compresses each patch through an aggressive bottleneck. The mixing operator, however, has remained softmax attention, inherited from sequence modeling rather than derived from denoising. For Gaussian noise, the optimal denoiser admits a canonical expansion in polynomial functionals of the observation, the Wiener--Volterra series. We prove that softmax attention, and any normalized mixing rule (including linear attention), cannot realize any polynomial map beyond an affine one. In contrast, Volterra layers realize this class directly and, on a data manifold, achieve approximation rates governed by the manifold dimension rather than the ambient pixel dimension. We therefore propose the architecture these results prescribe; raw image patches passed through a bottleneck embedding, followed by cascaded low-rank Volterra layers, which carry no activation function because each layer is nonlinear in its own right. To our knowledge this is the first pixel-space diffusion architecture whose mixing operators are Volterra layers throughout, with no attention, no convolution, no pretrained components, and no activation function on the residual path. On ImageNet $256 \times 256$ and $512 \times 512$, the resulting model is competitive with self-contained pixel-space diffusion at matched parameter and FLOP budgets.

0.15

Intra-Option Fitted Q-Evaluation: Evaluating Hierarchical Policies from Non-Hierarchical Data

cs.LG ·2026-11-07 ·Yunfu Deng, Josiah P. Hanna
Abstract

Off-policy evaluation (OPE) estimates the expected return of a target policy from previously collected data without additional environment interaction. While OPE methods for flat Markovian policies are well studied, little work has addressed evaluating hierarchical policies in which a high-level policy selects temporally extended options that generate primitive actions until a termination condition is met. The primary existing approach applies per-decision importance sampling at the option level, but requires option-annotated trajectories and exhibits variance that still grows with the horizon. For non-hierarchical policies, fitted Q-evaluation (FQE) typically achieves lower error than importance sampling by leveraging the classic Bellman equation for policy evaluation; however, a naive extension of FQE to hierarchical policies results in a biased policy value estimate. In this paper, we introduce Intra-Option Fitted Q-Evaluation (IO-FQE), which estimates a value function in the augmented state-option space; this approach enables OPE of hierarchical policies even when we lack annotations of what option was run in the data. We develop two continuous-action instantiations of IO-FQE and show on long-horizon hierarchical continuous control tasks (AntMaze navigation and OGBench Puzzle manipulation) that IO-FQE eliminates the bias incurred by a naive application of FQE to hierarchical policies and substantially lowers mean squared-error compared to option-level importance sampling.

0.15

Where Should Models Run? A Collaborative Inference Framework for Embodied Intelligence in 6G

cs.LG cs.AI ·2026-11-07 ·Wanlu Sun, Zhenhui Yuan, Xiang Ji, Yi Han, Qing Li +1 more
Abstract · Version 1 - Under Review

Embodied intelligence systems rely on fast and reliable inference to enable real-time perception, reasoning, and control. Recent advances in large-scale artificial intelligence (AI) models, including Large Language Models (LLMs) and Vision-Language-Action (VLA) models, have significantly enhanced the capabilities of embodied agents in complex environments. However, deploying these models remains challenging due to the limited computational resources and energy budgets of robotic platforms. A fundamental question is where inference should be executed. Collaborative inference, which partitions models across end devices, edge servers, and the cloud, has emerged as a promising solution. Yet determining the optimal partitioning point remains an open problem, as it depends on wireless network conditions, model structure, and privacy requirements. This article reviews the state of the art in collaborative inference for embodied intelligence and proposes an edge-centric framework in which the edge coordinates model partitioning and execution. Through both simulation and real-world experiments over WiFi and 5G networks, we demonstrate that the optimal partitioning strategy varies with network conditions and model structure. We further identify latency-minimizing partition points for ResNet-101, Qwen2.5-VL-7B, and OpenVLA-7B in collaborative inference between NVIDIA Jetson Nano or AGX Xavier devices and an AGX Thor edge server under varying wireless conditions. Finally, we discuss the future directions for collaborative inference in 6G-empowered embodied intelligence systems.

0.15

Guiding future global temperature distributions with simulations and observations

cs.LG physics.ao-ph ·2026-11-07 ·Francesco Immorlano, Elijah Tavares, Felix Draxler, Juan Nathaniel, Violette Launeau +4 more
benchmark
Abstract

Assessing climate risks requires projections of daily temperature distributions, including rare and spatially widespread extremes. Climate simulations are costly and biased, while correcting them toward observations risks suppressing inter-model differences. We train a single machine-learning emulator, a conditional diffusion model, on seventeen CMIP6 models and ERA5 reanalysis to generate spatially coherent daily temperature fields conditioned on model identity, date, and CO2-equivalent concentration. At sampling time, diffusion guidance reduces model biases by steering each emulated climate-model distribution toward observations, with its strength selected through perfect-model experiments using withheld simulated futures. Observational influence decreases with projection horizon and forcing. Across seventeen synthetic targets, guidance more than halves median biases in the 1st and 99th percentiles of daily temperature, improves probabilistic scores and reliability, and reduces median annual global-mean temperature RMSE by 32-57% across three emission scenarios. In a detailed benchmark, guidance also outperforms established bias corrections on daily-field probabilistic scores. Efficient sampling generates large ensembles without further climate simulations. For example, our framework corrects CMIP6's roughly threefold historical overestimate of extreme-heat coverage, yet projects that concurrent extremes will affect nearly half of global land on a typical day in the 2080s under SSP3-7.0.

0.15

A Topological Encoder Decoder Framework for Temporal Graph Learning

cs.AI cs.LG ·2026-11-07 ·Ronan Buck, Kiarash Shamsi, Tran Gia Bao Ngo, Astrit Tola, Baris Coskunuzer +1 more
Abstract

Most temporal graph learning methods reduce future prediction to discriminative inference over past interactions, focusing on edges or node labels for a largely persistent set of nodes. This view breaks down when new nodes and edges appear, since the future graph becomes a new object with changing size, composition, and structure. We address this gap by framing temporal graph prediction as an inverse topology problem. Instead of predicting edges directly, we first predict a multiscale topological descriptor of the future graph and then reconstruct a plausible future snapshot that realizes this descriptor under inductive constraints. This approach makes global structure and node churn explicit prediction targets and produces forecasted graphs that downstream tasks can evaluate without retraining. Across 14 temporal graph datasets, we evaluate our method, TOPOGED, on node, link, and graph property prediction and compare it against state-of-the-art temporal graph models. TOPOGED significantly improves node-level forecasting accuracy over the strongest baseline, with a macro average of 0.49 versus 0.05. It also outperforms baselines in 60% of graph-structure metric evaluations and substantially increases macro-average edge prediction metrics, from near-zero to 0.12. Our results show that topology-guided graph forecasting can predict inductive future snapshots whose structure supports multiple downstream evaluations.

0.15

Tail-risk-aware Agent Safety: Constraining Tail Risks Across Multi-step LLM Agent Workflows

cs.AI ·2026-11-07 ·Shilong Pan, Zhiliang Tian, Zihan Chen, Wentao Zou, Weiguo Chen +1 more
position
Abstract

LLM-based agents extend LLMs with tool calls, memory, planning, and retrieval to act in external environments, but they are vulnerable to attacks at different workflow steps. Existing defenses, including prompt engineering and post-training alignment, primarily improve LLM-only safety and often overlook risks that arise from multi-step execution. External verification adds guardrails to check each step separately, but it typically ignores how risks can propagate across steps. We argue that we should consider risk transitions across dependent steps through the agent workflow, rather than separate verification on a single step. In this paper, we propose \textbf{T}ail-risk-aware \textbf{A}gent \textbf{S}afety (TAS), a tail-risk-aware safety framework that models agent workflow safety as a sequence of dependent safety states and constrains the rare but extreme risks (i.e., tail risk) with a shared budget. TAS first models safety-state transitions sequentially along the workflow as an MDP to trace risk propagation, then measures a tail-sensitive risk to target rare but high-cost risks (i.e., tail risk), and constrains the risk with a shared budget to derive a step-wise safety strategy. By constraining tail risk rather than the overall average, TAS suppresses rare high-cost failures while preserving utility in normal cases. Experiments show that TAS achieves better performance compared with eleven baselines across three benchmarks.

0.14

DPDP: Diffusion Policy-Steered Diffusion Planner for Offline Reinforcement Learning

cs.LG ·2026-11-07 ·Ruyi An, Mingyuan Zhou
Abstract · 23 pages

In offline reinforcement learning, diffusion models fill two roles with complementary strengths. Typically guided by a critic, a diffusion policy is value-aware and favors high-return, behavior-supported actions, but it is short-sighted: it never lays out the states it intends to visit. A diffusion planner is far-sighted, synthesizing such a long-horizon trajectory, but it is only behavior-driven at execution, realizing each planned transition using only offline behavior.Each is strong exactly where the other is weak. We introduce DPDP to bring them together at the planner's execution step, where a planned transition becomes an action. Reading the planner's inverse-dynamics executor as a posterior over actions: the environment's transition likelihood relates an action to the planned next state, while the behavioral prior determines its weight under the offline data. Replacing this prior with a value-optimized diffusion policy preserves the planned transition likelihood and favors valuable actions that can serve the plan. We target the resulting clean-action distribution in two ways. A training-free sampler composes policy and inverse-dynamics scores at matched action noise levels and applies Langevin correction; this route uses a shared action noise schedule and an approximately common terminal prior. A preference objective instead amortizes the transfer into a single diffusion executor. Across D4RL, DPDP improves the planner it builds on, staying competitive with reactive diffusion policies on locomotion and leading on long-horizon navigation.

0.14

How Short Is Too Short? Length-Conditional Calibration and Abstention for Detecting AI-Generated Text

cs.AI ·2026-11-07 ·Ritesh Janga, Justin Zhan
Abstract

Detectors of AI-generated text are usually tuned and tested on passages of several hundred tokens, yet many real cases are short. We study score-based detection as a function of passage length. We give a closed form for how the false-positive rate of a threshold set on long text changes as length falls, and a formula for the number of tokens a detector needs to reach a target true-positive rate. We calibrate thresholds within length bins with conformal p-values, and we add a bin-level rule that abstains when the detector cannot be informative at that length. On synthetic scores the formulas agree with simulation, and binned calibration holds the false-positive rate at its nominal value at every length. On three zero-shot detectors (log-likelihood, log-rank, Binoculars), three corpora and two open-weight generators, binned calibration gives a mean false-positive rate of 5.0% in every detector-length cell from 10 to 200 tokens, averaged over 300 random document-level resplits. A single threshold set at 200 tokens gives 29% for Binoculars at 10 tokens and about 1% for log-likelihood, as the closed form predicts. No detector reaches an 80% true-positive rate, and the best, Binoculars at 200 tokens, reaches 59%, so the rule abstains in every cell at that floor and releases only the 200-token bins at floors of 0.3 and 0.5.

0.14

Zeroth-Order Langevin Dynamics for Stochastic Nonconvex Optimization: Curvature Regularization and Non-Asymptotic Guarantees

cs.LG math.OCmath.PRstat.ML ·2026-11-07 ·Stefano Bruno, Dong-Young Lim
Abstract

Zeroth-order methods provide a way to optimize stochastic nonconvex objectives using only noisy function evaluations. However, the smoothing inherent in their gradient estimators effectively modifies the objective being optimized. In particular, Gaussian smoothing with smoothing parameter $η>0$ induces, to leading order, an implicit Hessian-trace regularization, which motivates the study of the curvature-regularized objective of the form $f(x)+(η^2/2) tr(\nabla^2 f(x))$. We develop a non-asymptotic theory in the nonconvex setting for a class of algorithms called stochastic zeroth-order Langevin dynamics (ZOLD), which use only noisy function evaluations and do not have access to first- or higher-order derivative information, and derive excess-risk guarantees for the curvature-regularized objective. We first control the finite-time distribution of the ZOLD iterates in Wasserstein distance with respect to the Gibbs measure associated with the Gaussian-smoothed objective. An explicit coupling between the smoothing and inverse temperature parameters then quantifies the proximity of this measure to the Gibbs measure associated with the corresponding Hessian-trace-regularized objective. These distributional estimates in Wasserstein distance lead to excess-risk bounds for the curvature-regularized objective. Our analysis accommodates assumptions under which the stochastic function evaluations are allowed to have non-uniform dependence on the data stream. In addition, the constants in the non-asymptotic bounds do not rely on log-Sobolev or Poincaré inequalities as in prior zeroth-order Langevin analyses. To the best of our knowledge, this is the first non-asymptotic analysis of a zeroth-order Langevin method that simultaneously establishes excess-risk guarantees for a curvature-regularized objective and Wasserstein guarantees for the associated Gibbs measure in the nonconvex setting.

0.14

Optimal Byzantine-resilient Federated Learning with User-level Differential Privacy

cs.LG cs.DCmath.OC ·2026-11-07 ·Ming Xiang, Stratis Ioannidis, Edmund Yeh, Carlee Joe-Wong, Lili Su
publishedtheory
Abstract · NeurIPS 2026

Real-world deployment of federated learning (FL) requires resilience against adversarial clients and data privacy. Byzantine fault captures the worst-case behavior of adversarial clients, while differential privacy (DP) provides statistical guarantees against privacy loss. In particular, user-level DP guards a client's entire contribution and provides more realistic protection against information leakage. Despite intensive efforts, the minimax-optimal convergence bound of Byzantine-resilient FL with user-level DP remains open. We establish an information-theoretic lower bound for any Byzantine-resilient empirical risk minimization problem under user-level DP and heterogeneous clients. We propose ByzULDP, a central DP algorithm featuring a two-level momentum across the client and server. Client-side momentum mitigates stochastic gradient variance during robust aggregation. Unlike the fault-free case, naively combining non-linear Byzantine-resilient aggregation with central DP noise can be insufficient to guarantee convergence to a bounded neighborhood of a stationary point. Our key contribution is a server-side momentum that leverages a property specific to user-level DP, where privacy sensitivity depends on how client updates are constructed, to empower convergence. ByzULDP achieves the order-optimal convergence bound for strongly-convex objectives, with vanishing error terms decaying at the optimal $O(1/T)$ and $O(1/\sqrt{T})$ rates for strongly-convex and non-convex objectives, respectively. We corroborate our analysis with numerical experiments on real-world datasets across diverse privacy budgets.

0.14

Bayesian transfer for offline Q-learning with multi-task Gaussian processes

cs.LG stat.ML ·2026-11-07 ·Wenxing Guo
Abstract · 88 pages, 13 figures

Offline reinforcement learning can benefit from data collected in related tasks, yet indiscriminate knowledge transfer may lead to negative transfer under task mismatch and limited target-data coverage. We propose a target-aware Bayesian transfer framework for offline Q-learning based on multi-task Gaussian processes that jointly model task-specific Q-functions. Source tasks are assessed using target predictive evidence and maximum mean discrepancy (MMD), while posterior uncertainty and behavior-supported action sets guide conservative Bellman updates and policy extraction. To account for the dependence induced by Bellman bootstrapping, we distinguish Bellman innovations from bootstrap errors and establish stability and predictive source-selection results under explicit regularity conditions. We further derive selection-adaptive posterior contraction and offline policy-performance bounds, identifying the structural conditions under which source information can be pooled to achieve minimax-optimal rates in a shared-function Gaussian sequence experiment. Simulation studies with heterogeneous source tasks demonstrate improved target Q-function estimation and reduced material negative transfer compared with indiscriminate source pooling. An application to human decision-making data further shows higher average expected immediate payoff than both target-only learning and matched-size random source selection. These results highlight the benefits of combining target-aware source selection with uncertainty-aware conservative learning for reliable and sample-efficient offline reinforcement learning.

0.13

Agent Architecture Cost-Success Trade-offs: A Controlled Evaluation across Enterprise-Style SOPs

cs.AI ·2026-11-07 ·Subhrangshu Nandi, Hitesh Arora, Yunfei Bai, Udita Patel, Chirag Shah +9 more
Abstract

Enterprises increasingly use large language model agents to execute standard operating procedures (SOPs): long, branching, tool-intensive workflows. Yet there is little controlled evidence for choosing among competing agentic architectures. Holding the model, tasks, and tools fixed and varying only the architecture, we compare four designs (a function-calling single agent, a skill-augmented agent, centralized multi-agent orchestration, and a triage swarm; eight arms) across 13 enterprise SOPs, and introduce a skill-definition quality rubric to build matched high and low-quality skill corpora. Three takeaways emerge. (1)~\textbf{There is no universal best architecture; the right choice is SOP-dependent.} On a given SOP, task success ranges from under $10\%$ on the hardest to near $100\%$ on the easiest; no architecture is statistically separable in aggregate, and added structure only plausibly pays off on complex, branching SOPs (a directional, not yet significant, trend). (2)~\textbf{Added architecture cost must be justified by a task-success gain.} Every coordinated arm (multi-agent or skill-augmented) spends $1.4$--$4.7\times$ the tokens and $2.6$--$5.8\times$ the latency of the single agent for gains that are usually small and rarely significant, making the single agent the efficient default; high-quality skills do lift success (to $73.1\%$ from $67.4\%$) while degraded skills fall below it. (3)~\textbf{Implementation quality outweighs the single- versus multi-agent choice.} A plausible multi-agent implementation loses to the single agent, yet a targeted redesign of it becomes the only arm to significantly beat the baseline ($75.6\%$ vs.\ $67.4\%$).

0.13

On the universal structure in deterministic generative diffusion

cs.LG cond-mat.stat-mechcs.CVstat.ML ·2026-11-07 ·Defang Chen, Zhenyu Zhou, Can Wang, Zeyu Zheng, Siwei Lyu
Abstract

Diffusion-based generative modeling has emerged as a general computational framework for simulating high-dimensional, structured data across various domains. Understanding its generative mechanisms is paramount for advancing model training and inference in a principled way. Here, we demonstrate that deterministic sampling trajectories share a universal geometric structure invariant to data modalities and diffusion model implementations. This structure can be largely explained by an exactly solvable model based on the optimal linear denoiser.

0.10

Keeping LLM Agents on Procedure: Executable Finite-State Control for Long-Horizon Instruction Fidelity

cs.AI ·2026-11-07 ·Yutao Sun, Haokun Li, Mingshuai Chen, Guanyan Li, Zhongyi Wang +2 more
Abstract

Much of the automation handed to language-model agents is procedural: each step needs little intelligence, but a long unattended run must follow a skill or SOP stated once at the start. We show that agents drift from such procedures over long runs, and that restating the instructions or resuming a stopped agent only moves the drift to another shortcut. We study an executable finite-state machine (FSM) that keeps the workflow state outside the model's context and enforces it at three points: tool calls, state transitions and the end of the agent's turn. In a pre-registered experiment on 200-account SOP-Bench batches, Claude Haiku 4.5 with the FSM decided on average 139 accounts correctly and by the procedure, against 48 for the strongest baseline (95% CI of the difference [46, 130]), and kept this rate to the end; across three models, the gain tracked how much the plain agent drifted. Because the runtime, not the model, supplies this long-horizon reliability, the FSM can let small models carry procedural work: Haiku 4.5 with it averaged more such decisions than plain Claude Sonnet 4.6 at 38% of the cost. Analyses on SWE-bench Verified, $τ^2$-bench, and staged and cross-context workflows further suggest that reliable long-horizon automation can come from stronger execution control, not only stronger models.

0.10

Multimodal Models Can See, But Can They Think? A Study of Turkish Visual Reasoning

cs.AI ·2026-11-07 ·Berke Bunyamin Sule, Yavuz Selim Aygan, Ahmed Zeer, Elif Ince, Ali Mert Temizsoy +4 more
Abstract

Vision-Language Models have had massive leaps in recent years. Yet, when tested on high-level cognitive tasks like pattern completion, abstract logic, and relational reasoning, they still fall short. This limitation stems from the models' lack of ability to find out underlying rules and maintain constant intermediate steps, rather than merely deciding on the correct option. To research these shortcomings, this study analyzes visual reasoning performance throughout a multilingual evaluation framework encompassing the BİLSEM, MM-IQ, Visual Puzzles, and PuzzleVQA datasets. We evaluate various data mixing strategies and supervised fine-tuning (SFT) methods. We observed that a correct answer does not always mean the reasoning chain is valid. Because of this, we adopt an LLM-as-a-Judge approach to assess the quality of reasoning beyond final answer accuracy. Our experimental results reveal that baseline models often exhibit more stable reasoning compared to fine-tuned counterparts. Most importantly, we observe that SFT can lead to model collapse, manifesting as infinite text loops or empty outputs. The study concludes that current VLMs still struggle with abstract visual reasoning.

0.09

Look Ahead: Search-Driven Reasoning in Embodied AI

cs.AI ·2026-11-07 ·Ahmed Zeer, Osama Shbib, Muzaffer Kaan Yuce, Himmet Toprak Kesgin, Mehmet Fatih Amasyali
Abstract

Search allows a robotic policy to evaluate candidate futures before committing to an action. Motivated by test-time planning for vision-language action (VLA) models, we study selective search over action chunks in PushT using a frozen Action Chunking with Transformers (ACT) policy. Beam search generates perturbed alternatives around the policy prediction and scores them through rollouts from cloned simulator states. Across 50 evaluation episodes, beam search raises task success from 0% to 18% and increases average cumulative reward from 39.44 to 103.2. We introduce One Step Ahead (OSA), a gate that simulates the nominal policy chunk and invokes full search only when its predicted coverage drop exceeds a threshold. In the reported ablations, OSA preserves or improves success while reducing runtime by approximately 3x relative to full beam search. Ablations over tree depth, branching factor, beam width, and replanning frequency show that larger search budgets do not consistently improve performance. The results support selective test-time search for action-chunking policies; evaluation on language-conditioned VLAs and physical robots remains future work.

0.08

Trainability-Simulability Tradeoff in Parameterized IQP-based Reinforcement Learning Policies

cs.LG quant-ph ·2026-11-07 ·André Sequeira
Abstract

Parameterized instantaneous quantum polynomial-time (IQP) circuits generate distributions conjectured to be hard to sample classically. As reinforcement learning (RL) policies, their $2^n$ measurement outcomes are typically mapped to a smaller action space by a classical decoder. This aggregation can mitigate initialization concentration while making the policy classically simulable. We characterize this tradeoff through the decoder's Fourier--Pauli expansion. Under independent uniform initialization, we express policy variance as an exact sum over groups of nonconstant Pauli components with identical active-generator signatures. Each group contributes its squared aggregate coefficient times $2^{-r}$, where $r$ is the binary rank of its active generators. An inverse-polynomial aggregate coefficient with logarithmic rank prevents exponential policy concentration. Sparse Pauli expansions and contiguous action bins permit efficient classical additive estimation of action probabilities. We introduce a majority decoder combining protected low-weight components with exponentially large exact Fourier support, preserving initialization nonconcentration while excluding polynomial-size enumeration of the exact expansion. Numerical experiments corroborate the variance laws. In ten-qubit CartPole experiments, majority achieves higher returns than recursive parity and the tested local readouts, while higher-weight Pauli sectors contribute an increasing relative share of gradient magnitude during training. A complementary 21-qubit block-chain policy outperforms an extensive pool of fully connected classical neural networks in episode efficiency, even though its bounded-width structure permits efficient classical simulation. These results guide decoder design for quantum and quantum-inspired RL, distinguishing policy nonconcentration, learning performance, and classical simulability.

0.08

BRAT: Bounded Residual-Aligned Training for Stable Multi-Pass LLM Post-Training

cs.LG ·2026-11-07 ·Xunlan Zhou, Yuzhe Zhao, Xuhong Huang, Feng Chen, Xuanlin Chen +2 more
Abstract

Reinforcement learning for large language model post-training often reuses each batch of sampled rollouts for several gradient passes to amortize the cost of generation. Under such multi-pass reuse, existing objectives fail in different ways: unclipped updates collapse as the policy drifts away from the sampling policy, adding a negative baseline avoids collapse but reinforces every sampled response, and PPO-style ratio clipping cuts off gradients abruptly. We first ask why reuse is fragile. Measuring how well an update estimated from a small rollout group aligns with the KL-regularized target on held-out rollouts of the same prompts, we find that this alignment drops by nearly an order of magnitude from in-sample to held-out rollouts across three base models, suggesting that repeated passes largely overfit the sampled group. Motivated by this finding, we propose Bounded Residual-Aligned Training (BRAT), motivated by a proximal update toward a bounded reweighting of the sampling policy. BRAT assigns each token a gradient weight that decreases linearly with its probability ratio and vanishes at a trust-region boundary, and normalizes group-centered rewards by their maximum magnitude, keeping advantages zero-mean and bounded without extra hyperparameters. With four reuse passes, BRAT avoids training collapse, limits policy drift, and achieves higher average accuracy with lower seed variance on most benchmarks.

0.07

CONSTRAINER: Promptable Graph-Structured Optimization via Constraint Conditioning

cs.LG ·2026-11-07 ·Zeeshan Memon, Mingke Tian, Xinyuan Song, Hongwei Jin, Kibaek Kim +1 more
Abstract

Neural methods for graph-structured optimization have achieved strong performance within individual problem classes, but are typically trained for fixed formulations, limiting reuse across related problem variants. Recent progress in foundation models has motivated shared neural backbones trained across multiple graph-structured optimization tasks; however, existing approaches primarily share instance-level representations while treating feasibility constraints implicitly. A key source of shared structure across constrained optimization problems is the constraint set itself, which often exhibits substantial overlap across tasks. We propose CONSTRAINER, a constraint-conditioned framework for graph-structured optimization that treats constraints as structured inputs. Constraint specifications are parsed into operator graphs and encoded using a graph neural network to produce invariant constraint embeddings, which condition a shared backbone model for solution prediction. We evaluate our approach on power grid optimization and mixed-integer linear programming problems, where it shows improved solution optimality and enables efficient transfer learning and generalization compared to baseline methods.

0.07

AbductionBench: Benchmarking Abductive Reasoning in Large Language Models

cs.LG ·2026-11-07 ·Moein Salimi, Amir Qeysarbeigi, Babak Hosseini Mohtasham, Faraz Doagoye Tehrani, Mahdi Naieni +6 more
Abstract

Abductive reasoning is the inference of plausible explanations for observed evidence. Evaluating this capability requires more than a single task: a system may need to generate an explanation, discriminate among competing hypotheses, revise its hypotheses as evidence arrives, or decide what evidence to seek next. Yet these capabilities are typically evaluated through isolated benchmarks with different task formulations and scoring procedures, so claims that a model is ``good at abduction'' may reflect only a narrow aspect of this capability. We introduce AbductionBench, a unified framework whose taxonomy characterizes abductive tasks. We instantiate this framework on 27 datasets spanning ten domains and evaluate eight LLMs from four model families evaluated on single-choice selection tasks. Beyond harmonizing heterogeneous tasks, AbductionBench separates final-answer evaluation from analysis of the observable reasoning traces and interaction histories produced alongside those answers. We analyze written traces as observable model outputs rather than faithful representations of internal computation, characterizing evidence use, hypothesis comparison, uncertainty, contradiction handling, and answer commitment. Jev matches or exceeds small LLMs when selecting among explanations over compact evidence but falls substantially behind on longer cases, suggesting that candidate discrimination and extended evidence integration place different demands on a system. In interactive tasks, the proportion of actions judged relevant declines at later steps, and a relevant first question is associated with a correct final answer; in the passive task, models rarely revise their initial hypothesis as new evidence arrives. Together, these findings highlight the value of evaluating abductive reasoning through both final answers and observable evidence use, hypothesis revision, and information gathering.

0.06

Exploration via Exploitation: The Blessing of Reward Diversity in Personalized Federated RL

cs.LG ·2026-11-07 ·Jiin Woo, Gauri Joshi, Yuejie Chi
theory
Abstract

While federated reinforcement learning (RL) has traditionally guaranteed high efficiency in homogeneous environments, collaboration is often hindered when agents possess distinct goals. In this paper, we flip this paradigm, framing reward heterogeneity not as a hurdle to consensus, but as a structural asset that catalyzes exploration. We first propose Personalized Federated Upper Confidence Bound Value Iteration (PF-UCBVI), which achieves optimal linear speedup by decoupling shared dynamics from personalized objectives. To eliminate the risks of explicit exploration, we then introduce Personalized Federated Exploration-Free Value Iteration (PF-EFVI), a purely greedy algorithm that leverages reward diversity to ensure state-action coverage. We prove that PF-EFVI attains logarithmic regret without explicit exploration bonuses under sufficient reward diversity. Our results reveal that agent disagreement shifts the exploration burden from temporal complexity to spatial diversity, enabling efficient collective learning without explicit exploration.

0.05

Constrained Look-ahead Guidance for Interference-Aware Flow Editing

cs.LG cs.CV ·2026-11-07 ·Doudou Zhang, Qi Chen
Abstract

Diffusion and flow models enable powerful generative modeling, but controlled editing remains challenging due to the need to balance source fidelity and edit compliance along the generative trajectory. We propose a principled, inversion-free framework for flow-based image editing from a trajectory centric perspective. Rather than relying on inversion or heuristic guidance, we introduce a one-step look-ahead mechanism that evaluates the effect of candidate updates on both edit and fidelity rewards estimated via denoising, and adjusts the current update based on this evaluation. At each step, our method applies reward-driven corrections guided by the look-ahead evaluation under an explicit fidelity constraint, resulting in stable and coherent trajectory evolution. This look-ahead design mitigates gradient interference between edit and fidelity objectives by anticipating their interaction in future states, leading to improved trade-offs. Experiments demonstrate that the proposed approach achieves stronger semantic edits while better preserving source content compared to existing strong baselines.

0.05

D-REAL: Acceptance-Length Optimization with Unbiased Gradients for Speculative Decoding

cs.AI ·2026-11-07 ·Meng Wei, Yefei Chen, Antong Li, Yanan Gao, Xingliang Shi +5 more
benchmark
Abstract

Speculative decoding efficiency depends on the number of consecutively accepted draft tokens, yet token-level training objectives do not directly optimize this quantity. We introduce D-REAL (Drafter Reinforcement of Expected Acceptance Length), a post-training method that directly optimizes expected acceptance length for a fixed decoding context. D-REAL weights position-wise draft-target distributional overlap by prefix survival probabilities along a draft-sampled prefix, faithfully capturing the sequential dependence of verification on earlier proposals. By accounting for the parameter dependence of the proposal distribution, we derive an unbiased gradient estimator for this single-round objective under stochastic verification. We evaluate D-REAL on DSpark and DFlash2 with Qwen3-8B as the target model across ten benchmarks. Starting from baselines whose mean acceptance length has plateaued, one epoch of D-REAL post-training consistently improves the benchmark-averaged mean acceptance length by up to 0.23 tokens across both architectures under greedy (T = 0.0) and stochastic (T = 1.0) sampling, translating into up to a 9.4% end-to-end serving throughput improvement.

0.05

Language, Vision & Media 128

3D Segmentation in the Clinic: A Grand Challenge

cs.CV ·2026-11-07 ·Bram van Ginneken, Tobias Heimann, Martin Styner
publishedhighly cited
Abstract · 9 pages, 1 table. Published in the proceedings of the MICCAI 2007 workshop 3D Segmentation in the Clinic: A Grand Challenge

This paper describes the set-up of a segmentation competition for automatic and semi-automatic extraction of the liver from computed tomography scans and the caudate nucleus from brain MRI data. This competition was held in the form of a workshop at the 2007 Medical Image Computing and Computer Assisted Intervention conference. The rationale for organizing the competition is discussed, the training and test data sets for both segmentation tasks are described and the scoring system used to evaluate the segmentation is presented.

0.60

Disentangling Channel Semantics in Vision Transformers via Token Decorrelation and Composition-Aware Modulation

cs.CV ·2026-11-07 ·Daeun Kim, Hyesong Choi, Hyejin Park, Dongbo Min
top venuecode/data
Abstract · Accepted at NeurIPS 2026. 32 pages, 12 figures

Multi-channel images encode heterogeneous channel semantics aligned over a shared spatial structure, which requires models to jointly capture globally shared structure and channel-specific variations. Recent multi-channel Vision Transformers attempt to address this challenge by augmenting the [CLS] token with memory tokens to organize multi-channel representations under varying channel compositions. However, we find that these context tokens collapse onto a few dominant channels under high inter-channel redundancy, producing redundant representations and a biased global summary that fails to reliably support composition-aware channel interpretation. We propose MuCa-ViT (Multi-Channel Composition-Aware Vision Transformer), a unified framework that addresses these limitations through two coupled mechanisms. Factorial Token Learning (FTL) enforces token-level decorrelation among context tokens, encouraging the [CLS] token to capture globally shared semantics while memory tokens preserve complementary channel-specific cues. Channel Composition-Aware Modulation (CAM) uses the FTL-refined [CLS] as a semantic anchor to generate sample-wise modulation signals, which enable composition-aware interpretation of channels within a unified backbone rather than relying on static channel embeddings. Experiments on JUMP-CP, CHAMMI, and So2Sat show that MuCa-ViT outperforms prior multi-channel ViTs across all three benchmarks, and the magnitude of improvement aligns with the inter-channel redundancy of each dataset ($ρ\in [0.19, 0.77]$), validating the diagnostic analysis that motivates our design. Our code is available at https://github.com/kimdanni/MuCa-ViT .

0.55

Beyond Bit Matching: Orthogonal Watermarks for Collusion-Resistant Image Fingerprinting

cs.CV ·2026-11-07 ·Zhakshylyk Nurlanov, Tobias Weißberg, Florian Bernard
top venuecode/data
Abstract · Accepted at NeurIPS 2026

Image fingerprinting assigns each distributed copy a user-specific watermark for source tracing. Averaging collusion is a central threat: several recipients can average their differently marked copies to suppress each individual mark. Discrete bit-string watermarks are especially vulnerable because averaging drives bit evidence toward ambiguous decisions. We introduce \textbf{VectorMark}, which maps user keys to near-orthogonal high-dimensional unit vectors and identifies colluders by cosine similarity to the extracted watermark direction. Under ideal averaging of $K$ vector keys, the normalized average preserves a cosine signal of order $1/\sqrt{K}$ for each colluder, while unrelated keys remain concentrated near zero. The spherical geometry of random directions gives a shifted-Beta null distribution for cosine similarities, enabling analytic per-key false-positive-rate control after validation on unwatermarked images. VectorMark uses a neural watermark encoder and centered extractor trained with JND masking and a progressive curriculum for photometric, geometric, and collusion robustness. Experiments on MS-COCO, OpenImages, and AI-generated images show that VectorMark maintains robust watermark attribution under photometric and geometric distortions, and achieves the best strict all-colluder detection among evaluated methods, while preserving high visual quality. Our code is publicly available: https://github.com/nurlanov-zh/VectorMark

0.52

Imperfectly Imperfect: Evaluating Graded Cognitive Diversity in LLM Persona Simulations

cs.CL cs.AIcs.CY ·2026-11-07 ·Sion Weatherhead, Ben Newell, Aaron Belbasis, Flora Salim
top venuecode/data
Abstract · Accepted at NeurIPS 2026 (Evaluations & Datasets Track). 42 pages, 14 figures. Code: https://github.com/sion-w/CogSuite

Large language model (LLM) agents are increasingly used as synthetic participants in social, behavioural, and decision-making research, yet it remains unclear whether they can express graded cognitive diversity as assigned when engaging in simulated behaviour. We introduce a cognition-suite evaluation that tests whether simulated populations reproduce human variation in reasoning, self-reported traits, their associations, and downstream behaviour. Agents conditioned on synthetic or human-derived profiles of cognitive reflection, Need for Cognition, and Big Five personality answer the same suite as a cohort of 473 adults: two newly assembled tests, a Cognitive Reflection Test (CRT) and a Behavioural Reflection Test (BRT) whose decision scenarios measure reflection in behaviour, alongside an established CRT and the two trait questionnaires. Across seven full-battery simulation runs of four models, agents recover Need for Cognition and Big Five trait distributions on self-report questionnaires, but fall short on simulating diversity of cognitive task performance and downstream behaviours. In the four Claude Sonnet 4.6 runs, every agent scores full marks on both CRTs, whatever reflection level its persona was assigned. In the other three runs agents make intuitive-lure errors (the tempting wrong answers CRT items invite), but 49 to 85% score full marks, against under 10% in the human sample. Claude Sonnet 5 comes closest to people's item-by-item pattern of lures; the other two runs give fewer lures and vary less between items. In the BRT scenarios, every run over-produces 8 to 11 of 15 coded behaviours by at least 20 percentage points, and people's links between reflection and behaviour carry over only in part. Useful synthetic populations must preserve not only trait descriptors but the structured variation in reasoning, error type, and behaviour those descriptors are meant to organise.

0.51

MultiTurnRoute: A Platform for Building and Evaluating Multi-Turn LLM Routers

cs.CL cs.MA ·2026-11-07 ·Jiaqi Xue, Mengxin Zheng, Heng Huang, Qian Lou
top venuebenchmark
Abstract · Accepted by NeurIPS 2026

In a multi-turn conversation, the choice of language model at one turn can affect which model should answer the next. The selected model's reply shapes the context for the next query, and its cached prompt prefix can reduce the cost of using that model again. We introduce MultiTurnRoute, a platform for building and evaluating multi-turn LLM routers. A shared interface gives built-in and user-defined routers the current query, dialogue history, and an estimate of each model's reusable prefix length. Our routing algorithm, Chain-of-Route (CoR), uses a learned gate over recent and recurrent dialogue history to predict response quality and estimates cost from each model's reusable prefix. MultiTurnRoute Arena ranks routers on recorded conversations and compares session quality and cost in interactive runs where a simulated user responds to each reply. On a benchmark drawn from WildChat and LMSYS-Chat-1M with per-turn quality annotations for five LLMs, CoR reduces the normalized cost required to match the strongest model's quality by up to 20.3% relative to the strongest routing baseline, while matching or improving peak response quality.

0.51

PhysDNet: Physics-Driven Gradient Amplification for Real-Time Image Dehazing

cs.CV cs.LG ·2026-11-07 ·Rayen Dhahri, Lei Sun, Victor Javier Kartsch Morinigo, Alice Pagnoux, Luca Benini +2 more
top venuebenchmarkcode/data
Abstract · Accepted to NeurIPS 2026. Code, deployment scripts and pre-trained checkpoints: https://github.com/Heatdh/PhysDnet

Modern dehazing networks are often computationally expensive and overfit to synthetic haze, limiting real-time deployment. This underscores an urgent demand for efficient, high-fidelity restoration methods that can operate on resource-constrained edge devices in practical, dynamic environments. We introduce PhysDNet, a lightweight multimodal architecture that fuses NIR imagery with sparse LiDAR depth and is trained with a physics-constrained dual-head objective. Differentiating through a fixed Koschmieder inversion induces density-adaptive gradient amplification, acting as an implicit resource allocation mechanism that concentrates learning on heavily degraded regions without increasing inference cost. PhysDNet delivers superior restoration with minimal capacity and enhanced training efficiency, facilitating real-time deployment on edge devices without increasing inference overhead. On the Seeing Through Fog benchmark, PhysDNet outperforms competitors by 1.67 dB at 5.2 times lower compute while recovering 54.9% of fog-induced detection loss. Ablations and cross-domain evaluations show that physics-guided training improves generalization, including under non-homogeneous haze that violates the Koschmieder assumption. PhysDNet further demonstrates compatibility with fixed-function NPUs, running at 29/45/99 FPS at <4 W across L/M/S variants. Code, deployment scripts and pre-trained checkpoints are available at https://github.com/Heatdh/PhysDnet.

0.50

WordEval: Evaluating Word-Native Operation Fidelity in Document Editing

cs.CV ·2026-11-07 ·Jinyoung Park, Yupan Huang, Shimin Zhang, Jiayu Ding, Yilin Jia +16 more
top venuebenchmarkcode/data
Abstract · Accepted at NeurIPS 2026 (Evaluations and Datasets Track). Project page with code and data at https://aka.ms/word-eval

Microsoft Word is a central environment for office automation, and the faithful execution of its editing operations is an important capability for document agents. Unlike plain-text editing, editing in Word modifies a structured object model, where the requested effect is realized in object properties. Existing benchmarks, however, evaluate office workflows or desktop automation, treating word-processor documents as one task domain among many and verifying outcomes with keyword matching, per-task property checks, or screenshot comparison. We introduce WordEval, an operation-fidelity benchmark that instead centers evaluation on Word editing operations, testing whether each requested edit is realized as intended changes to Word objects and properties. It contains 642 cases across 15 Word feature areas, ranging from text and lists to charts, shape geometry, and SmartArt. 546 atomic cases cover 231 commands, 96 compositional cases combine multiple edits in a single document, and a Word-native oracle checks task-specific object properties and preservation conditions. Cases are constructed from editing-unit specifications, yielding controlled input documents, reference documents, and aligned oracle checks. Each compositional case is linked to corresponding atomic cases, supporting fine-grained diagnosis of editing failures. We evaluate eight agent systems and find that many struggle on the Atomic track, with performance varying across feature areas. Moreover, every system scores lower on Compositional than on Atomic, with gaps that vary substantially by system. Exact atomic-intent links and recorded failure diagnoses provide complementary views of this variation.

0.50

Adapting Large Language Models via Multi-Task Learning for Mental Health Problem, Emotion, and Sentiment Detection

cs.CL cs.AI ·2026-11-07 ·Amir Rafiq Bhat, Rahim Dehkharghani, Zeki Erdem
top venue
Abstract · 33 pages

This study investigates sequential multi-task learning using Parameter-Efficient Fine-Tuning (PEFT), LoRA, and 4-bit quantization on Llama 3.1 8B Instruct for mental health problem detection, emotion detection, and sentiment analysis. Using 4,243 English social media posts, it compares two task sequences and demonstrates that task ordering significantly affects knowledge transfer and task interference. The MHP-first sequence achieves the best mental health problem detection (Micro-F1: 0.7624), while the Emotion-first sequence performs better in emotion (F1: 0.6052) and sentiment analysis (F1: 0.9459). A Joint-MTL baseline further highlights the challenges of simultaneous learning, including task interference and generative format collapse. Overall, the findings demonstrate the effectiveness of sequential PEFT for mental health NLP while showing that task order is critical to balancing performance across related tasks. The integration of LoRA and 4-bit quantization also reduces computational and memory requirements, supporting the adaptation of large language models in resource-constrained environments.

0.50

PLS-GS: Probabilistic Lidar Scaffolds for Compact, Surface-Aligned Gaussian Splatting

cs.CV cs.GRcs.RO ·2026-11-07 ·Krishnan Harikumar, Alejandro Fontan, Thierry Peynot, Emili Hernández, Niko Sünderhauf
publishedcode/data
Abstract · Accepted to WACV 2027. Main paper and supplementary material. Code: https://github.com/k-harikumar/PLS-GS

While 3D Gaussian Splatting (3DGS) is typically optimized from images only, many robotics and mapping applications provide both camera and high-definition LiDAR data, combining rich visual appearance with accurate geometric structure. We introduce PLS-GS, a probabilistic LiDAR scaffold for Gaussian Splatting that exploits this complementary sensing setting to construct a compact 3DGS representation aligned to the measured surface. PLS-GS fits a Hierarchical Gaussian Mixture Model jointly over position and color, yielding a continuous and adaptive scaffold that eliminates the need for discrete scene partitioning such as voxel grids or octrees. The fitted mixture then guides the renderable 3DGS Gaussians: each rendering Gaussian is associated with a GMM component and constrained in the local surface frame induced by that component's covariance. A Mahalanobis support loss bounds the deviation of the rendering Gaussians from the corresponding LiDAR-derived surface. Across five large outdoor scenes, PLS-GS places primitives markedly closer to the LiDAR measurements than both image- and LiDAR-initialized baselines, while employing substantially fewer primitives, yielding a considerably more compact model and faster rendering. At matched capacity, PLS-GS places primitives closer to the LiDAR measurements, with comparable rendering quality. Code is available at https://github.com/k-harikumar/PLS-GS.

0.49

Locating and Repairing Domain Shift in VLM Trajectory Planning

cs.CV cs.RO ·2026-11-07 ·Zhihong Cui, Hengyu Liu, Michael Alexander Riegler, Guandong Xu, Amir Taherkordi +1 more
top venuecode/data
Abstract · Accepted at NeurIPS 2026 (Main Track). 30 pages, 7 figures, 16 tables. Code: https://github.com/ZhihongCui/MoLaRx

Vision-Language Models (VLMs) are increasingly adopted as end-to-end driving planners, yet they suffer from cross-city domain shift. Existing methods treat this shift as a single quantity addressed by uniform alignment or single-site editing. We challenge this view on two counts: (i) domain shift in VLM trajectory planners is inherently structured across modality and layer, and (ii) a locate-then-repair recipe matched to this structure outperforms every uniform-objective baseline. We analyse domain shift via activation patching and uncover two regularities. (F1) Modality axis. Image and text tokens exhibit distinct shift patterns in each VLM. (F2) Layer axis. The layer-wise shift partition is architecture-dependent and dataset-stable. These findings recast domain shift as a two-dimensional tensor indexed by modality (image vs. text tokens) and layer, motivating a simple principle: locate where shift occurs, and repair only there. We instantiate this principle in MoLaRx (Modality-Layer Repair), a locate-then-repair framework. The Causal Importance Score (CIS) estimates the domain shift as a modality-layer tensor and partitions the layers into functional zones. CIS-Guided Selective Repair (CISR) then applies a matched operator (Maximum Mean Discrepancy (MMD) / low-rank adaptation (LoRA) / freezing) to repair each zone, with the composition adapting automatically to each architecture. We evaluate MoLaRx on three architecturally distinct VLMs (LFM2-VL-1.6B, Qwen2-VL-2B, and PaliGemma2-3B) across two cross-city benchmarks (nuScenes and Argoverse 2). MoLaRx attains the lowest cross-city trajectory-error gap among adaptation methods in all six main-table settings, while using 50-60% fewer trainable parameters and 8.4-10.9 ms lower inference latency. In contrast, uniform MMD-only adaptation collapses in all six settings, highlighting the importance of structure-guided repair.

0.47

On the Geometric Bottlenecks of Cross-Lingual Knowledge Editing

cs.CL cs.LG ·2026-11-07 ·Aastha A K Verma, Anwoy Chatterjee, Sparsh Jain, Tanmoy Chakraborty
top venue
Abstract · Accepted at NeurIPS 2026

Transformer-based autoregressive multilingual language models (LMs) are often assumed to store factual knowledge in an interlingua: a language-agnostic latent space. If this assumption holds, a parameter update applied in a source language during knowledge editing should inherently propagate to its semantic equivalents in target languages. However, empirical benchmarks show that cross-lingual propagation of edits often degrades substantially, a degradation that prior work treats largely as an algorithmic symptom rather than a structural bottleneck. In this work, we investigate whether representational anisotropy creates a structural bottleneck for source-only associative-memory editing, and perform a geometric dissection of the cross-lingual propagation gap. Using associative memory-based "locate-then-edit" methods, including ROME, MEMIT, and AlphaEdit, as analytical frameworks, we derive a closed-form decomposition of this gap into key-space propagation, parallel residual mismatch, and orthogonal residual mismatch. Our empirical findings show that cross-lingual propagation is primarily bottlenecked by a large orthogonal residual component, indicating that source-language edits often fail to span the target-language update directions required for successful transfer. Furthermore, we show that post-hoc correction using only source-language residual statistics encounters a large residual-prediction barrier. Finally, we perform a structural realignment via a targeted contrastive alignment objective and find that it substantially improves key-space propagation and partially improves cross-lingual edit transfer, while leaving value-space mismatch as a remaining bottleneck.

0.46

Permute-then-Adapt: Weak-to-Strong Contrastive Image-Text Adaptation

cs.CV ·2026-11-07 ·Jinhao Li, Sarah Erfani, Lei Feng, Guangrui Li, James Bailey +1 more
top venuecode/data
Abstract · 28 pages; accepted at NeurIPS 2026 (Spotlight)

Image-text alignment models such as CLIP are typically trained with contrastive learning, where unpaired examples are treated as strict negatives. While this assumption introduces noise by penalising semantically related pairs, existing solutions often rely on heuristic similarity thresholds that lack statistical grounding and sensitivity to dataset-specific noise distributions. In this paper, we propose Permute-then-Adapt (PTA), a framework that leverages efficient weak models to identify missed positives while controlling Type-I error relative to a conditional permutation reference, up to calibration approximation error. Unlike standard distillation or arbitrary thresholding, PTA estimates a random-pairing reference distribution from the weak teacher's similarity scores and uses its upper tail, determined by $α$, to select auxiliary positives. The calibrated positive set is then trained against by a single multi-positive contrastive objective: each anchor maximises the total probability mass it assigns to its positive set. Offline calibration with the weak model adds negligible training overhead. Experiments show gains over heuristic soft-label approaches in object recognition, cross-modal retrieval, and data efficiency. Project repository: https://github.com/JinhaoLee/Permute-then-Adapt.

0.45

Can small transformers learn the language of Coxeter groups?

cs.CL cs.LGmath.GRmath.RT ·2026-11-07 ·J. Elisenda Grigsby, Yiyao Huang, Arushi Katyal, Ryan Lin, Pranav Ramakrishnan +1 more
published
Abstract · 13 pages, 13 figures, accepted as a poster at the 6th workshop on Mathematical Reasoning and AI; NeurIPS 2026, Atlanta; This version is upda…

The field of mechanistic interpretability is well-suited to collaborative projects between mathematicians and AI researchers. This work describes one such project. We trained a small transformer to learn the language of the Coxeter group affine A2, best known as the symmetries of the tiling of the Euclidean plane by equilateral triangles. We then formulated and proved a word-theoretic conjecture about affine A2 after observing the attention patterns of the trained model.

0.44

Matchability-Aware Conformal Prediction for Open-Ended Language Model Generation

cs.CL stat.MLcs.LGstat.ME ·2026-11-07 ·Ilia Azizi
top venuecode/data
Abstract · Accepted at NeurIPS 2026. Project page: https://unco3892.github.io/caelum/

Large language models (LLMs) often do not produce calibrated uncertainty estimates for their answers. We propose CAELUM, a framework that builds semantic prediction sets while separating aleatoric from epistemic uncertainty. CAELUM samples answers using an ensemble of clarification prompts, clusters them by semantic equivalence, and estimates the aleatoric and epistemic components via kernel-based von Neumann entropy before calibrating nonconformity scores with split conformal prediction. We use a weight $w_λ(x)=1+U_{\mathrm{ale}}(x)+λ\,U_{\mathrm{epi}}(x)$ to control for the contribution of the epistemic uncertainty of each instance and fit a coefficient $λ^\star$ to minimize prediction set size. When the objective is nonflat and selection is stable across splits, $λ^\star$ could be a diagnostic, where large values may indicate epistemic dominance (retrieve, abstain) and $λ^\star\to 0$ aleatoric dominance (clarify). For individual queries, a log ratio $R(x)$ compares the weighted epistemic component with the aleatoric component. Theoretically, we demonstrate that the coverage is capped by matchability, the fraction of items whose reference answer appears in any sampled cluster. Empirically, this matchability can be predicted from the original question features using a lightweight classifier. We use those predictions to define a stratum and derive a conformal coverage bound within it. Across 30 LLM and dataset combinations (4-32B), mean coverage is 0.74 in the learned stratum versus 0.65 overall, averaged over the 78 evaluations using the learned classifier. Project page: https://unco3892.github.io/caelum/

0.44

VAComp: View-Aware Full Projector Compensation

cs.CV ·2026-11-07 ·Kai Zhou, Qingyue Deng, Bingyao Huang
code/data
Abstract · 14 pages, 13 figures

Full projector compensation adjusts a projector's input image to compensate for both geometric and photometric distortions introduced by the projection surface. Existing methods generally treat the viewing condition as a static camera viewpoint. When applied to a moving viewer, existing methods face two key challenges: (1) view-dependent methods require model retraining or recalibration to compensate for novel viewpoints; and (2) NeRF/GS-based view-agnostic methods use structure from motion (SfM) to calibrate the projector-camera systems (ProCams) with an unknown global scale, while view-aware tracking requires calibrating a camera oriented opposite to the projector, which introduces an additional calibration challenge. To address these challenges, we introduce VAComp, the first view-aware framework for full projector compensation for a moving viewer in a naked-eye setting. Using a projector, VAComp enables the viewer to perceive a virtual object that visually replaces the real object or to observe view-consistent modifications of the real object's appearance and material properties. To resolve the SfM scale ambiguity, we use metric calibration as a reference to recover the global scale; to calibrate the oppositely facing camera, we use two checkerboards, one facing the RGB camera and the other facing the depth camera. To balance compensation quality and runtime efficiency, we develop two non-iterative hybrid methods, VAComp-SD and VAComp-SI. Both combine 2D Gaussian-based view-conditioned geometric correction with learned photometric compensation, predicting each compensation image through a single forward pass without per-frame iterative optimization. Real-world experiments show that both methods achieve competitive compensation quality, with VAComp-SD enabling real-time updates and VAComp-SI operating at interactive rates. Project page: https://realkaizhou.github.io/VAComp/.

0.44

They Can See It but Not Say It: Iterative Self Knowledge Re-expression in Visual Reasoning Relative Pose Identification

cs.CV ·2026-11-07 ·Mengyu Wang, Ken Deng, Yftah Ziser, Shay B. Cohen, Tiejun Ma
top venuecode/data
Abstract · Accepted as NeurIPS 2026 Oral

The task of visual reasoning relative pose identification (VRRPI) has exposed critical weaknesses in the multi-view 3D spatial reasoning capabilities of modern vision-language models (VLMs). We present a systematic analysis of VLMs' behavior in VRRPI, uncovering stable output biases where VLMs disproportionately predict specific motion patterns regardless of visual evidence. To mitigate these inherent biases, we propose self knowledge re-expression with iterative debiasing (SKR-ID), an annotation-free pipeline that adapts VLMs for VRRPI using only randomly extracted unannotated image pairs. SKR-ID transitions VLMs' output mechanism from generic next-token prediction to task-optimized classification heads. The pipeline proceeds in two stages: (1) a cold-start initialization using debiasing object-centric prompts to anchor motion inference in 2D visual evidence; and (2) a recursive logit-ranking mechanism that enforces statistical parity by using the median logit value as a dynamic threshold. Experimental evaluation across frontier VLMs on standard benchmarks demonstrates that SKR-ID significantly mitigates systematic biases. Our method improves open-source VLMs' Macro-F1 scores from 35.0-45.4% to 53.7-58.0%, without using any annotated data. These results show that SKR-ID can effectively improve 3D reasoning performance from a severely collapsed baseline. The code is publicly available at https://github.com/astudentuser/Codes_of_SKR-ID.

0.44

SRA: Spatial Reasoning Adapter via Evolving Social Interaction Graphs for Trajectory Prediction

cs.CV ·2026-11-07 ·Jaewoo Jeong, Seonkyu Song, Hyeonwoo Park, Jegyeong Cho, Yujin Bae +3 more
top venuecode/data
Abstract · Accepted to NeurIPS 2026

Predicting where multiple agents will move next is fundamental to autonomous driving, robotics, and any system that must share space with people safely. Modern stochastic predictors read off inter-agent interactions once from the observed past and keep that view fixed throughout the generative rollout, so the reciprocal geometric constraints that emerge as the joint future sharpens are not re-examined. We address this gap with SRA (Spatial Reasoning Adapter), a modular graph denoiser that is integrated into existing diffusion and flow-matching predictors without redesigning their generative cores. At every reverse step, SRA reads the host's current future estimate, builds a sparse Euclidean interaction graph on the predicted trajectories, and returns a residual correction to the host's decoding path. Our adapter consists of three core components: a sparse neighborhood retrieved with a learned query-key score and a geometric bias, a future-conditioned relational feature recomputed on the evolving prediction, and uncertainty-weighted message passing where confident agents act as anchors for uncertain ones. We instantiate SRA on three structurally different hosts (LED, MID, and MoFlow) across four multi-agent benchmarks: SDD, NBA, Soccer, and Football. SRA improves ADE for every host on every dataset, with average gains of 5.8% in ADE and 7.0% in FDE. A cumulative ablation shows that the three components act as one mechanism: only together do they improve every host on every sports dataset, indicating that spatial reasoning over the model's own evolving future, weighted by how far that future can be trusted, is broadly useful across denoising architectures. Code is available at https://github.com/Jaewoo97/SRA.

0.43

Pole or Vector? Encoders Capture Bangla Compound Verbs, but LLMs Read Them Literally

cs.CL ·2026-11-07 ·Md. Asaduzzaman Shuvo, Md Mehrab Karim Antor, Md. Adnanul Islam, Shahanur Rahman Bappy, Sajid Ahamed +2 more
benchmarkcode/datahot direction
Abstract

In Bangla, a main verb is often followed by one of a small set of vector verbs. These verbs lose their own meaning and add a shade of meaning instead, such as completion. For example, likhe phello means that someone finished writing, not that they wrote and threw something. The same two verbs can also describe two separate events, so only the context shows which meaning is meant. This makes compound verbs a hard test of how language models handle multiword expressions. We present BanglaCV-Probe, a benchmark of 17 vector verbs with 1,044 minimal pairs, 204 entailment items and 197 pairs of correct and incorrect combinations, all checked by native speakers. We test 26 models (encoders, LLMs and translation systems) on four skills: telling the two meanings apart internally, understanding what a vector adds, preferring correct combinations and keeping the vector's meaning in translation. Probes on five encoders tell the two meanings apart and still reach up to 93% accuracy when trained on pairs without surface cues. Prompted models rarely use this knowledge. Entailment accuracy stays between 59% and 81%. Several LLMs answer Yes or No to almost every item, and nine of ten models miss that someone who ate up the rice left none. Most instruction-tuned LLMs translate vectors literally in 15-20% of vector sentences and in up to 55% for mora (die). On correct combinations, a small Bangla-only model (88%) matches BLOOM-7B1 (87%) and both beat Qwen2.5-7B (74%). Data and code: https://github.com/ashuvo25/Pole_OR_Vector

0.43

Distinguishing Extraction Failures and Clinical Discordance with LLM Uncertainty on Radiology Reports

cs.CL ·2026-11-07 ·Jerjes Aguirre-Chavez, Shant Malkasian, Roshun Sankaran, Brendan Crabb, Albert Hsiao
top venue
Abstract · Accepted as a spotlight presentation at ML4H 2026, published at Proceedings of Machine Learning Research 346

Large language models (LLMs) can structure clinical datasets, but extraction error must be distinguished from discordance in the underlying data. From a real-world cohort of 1{,}045 patients undergoing coronary CT angiography (CCTA), we evaluated whether LLM uncertainty during vessel-level coronary artery disease (CAD) stenosis extraction could identify extraction failure and clinical discordance. References were expert CCTA annotations (253 patients) and subsequent invasive coronary angiography (ICA; 729 patients). We quantify extraction failure as disagreement between stenosis severity in LLM extracted versus expert annotations, and clinical discordance as disagreement in stenosis severity between CCTA and subsequent ICA. Stenosis Margin, selected from ten candidate measures, was used to quantify LLM uncertainty, and generalized estimating equation logistic regression and AUC evaluated its association and incremental discrimination for each failure mode. LLM uncertainty was strongly associated with extraction failure (adjusted OR 4.15 [95% CI 2.57--6.69], $P<0.001$) and substantially improved discrimination beyond LLM-extracted CCTA stenosis severity alone (AUC 0.81 to 0.86; $Δ$AUC $+0.06$ [$+0.03$, $+0.09$], $P=0.001$). In contrast, LLM uncertainty weakly associated with clinical discordance (adjusted OR 1.22 [1.04--1.43], Holm-adjusted $P=0.118$) but provided no incremental discrimination beyond LLM-extracted CCTA stenosis severity ($Δ$AUC $+0.01$ [$-0.00$, $+0.02$], $P=0.245$). Moreover, extraction errors and clinical discordance showed little overlap, with 38 of 42 clinically discordant vessels (90%) having been extracted correctly. LLM uncertainty may identify extraction failures without identifying underlying clinical discordance, supporting failure attribution when evaluating LLM-derived clinical datasets.

0.43

Same Contacts, Different Rooms:Meeting Format Shapes LLM Group Influence Beyond Pairwise Contact Counts

cs.CL physics.soc-ph ·2026-11-07 ·Zhexi Gu, Shusheng Cao, Runping Zhu
code/data
Abstract · 10 pages, 5 figures, 9 tables. Code: https://github.com/ZachGu-00/llm-meeting-format

Simulations of LLM agent collectives usually summarise interaction as a pairwise contact network. We test that summary with a contact-matched intervention: the members, their initial stances and every pairwise contact count stay fixed while the meetings are reorganised, and each member gives a one-step support readout after reading the arguments of everyone present. Across six open-weight LLMs and 14 debate topics, merging triads into five-person meetings (resize) moves group support more than reshuffling who meets in triads (regroup), although both move the same 60 dyadic exposures. The root-mean-square change is 4.36 versus 2.59 percentage points (pp), larger for resize in all six models and in 81 of 84 model-topic cells. Resizing also has a consistent direction: it amplifies the initial majority of the merged members by 1.72 pp, in every model, whereas regrouping has no direction (<= 0.03 pp). A contact network cannot represent regrouping at all, yet pairwise rules applied meeting by meeting explain up to 76% of its effect, and with identical weights they raise resize R^2 by 0.51-0.58 for two models. We show that the gap between the two placements is an exact Jensen gap of the sigmoid readout, bounded by how much a member's predicted logits vary across meetings; this dispersion tracks the gap with log-log r=0.97 and tells a modeller when contact counts are a safe summary. Meeting format is a first-class variable for LLM social simulation.

0.41

DAPointRWKV: Domain Adaptive Vision RWKV for Point Cloud Completion

cs.CV ·2026-11-07 ·Yinghui Li, Qianyu Zhou, Hao Yang, Richard Dazeley, Ye Zhu +1 more
top venue
Abstract · This paper has been accepted by 34th ACM International Conference on Multimedia

Domain adaptive point cloud completion (DA PCC) aims to bridge the geometric discrepancy between labeled source domains and unlabeled target domains for accurate 3D shape completion. Existing methods based on CNNs, Transformers, and Mamba-like sequence models still face challenges from limited receptive fields, high computational complexity, insufficient geometric bias, and serialization sensitivity. Recently, Receptance Weighted Key Value (RWKV) provides a linear-complexity alternative for long-range sequence modeling. However, adapting RWKV to DA PCC remains challenging due to the lack of explicit geometric priors, the mismatch between recurrent modeling and coarse-to-fine generation, and unstable target-domain adaptation. To address these challenges, we propose DAPointRWKV, a novel RWKV-based framework for DA PCC. DAPointRWKV consists of three components: Geometry-Aware RWKV (GA-RWKV) Backbone, RWKV-Oriented Progressive Decoder (RWKV-PD), and Dual Structural Regularization (DSR). Specifically, GA-RWKV introduces geometry-aware recurrent representation learning, RWKV-PD bridges RWKV token representations with coarse-to-fine reconstruction through local token-guided refinement, and DSR stabilizes target-domain transfer through visible-structure anchoring and complete-shape hypothesis stabilization. Extensive experiments on synthetic and real-world benchmarks demonstrate that DAPointRWKV consistently outperforms state-of-the-art methods.

0.41

Anatomy-Activated Mixture-of-Experts for 3D Medical Vision-Language Pre-training

cs.CV cs.AI ·2026-11-07 ·Szymon Płotka, Gizem Mert, Pedro R. A. S. Bassi, Wenxuan Li, Zongwei Zhou +7 more
publishedcode/data
Abstract · NeurIPS 2026

Medical Vision-Language Pre-training (VLP) leverages the semantic richness of radiology reports to learn 3D representations, yet current architectures remain fundamentally anatomy-agnostic. While clinical diagnostics rely on organ-specific context, standard encoders apply a uniform, dense parameterization to all volumetric patches, and existing Mixture-of-Experts (MoE) routers gate on isolated local features without explicit access to the organ composition of the scan or crop. We propose Anatomy-Activated Mixture-of-Experts (A$^{2}$MoE), a framework that incorporates anatomical inductive biases within the model's computational path. A$^{2}$MoE introduces two key components: (1) a Histogram Representation Router (HRR) that conditions expert selection on global visual context and a predicted anatomical composition histogram, encouraging anatomy-dependent expert weighting; and (2) an Organ-Query Multi-Scale Projector (OQMSP) that utilizes learnable queries to extract organ-level embeddings for fine-grained alignment without requiring manual annotations at inference. By routing at the patch level rather than the token level, A$^{2}$MoE reduces routing complexity by orders of magnitude while satisfying load balancing by construction. Pre-trained on over 60,000 multi-phase CT scans, A$^{2}$MoE achieves state-of-the-art performance across segmentation, classification, and detection benchmarks. Furthermore, we demonstrate that our anatomically grounded inductive biases support robust generalization to MRI and ultrasound, establishing a scalable foundation for medical representation learning. The code is available at https://github.com/gmum/A2MoE.

0.41

Beyond Row Alignment: Virtual-Camera-Aware Online Stereo Rectification

cs.CV ·2026-11-07 ·Yizhao Peng, Rui Gong, Zaiwang Gu, Xudong Jiang, Jun Cheng
publishedhot direction
Abstract · Accepted as poster in NeurIPS 2026

Stereo matchers estimate depth by comparing the left and right images from a stereo camera pair, but they assume that the two cameras are accurately calibrated and rectified. In real deployments, small physical shifts after calibration, caused by camera motion, mechanical drift, vibration, or installation changes, can break this assumption and degrade depth estimation. Online stereo rectification is therefore needed to correct the image pair during operation and recover a stereo geometry where standard stereo matching can work reliably. Recent online rectification methods mainly optimize row alignment, making the left and right views of the same scene point fall on the same image row. We show that this target, while necessary, is insufficient for deployable stereo depth estimation. A rectifier may make a small set of matched points look well row-aligned, but still distort the full images in ways that hurt stereo matching: useful image regions may be cut off, large blank areas may appear, objects may be stretched or squeezed, or image scale may change unevenly across the view. We propose VirtualRect, an online stereo rectification method designed for reliable real-world deployment. Rather than treating rectification as simply making the left and right image rows line up, this method estimates a rectified virtual stereo camera and combines image-wide supervision of image warp behavior. This makes the rectified images and the camera parameters come from the same geometry, so the images used for matching and the model used for depth computation remain consistent. Experiments on multiple datasets show that, compared with the prior state-of-the-art online rectifier, VirtualRect reduces vertical flow error by 17% on Carla-Flowguided and 7% on Semi-Truck Highway, and lowers depth MAE by 30-41% on Carla-Flowguided across three downstream stereo matchers.

0.40

Action Transfer Across Character Body Plans in Latent-Action World Models: An Empirical Study of Animated Characters

cs.CV cs.GRcs.LG ·2026-11-07 ·Amarnath R
benchmarkcode/data
Abstract · 28 pages, 13 figures, 8 tables. Dataset: https://huggingface.co/datasets/amarnathresearch/synthetic-2d-body-plans

Latent-action world models learn actions from video without labels, and recent methods aim to share these actions across bodies. Transfer between 2D characters with different body structures remains underexplored. In this paper, we study whether an action learned from one character can make a character with a different body perform the same action. A cat, a blob and a human can all jump, but each moves its body in its own way. The usual transfer gain mixes body motion with screen movement and can assign a perfect score to a model that ignores its action. To address this, we introduce a transfer score that compares the correct action with wrong ones from the same source. We test five training objectives on a new synthetic benchmark of 2D cartoon characters with six body plans, the standard Sprites benchmark, and 431 selected animal and human animations from DeformingThings4D. With a fixed camera, a baseline that only copies the source's screen movement scores 0.71, as high as most learned models. In contrast, with a camera that follows the body, the best model reaches only 0.46 for an unseen animal family and 0.20 for humans. With this camera, a linear probe identifies the action class of unseen animal families with 66% to 86% balanced accuracy (chance 17%). The learned actions therefore retain information about which action is performed, but the correct action closes only part of the prediction error gap for a new character.

0.40

SCorPion: A Strategic, Controllable and Personified Conversational Companion

cs.CL ·2026-11-07 ·Luo Ji, Xiyuan Chen, Dan Wang, Qingqing Gu, Teng Chen +10 more
published
Abstract · NeurIPS 2025 PersonaLLM Workshop

Fixed prompt-based language models face challenges in open-domain conversations, where versatile topics require different expert advice or demonstrations to maximize their universal performance. Motivated by the dynamic prompt studies in language generation, we extract different chatting strategies from varied forms of text corpus, which steer the language model to different styles based on human principles. We design and deploy a conversational companion called SCorPion, which uses GraphRAG to extract and summarize the knowledge graph globally, yield community-summarized strategies, and produce query-focused optimal strategies. SCorPion has strategic thinking, yields personified and stylized responses, and can be controlled by explicit factors. Code and further results can be found in https://bachozean.github.io/scorpion-page/.

0.40

REDiff-AD: No Training and No Target Masks for Industrial Anomaly Image Generation via Diversity-Driven In-Context MMDiT

cs.CV ·2026-11-07 ·Chih-Yu Tsai, Chih-Heng Chang, Jian-Jiun Ding
publishedcode/data
Abstract · Accepted to WACV 2027

Industrial anomaly detection is challenged by the scarcity and diversity of real defects, leaving insufficient anomalous data to effectively train downstream models. Existing anomaly generation methods often require product-specific fine-tuning, target mask annotation, or costly procedures to achieve realistic and diverse defects. We propose REDiff-AD, a framework that requires neither model training nor target masks and introduces Rectified Flow MMDiT into industrial anomaly generation for efficient sampling with fewer generation steps. Given a normal product image and a single masked reference anomaly, REDiff-AD employs in-context learning to transfer the defect appearance from the reference into the normal image. Internal attention signals generated during this process are further reused to localize the synthesized anomaly, while edge and color cues refine the localization into pixel-level pseudo-masks. Two lightweight modules further improve generation quality: adaptive reference conditioning enhances defect fidelity, while diversity guidance promotes varied defect patterns without additional training. Experiments demonstrate that REDiff-AD can well address the tradeoff between realism and diversity for anomaly pattern generation and achieve competitive downstream anomaly classification, detection, and localization performance. These results show that REDiff-AD is a simple, efficient, and effective solution for large-scale anomaly data augmentation and pixel-level data construction. Github code: https://github.com/efdefe4-droid/REDiff-AD

0.39

Decoupling Direction and Magnitude: Language-Steered Flow Matching for Super-Resolution in the Dark

cs.CV ·2026-11-07 ·Jiaxin Gao, Ziyu Yue, Yaohua Liu, Danchen Cui, Zhihui Zhao +1 more
publishedcode/data
Abstract · Accepted at NeurIPS 2026 (poster). 32 pages, 12 figures, and 13 tables. Code and pretrained models: https://github.com/moriyaya/DeLang-SR

Image super-resolution in the dark is fundamentally challenged by extreme spatial heterogeneity: severely underexposed and noisy regions demand aggressive generative enhancement to reconstruct missing details, while relatively well-exposed areas require conservative updates to prevent hallucinated textures. Standard diffusion and flow-matching models, however, rely on globally uniform integration trajectories, making them sub-optimal for such spatially varying degradations. In this paper, we propose DeLang-SR (Degradation-Language steered Super-Resolution), a one-step flow-matching framework that explicitly decouples the restoration direction and magnitude. Rather than treating low-light corruption as arbitrary latent features, DeLang-SR translates measurable physical statistics into structured degradation language, harnessing the semantic prior of pretrained text-to-image models. The degradation-language intent, complemented by image-specific visual tokens, steers the velocity field (direction) toward restoration-relevant manifolds. Simultaneously, a continuous spatial update-strength field acts as a locally varying Euler step-size field (magnitude), applying stronger integration steps in heavily degraded shadows while enforcing conservative updates in reliable regions. Experiments on paired and real dark images, together with ablations and diagnostic analyses, show that DeLang-SR provides an effective and interpretable one-step route to super-resolution in the dark. Code and pretrained models are available at https://github.com/moriyaya/DeLang-SR.

0.39

Perception, Not Reasoning, Limits Visual Theory of Mind

cs.CV cs.AIcs.CLq-bio.NC ·2026-11-07 ·Mohamed Rayan Barhdadi, Syed Talal Wasim, Juergen Gall, Erchin Serpedin, Hasan Kurban
top venueposition
Abstract · 36 pages, 6 figures. Accepted to NeurIPS 2026 (E&D). Code and dataset: https://kurbanintelligencelab.github.io/perception-not-reasoning/

Vision-language models pass text-based false-belief probes at near-human levels but drop by over twenty percentage points on abstract panels the moment the same problem is conveyed visually. The consensus diagnosis is conceptual, prescribing new belief-reasoning modules. We argue the dominant bottleneck is perceptual: not what the model cannot reason about, but what it cannot see. We propose a five-condition dissociation protocol that holds the reasoning task constant while varying perceptual fidelity from visual only to text only, and quantify the result with the Perceptual Contribution Ratio, the share of the visual-to-text gap a given fidelity closes. We test this on two kinds of scenes: abstract 2D panels in the Heider--Simmel tradition, and naturalistic 3D game scenes from MuMA-ToM and MindPower. The evaluation covers seven VLMs and roughly 159,000 API calls. Ground-truth scaffolds substantially close the gap in every setting, and on the abstract panels and MuMA-ToM they match or exceed the text-only reasoning. Whereas scaffolds built by automatic parsers, including frontier VLMs, fail. The most common error in the visual-only condition is \emph{reality bias}, which accounts for 60\% of false-belief failures. Models report where an object actually is, not where an absent agent believes it is. This points to a perceptual front-end that fails to track which agents were present at each moment. Under the evaluated protocol, structured visual scene-state extraction, rather than belief inference, is the dominant bottleneck for visual ToM.

0.38

Slim Foundation Models for Latency-Critical Recommendation Serving

cs.IR cs.AI ·2026-11-07 ·Danyal Aftab, Anas Zafar, Assadig Abakr, Steven Davy
published
Abstract

Click-through rate (CTR) prediction is a core component of recommendation and advertising systems. Large language models (LLMs) offer rich semantic priors that improve prediction quality, particularly in cold-start and long-tail settings. Injecting a full-scale LLM into a CTR pipeline is, however, prohibitively expensive at inference time: existing LLM-enhanced CTR methods avoid this cost by caching LLM outputs offline rather than serving the LLM online. We propose a resource-efficient pipeline, SlimFM, that makes online LLM-based semantic modeling practical: (1) dependency-aware structured pruning compresses LLaMA3-8B by 44% to a 4.5B-parameter backbone; (2) LoRA-based field-aware fine-tuning trains the pruned model to produce discriminative field-level semantic embeddings from lightweight textual prompts; (3) a semantic-guided fusion module injects these embeddings into the feature interactions of a conventional CTR model. Across four datasets and six CTR backbones, SlimFM retains over 99% of uncompressed-model accuracy while reducing inference latency by up to 3.4x relative to existing LLM-enhanced baselines.

0.38

STAR-VF: Spatio-Temporal Aware Representation Condensation for Data-Efficient Video Fusion

cs.CV ·2026-11-07 ·Yuchen Xian, Yunqiu Xu, Guosheng Lin, Yang He
top venuecode/data
Abstract · Accepted at ACM Multimedia 2026 (MM '26)

Video fusion aims to integrate complementary cues from heterogeneous sources, producing visually informative videos with coherent spatio-temporal structures. Jointly learning cross-modal complementarity and temporal dependencies typically relies on large-scale, high-quality video datasets, whose storage and repeated use during optimization incur substantial costs, making data-efficient learning highly desirable. To address this challenge, we rethink video fusion from the perspective of dataset condensation and propose STAR-VF, the first framework that condenses spatio-temporal representations for data-efficient video fusion. STAR-VF condenses the original training set into a compact set of learnable spatio-temporal representations, which are decoded into synthetic video clips and optimized using feature-level supervision provided by a frozen proxy model. The resulting condensed data preserves fusion-relevant spatial and temporal cues without requiring ground-truth fused videos as supervision. To adapt fusion learning to condensed representations, we equip the final fusion model with flow-free inter-frame temporal modeling and query-driven intra-frame aggregation, enabling direct temporal interaction and adaptive selection of fusion-relevant information in representation space. Extensive experiments on infrared-visible, multi-exposure, multi-focus, and medical video fusion benchmarks demonstrate that STAR-VF, trained with only approximately 1/16 of the original temporal training set, achieves fusion quality comparable to methods trained on the full data. Further analyses validate the contributions of learned condensation and flow-free temporal modeling, and demonstrate the robustness of the condensed supervision across different proxy encoders and downstream fusion models. Project page: https://zju-xyc.github.io/STAR-VF-Project-Page/.

0.38

AmodalBox: Automated 3D Bounding Box Annotation under Partial Visibility

cs.CV ·2026-11-07 ·In-Jae Lee, Namtae Kim, Robert Exley, Kwonyoung Ryu, Jaesik Park
publishedcode/data
Abstract · Accepted by CoRL 2026

Recent progress in vision foundation models (VFMs) has made automated open-vocabulary 3D bounding box annotation increasingly feasible by jointly leveraging LiDAR point clouds and images. However, partial observations remain a major challenge in sparse driving scenes, where visible LiDAR points often cover only a small fraction of the object's surface and provide insufficient cues for estimating its full spatial extent. As a result, existing methods often produce mislocalized boxes or leave objects unlabeled. To address this limitation, we propose \textbf{\Ours{}}, an automated annotation framework that leverages image-generated 3D assets as amodal geometric priors for partially observed objects. By establishing correspondences between input images and rendered views of 3D assets, \Ours{} robustly aligns canonical amodal geometry with partially observed world-space LiDAR point clouds. To improve temporal consistency and reduce computational cost, \Ours{} generates a single asset per instance, estimates its metric scale once at a single frame and propagates it across frames for dynamic instances, and aggregates multi-frame observations for static instances. 3D object detectors trained on annotations from \Ours{} improve performance over strong baselines by 14.1 mAP on the nuScenes dataset and by 12.45 AP at IoU 0.5 for the vehicle class on the Waymo Open Dataset (WOD). Our project page is available at: \href{https://oliver0922.github.io/AmodalBox/}{https://oliver0922.github.io/AmodalBox/}

0.37

Large Language Models without QK-Normalization

cs.CL cs.AIcs.LG ·2026-11-07 ·Yongding Tao, Yitong Qiao, Tianqi Shao, Tian Wang, Zhixuan Chu +3 more
code/data
Abstract · Code is available at https://github.com/yongding-tao/QK-MAP

Query-Key Normalization (QK-Norm) is widely adopted in modern large language models to control the growth of attention logits and improve training stability. However, query and key norms are not merely sources of instability. The query norm controls the sharpness of the attention distribution, while the key norm determines how strongly a token attracts different queries. These norms therefore carry token-specific information about attention selectivity and token importance. By normalizing both, QK-Norm suppresses this information. This raises a natural question: can we control query-key representations without discarding their sensitivity to magnitude? To address this question, we propose QK-MAP, a Magnitude-Aware Projection for constructing queries and keys. Rather than removing magnitude through normalization, QK-MAP transforms the input representation using multiple learnable frequencies, so that changes in magnitude are encoded as bounded phase patterns. This allows magnitude to be represented through phase and direction rather than vector length alone. Its periodic features have a fixed norm while remaining sensitive to magnitude changes, separating magnitude encoding from norm growth. Multiple frequencies further help distinguish magnitudes and reduce periodic ambiguity. Experiments across model scales from 190M to 7B parameters show that QK-MAP achieves lower training and validation loss than both Standard attention and QK-Norm. Further analyses show that QK-MAP preserves sensitivity to input scale, assigns less attention to the first token, and produces stronger attention contrast for relevant evidence in synthetic long-context tests. These results demonstrate that query-key scale can be effectively controlled without sacrificing token-specific magnitude information.

0.37

LEGIT: Live Eligibility Gating over Identifier Tries for Constrained Decoding of Heterogeneous Pages

cs.IR cs.LGcs.PF ·2026-11-07 ·Felix Fang, Zheng Liu, Robert Saliba
hot direction
Abstract · 22 pages, 11 figures, 18 tables. Under review

A generative recommender names each candidate with a short token sequence, a Semantic ID, and recommends by generating those tokens. Some generative recommenders, including the Uber Eats search engine we are building, write a whole results page, stores with their menu items, in one decoding pass. Every Semantic ID on such a page must name a real store or item that is eligible for the search, such as a store that delivers to the user's location, and that is not already on the page. Each item must also belong to the store above it. A trie over the catalog keeps identifiers real, but eligibility and repetition change with every request and token, and the grammar-constrained decoder we tried could not enforce these changing rules exactly and quickly. We call this constraint LEGIT, for live eligibility gating over identifier tries. Numbering trie leaves depth first turns each rule into a range lookup over one store trie and one item trie, and every generated page passes an independent catalog check. On one NVIDIA B200 GPU with one or two requests per decode step, a direct implementation that walks the trie on the GPU makes each token about 5 times as slow as unconstrained decoding. Across 36 implementations, hardware counters and ablations are consistent with kernel launches and CPU-GPU waits as the source of this slowdown. Our fastest implementation walks the trie on the CPU while the GPU runs the model, needs no custom kernel, and cuts time per output token by 65.9 percent against the direct GPU implementation, to about 1.7 times the unconstrained cost. This implementation is also the fastest on a page layout that puts each item before its store.

0.37

MotionGraphicsBench: Evaluating Motion Graphics Generation with Programmatic Verification

cs.CV cs.AI ·2026-11-07 ·Jiaju Ma, Ashvini Kumar Jindal, Chuhua Wang, Sumithra Bhakthavatsalam
benchmarkcode/data
Abstract

Large language models can generate motion graphics from static images and natural language requests, but it remains difficult to evaluate how well models can follow requests that combine spatial, appearance, and temporal requirements. We introduce MotionGraphicsBench, an expert-curated benchmark of 100 instances for animating an input static SVG with a motion edit prompt. By supplying all required visual elements, our editing task reflects a designer-like workflow and isolates motion generation from graphic generation. Derived from real-world Lottie motion designs, our benchmark contains diverse combinations of spatio-temporal animation behaviors. Each benchmark instance pairs its model input with a deterministic, human-readable verification program that accepts different valid implementations and reports which prompt requirements pass or fail, rather than comparing against a reference animation or using a VLM judge. We construct these programs with a multi-agent pipeline using human-labeled animations and synthesized test cases. On a balanced test set of 180 animations, our programs achieve 93.3% raw agreement with human judgments, compared with 83.3% for a GPT-6 Astra VLM baseline, and outperform the baseline in both approval and rejection recall. Across 47 configurations from 32 LLMs and 12 providers, pass@1 ranges from 6.0% to 67.7%. Across 15 models evaluated at their highest available reasoning effort with 32 attempts per instance, additional attempts improve every model, but pass@32 still ranges from 38.0% to 95.0%. Our analysis of verification programs reveals that generated animations often fail to match reference poses, follow requested motion trajectories, or coordinate related events. We release our benchmark at https://huggingface.co/datasets/figma-ai/motion-graphics-bench.

0.36

JoVox: Joint Hierarchical Flow Matching for Lightweight Voxel Generation

cs.CV cs.GT ·2026-11-07 ·Xiangjun Tang, Jianqi Chen, Peter Wonka
Abstract

Generating high-fidelity voxels while keeping their lightweight nature remains an open problem in computer vision. Current generators are either high-fidelity but computationally heavy, or lightweight but of lower fidelity. Hierarchical methods keep the lightweight nature of voxels. However, their coarse levels, which decide where all finer cells can appear, are generated with the least information about the shape and are never revisited, so they are often plausible but wrong. Our key insight is to let finer cells take part in deciding coarse cells. To this end, we propose JoVox, a discrete flow matching framework that learns the joint distribution of all levels of the voxel hierarchy, instead of a separate distribution for each level, so that coarse cells can be predicted with finer cells already generated nearby. Within this framework, we further introduce a two-level prediction, which outputs the cells two levels finer at once instead of one level finer, halving the number of levels that the network processes and yielding voxels of higher fidelity. We also introduce an unmasking strategy that revisits early decisions once finer cells are generated and corrects inconsistencies in masked diffusion models. Our framework outperforms the baselines on both unconditional and image-conditioned voxel generation. Compared with the current state-of-the-art TRELLIS.2, we achieve 8x higher training throughput, 1/15 of the FLOPs in a forward pass, train on 4 vs. 32 GPUs, and obtain higher quality with an 11% lower Chamfer distance.

0.36

Fine-resolution land-use segmentation and synthetic image-label augmentation in the Kathmandu Valley

cs.CV cs.AI ·2026-11-07 ·Pranil Parajuli, Sugam Dahal, Kiran Chandra Dahal
benchmarkhot direction
Abstract · 21 pages, 26 figures

Parcel-scale land-use data is missing for most rapidly urbanizing Global South cities, as current segmentation architectures rely on multispectral imagery and planned layouts. We introduce a hand-annotated six-class land-use corpus covering 202 km$^2$ of Kathmandu Valley across 408 fine-resolution RGB scenes. The dataset is uniquely challenging: RGB spectral similarity is high, class imbalance approaches 100:1, and key structural features like rivers measure just two to three pixels wide at a standard hierarchical decoder's prediction stride. Evaluating five architectures reveals that standard random data splitting conceals massive spatial leakage. Switching to a strict sequence-block split drops performance by up to 0.095 mIoU. This penalty exceeds the performance gap between the architectures themselves. We modified the SegFormer baseline by integrating a half-resolution residual detail path, boundary-reliability reweighting, a topology-aware loss, and a presence-balanced sampler. Adding only 1.9% more parameters, the enhanced architecture improves mean IoU across every test fold, with gains concentrated in rare, thin-structure classes. However, both models achieve identical results on the external LoveDA benchmark. Finally, augmenting with diffusion-generated pairs at matched compute degraded accuracy across all architectures, especially harming rare classes it aimed to improve. This demonstrates that blindly incorporating synthetic data can severely damage performance, highlighting a critical need to rigorously evaluate and filter generated samples before training.

0.36

ClearForm: Shared-Surface Gaussian Splatting for Transparent and Opaque Scenes

cs.CV ·2026-11-07 ·Hajeong Lee, Jong-Il Park
code/data
Abstract · 15 pages, 9 figures

Transparent-interface reconstruction must recover glass surfaces even when most observed color comes from the scene behind them. We present ClearForm, a shared-planar-Gaussian method for jointly reconstructing transparent interfaces and the appearance seen through them. We identify and analyze the Geometry-Appearance Compositing Dilemma (GACD): when the same transparency-dependent alpha controls both color and geometry, revealing the background can weaken a transparent surface's contribution to geometric reconstruction. We derive its effect on plane-based depth estimation, identify a mechanism that can weaken geometric learning, and quantify the loss of interface support through controlled comparisons. To address this conflict, we develop geometry-preserving dual-opacity surface reconstruction. Shared planar Gaussians retain interface contributions in geometry compositing while modeling transmission and sparse reflection in RGB, supported by transparency-aware geometric constraints. To manage the different primitive-visibility requirements of these compositors, we introduce Dual-Evidence Lifecycle (DEL), which retains Gaussians supported by multi-view geometry or RGB reconstruction evidence. On 3D-FRONT-T, ClearForm achieves 27.05 mm Chamfer distance (CD) and 0.928 F1, compared with 59.18 mm CD for our PGSR reproduction and the reported 34 mm CD and 0.836 F1 for GLINT. On TransLab, it achieves 1.65 mm CD, 0.963 F1, and 40.14 dB PSNR. In matched ablations, dual-opacity compositing reduces CD from 59.25 to 27.09 mm, while DEL improves PSNR by 2.06 dB. Our project page is available at https://hajeongx.github.io/clearform/.

0.35

When Greedy Is Sorting: Exact and Certified Ideal Rankings for Novelty Metrics

cs.IR ·2026-11-07 ·Kehan Liu, Zhaoyun Chen, Xianquan Yan
Abstract

The ideal ranking that normalizes novelty metrics is NP-hard to compute in general, so widely used $α$-nDCG evaluators approximate its denominator with a greedy rule that rescans every remaining document at each rank. We identify a regime in which this ideal ranking is exact and cheap for both $α$-DCG and intent-aware expected reciprocal rank (ERR-IA). When each document is positively judged for at most one intent, each metric reduces to the classical allocation of a fixed budget across chains of decreasing marginal gains. A metric-specific virtual-score transformation followed by a multiway merge, which we call ChainMerge, returns the globally optimal top-$K$ ranking in $O(G+K\log G)$ time from $G$ preordered intent groups, or in $O(n+K\log G)$ time from $n$ globally preordered candidates, instead of the $Θ(nK)$ of the greedy rule. When intent judgments overlap, exact $α$-DCG optimization is NP-hard under the standard logarithmic discount and $α=0.5$, even when every document has three intents and every intent occurs in two documents. For arbitrary overlap, a polynomial-size LP over nested ranking prefixes upper-bounds the ideal score and certifies the optimality gap of any feasible ranking. In the official TREC Web 2009-2012 judgments, 64.8% of positive documents serve a single intent, yet only 20 of 198 complete queries do so throughout. Beyond this exact regime, the LP certifies 189 of the 198 greedy denominators as globally optimal, and the resulting score intervals preserve the winner in all eight matched comparison groups. On 65,238 MIND-small candidates, ChainMerge returns identical optimal objectives 17-1,011 times faster than the greedy rule. For $α$-DCG, these results trace the hardness of the ideal ranking to intent overlap.

0.35

FEROM: Frontier Endogenous Reveal-Order Marginal Policy Optimization for Masked Diffusion LMs

cs.CL cs.AI ·2026-11-07 ·Zian Su, Ziyang Huang, Kaiyuan Zhang, Xiangyu Zhang
published
Abstract · NeurIPS 2026

Masked diffusion language models generate text by iteratively unmasking positions, with practical samplers often selecting reveal positions from the model's own logits. The reveal order is therefore an endogenous latent variable of the rollout process with an enormous discrete latent space. Policy optimization objectives relying on pre-defined schedules or sampled trajectories introduce either bias or high variance. In this paper, we propose FEROM, Frontier Endogenous Reveal-Order Marginal Policy Optimization, which approximates policy gradients for the rollout-induced marginal response policy of masked diffusion LMs. FEROM derives a Rao-Blackwellized policy-gradient identity over latent reveal orders and expresses the resulting estimator as posterior edge occupancy on a reveal-state DAG. To make marginalization practical, we introduce Frontier Reveal Marginalization, a budgeted estimator that approximates this marginal edge structure by combining CDF-based computation of local probabilities of the endogenous reveal process with frontier expansion of high-mass partial states. Integrated into a GRPO-style objective, FEROM replaces single-path log-scores with a detached token-score surrogate over posterior-relevant reveal edges. Experiments on math and coding tasks show comparable or improved results over existing methods under matched scoring-state budgets. An offline proxy study further shows improved marginal-gradient approximation with increased scoring budgets.

0.35

Patchbay: a block model for the sonification of ADS-B aircraft data

cs.SD cs.HC ·2026-11-07 ·Arquimedes Canedo
publishedunusual
Abstract

Patchbay is a design for turning the data that an ADS-B receiver collects into sound by wiring typed blocks together. ADS-B reports are the position and altitude messages that aircraft broadcast and that a small ground receiver can collect. We first describe what such a receiver reports: signals that differ in shape (continuous, stepped, circular, categorical), in resolution, in who produces them, and in how long they stay put, which runs from about ten seconds to the whole time an aircraft is in view. From this we derive a small model. Four kinds of connection carry an aircraft set, a value, an event or audio, and 27 blocks select aircraft, read fields, decide when a note starts, shape values and make sound. One of these blocks makes the receiver act as a radar: scanning by bearing it is a line that rotates about the receiver, and scanning by range it is a ring that grows from the centre or shrinks inward. Each aircraft sounds at the moment the scan reaches it, and the scan's period is set in bars of a beat clock. The model does not depend on a browser. Our reference implementation is a browser editor, contained in one HTML file, that replays two recorded captures and records both an audio take and a MIDI take, a file of the notes played rather than of the sound. We say what is not built and place the work beside the closest systems we found, among them a radar-sweep sequencer for tangible objects from 2006 and a range-ring sonification of ADS-B aircraft published in 2026. Sweeps that trigger sound, and typed connections, both exist in earlier systems. To our knowledge, none of the sources we reached carries the aircraft in view, with every field attached, as one kind of connection in an editor.

0.34

SHIN: Scalable Hierarchy for INcremental 3D Gaussian Splatting

cs.CV ·2026-11-07 ·Reagan Koo, Yeong-Gyu Kim, Eun-Seok Ryu
Abstract · 8 pages of main text (13 pages including references and supplementary material), 7 figures

3D Gaussian Splatting (3DGS) provides high-quality real-time novel-view synthesis, but deployment budgets vary across devices and network conditions. Most compact methods target one operating point, so another budget requires a separate representation. SHIN is a scalable 3DGS hierarchy with a reusable Base and additive enhancement layers, so each higher-quality operating point extends rather than replaces the decoded representation. Constructing this hierarchy requires deciding which Gaussians each stage adds and how much spherical-harmonic (SH) appearance each retains, because these choices remain fixed in later prefixes. SHIN derives one shared Gaussian order from multi-scale position-scale sensitivity and divides each stage budget between adding Gaussians and retaining SH bands using calibrated gradient-based scores per byte. Lightweight opacity updates adjust compositing while transmitted geometry and SH coefficients remain unchanged. Across three benchmarks, SHIN progressively improves one cumulative representation and achieves the best PSNR-size efficiency in every overlapping comparison with the multi-rate and progressive baselines.

0.34

Evaluation protocols for historical script classification: the cost of manuscript-level leakage and the role of soft labels

cs.CV ·2026-11-07 ·Daniil I. Naryzhnyi, Roman A. Popkov, Daniil A. Bakalin
Abstract · 29 pages, 8 tables, 2 figures

A paleographer applies a script type classifier to unseen manuscripts. We measure how much testing on other pages of training manuscripts inflates accuracy. On VML-HP-ext (715 pages, 171 Hebrew manuscripts, 14 script classes) we compare page-level and manuscript-level splits. Page-level accuracy drops from 97% to 54% (59% at manuscript level), and is 39% on the official blind test set. The gap of 40 to 47 percentage points (pp) at patch, page and manuscript level is statistically significant and reproduces with ResNet18; on binarized patches (ink and background only) it is 34 to 42 pp, also significant. On unseen manuscripts the model is overconfident: mean confidence 70% against 54% correct pages; patch-level expected calibration error (ECE, mean gap between confidence and accuracy) rises from 0.065 to 0.444. Binarization leaves patch-level overconfidence but cuts it markedly at page level. Confidence still helps triage: in the more confident half of pages 67% are correct, 41% in the rest. Training on expert soft labels showed no evidence of an accuracy difference, though a small page-level loss cannot be ruled out. The predicted distributions are closer to the expert's assessment. ECE is lower at patch level (0.21 to 0.31) and in most runs at page level, but not significantly at manuscript level. Label smoothing (blurring the label, no expert input) also lowers ECE but comes no closer to the expert. Temperature scaling (post-training confidence correction) helps significantly only at patch level. On 97 pages of 16 manuscripts with mixed expert assessment, soft labels cut divergence from the expert by 22 to 30% (significant in four of five runs); the point estimate of accuracy is 16.5 to 23.7 pp below hard labels, but with 16 manuscripts the loss can neither be confirmed nor ruled out. We propose recommendations for the evaluation protocol; code will be released with the paper

0.34

AFVE-Bench: Measuring Abandoned-Frame Verbalization Errors in Multi-Turn LLM Revision

cs.CL cs.AI ·2026-11-07 ·Jiaming Chu
Abstract · 18 pages, 4 figures

Multi-turn revision often asks a model to replace an earlier organizing frame while preserving still-valid facts and artifact requirements. Existing instruction-following evaluations mainly assess the latest target and may therefore score an artifact as successful even when it adopts the new frame but exposes the superseded one. We call this failure an abandoned-frame verbalization error (AFVE). AFVE-Bench evaluates silent frame replacement in English and Chinese using 192 matched four-turn dialogues across eight writing domains. Each seed has zero, one, or two pre-target user requests to generate under the source frame. AFVE eligibility depends on the user-request history and applies to the one- and two-request conditions; the zero-request condition serves as a frame-bearing baseline. Eleven model configurations yielded 4,224 expected post-target response units. Two annotators independently scored 4,204 scorable responses for target adoption, source-frame verbalization, and direct usability; model and condition metadata were hidden, and disagreements were adjudicated. AFVE occurs in 25.54% (715/2,799) of history-conditioned responses. Of 2,388 target-passing history-conditioned responses, 472 (19.77%) still expose the abandoned source frame, most often through contrastive framing (423/472). For these standalone revision tasks, target adoption does not ensure source-frame silence.

0.33

Mental Imagery That Matters: Curing Latent Collapse in Multimodal Reasoning

cs.CV cs.AI ·2026-11-07 ·Mi Luo, Rohan Dugad, Zihui Xue, Alex Dimakis, Kristen Grauman
top venuecode/data
Abstract · Accepted by NeurIPS 2026, Project page: https://vision.cs.utexas.edu/projects/AnchorSim/

Mental simulation, the ability to reason by constructing and updating internal visual states, is central to flexible human cognition, but remains elusive in current multimodal large language models. Latent multimodal reasoning offers a promising route toward this capability by interleaving text with continuous visual latent blocks, allowing models to maintain internal visual workspaces without decoding them into pixels. In this paper, we ask whether these latent visual states actually matter. Through a direct perturbation analysis across representative latent-reasoning MLLMs, we find that they often do not: dropping, replacing, or shuffling generated visual latents has only minor effects on the model's own final-answer probability, while equivalent perturbations to the linguistic reasoning trace are substantially more damaging. We call this failure mode Mental Simulation Collapse (MSC): the model appears to perform interleaved visual reasoning, but routes most answer-relevant computation through text. We further diagnose that MSC follows naturally from existing training objectives: visual alignment can make generated latents resemble helper-image embeddings, but it does not require the decoder to rely on them; subsequent text-only supervision further leaves the semantic role of the latent pathway unidentifiable. To address this, we introduce AnchorSim (Anchored Mental Simulation), a contrastive distillation framework that trains models to make their generated visual latents consequential. Across multiple visual reasoning benchmarks, AnchorSim improves performance, while diagnostic re-evaluation shows that visual perturbations now meaningfully affect predictions. Our results expose a fundamental failure mode of latent multimodal reasoning and demonstrate that mental simulation must be anchored to model behavior in order to matter.

0.33

onóna: Twin Embeddings with Tied Unembeddings

cs.CL cs.LG ·2026-11-07 ·Cheng Luo, Xinnan Dai, Zhongzhu Zhou, Zefan Cai, Ruijie Zhu +2 more
Abstract · 18 pages

Weight tying reduces the parameter cost of language models, but the usual tied output head is limited to a single softmax. We introduce onóna, a two-table parameterization that increases output expressiveness while preserving input-output weight sharing. The Transformer takes the average of two embedding tables as input, and the output head combines the softmax distributions defined by those same tables. A gate conditioned only on the current input token determines the mixture weight. Its values can be precomputed for inference, and the Transformer body remains unchanged. We evaluate onóna across Dense, Gated DeltaNet hybrid, mixture-of-experts, and Attention Residual architectures at 142M, 330M, and 1B scales. Under matched token budgets, onóna lowers final validation loss and external perplexity in 11 of 12 comparisons and raises the average accuracy over eight downstream tasks in 10 of 12. The one exception on loss is the 1B MoE model. The cost is one extra embedding table and one extra vocabulary projection.

0.33

A Tool for Anonymized Clinical Knowledge Retrieval and Speaker Interaction Analysis of Healthcare Tele-mentoring Sessions

cs.IR ·2026-11-07 ·Parthaw Goswami, Erfan Zareei, Amy Vogelsmeier, Dima Dandachi, Ben Casterline +2 more
Abstract

We present a tool for anonymized retrieval of clinical knowledge and analysis of multi-speaker interactions in tele-mentoring sessions of Project ECHO (Extension for Community Healthcare Outcomes). Our tool's novel design includes: (a) a compound metadata filter that enforces strict speaker- and session-level isolation during retrieval, preventing cross-session context contamination; (b) MedGemma-4B as a locally-hosted language model for clinically grounded answer generation across three query modes; (c) interaction graphs to compute betweenness centrality, PageRank influence, and turn-taking patterns to quantify participation equity; and (d) a consistent alias-based anonymization scheme that prevents real participant names from appearing in any rendered output. During the demo, a healthcare provider (user) will interactively query multi-year ECHO sessions, explore participation dynamics through interaction graphs, compare sessions side-by-side, and assess answer quality via six reference-free RAG metrics, all without any cloud API dependency.

0.33

MaRiO: Multi-agent Collaborative Reasoning via Shared Observations in MLLMs

cs.CV ·2026-11-07 ·Nathaniel Redmond, Fidel Omar Tito Cruz, Devansh Sharma, Shehreen Azad, Sirshapan Mitra +2 more
publishedbenchmarkcode/data
Abstract · Accepted to the NeurIPS 2026 Evaluations & Datasets Track (poster). First two authors contributed equally

Multimodal Large Language Models (MLLMs) show strong visual perception capabilities, but existing evaluations largely focus on single-view or multi-view settings with shared camera parameters, leaving the challenge of multi-agent collaborative scenarios underexplored. In such settings, multiple agents observe a shared environment from independent viewpoints, and reliable decisions require reasoning across these independent observations of the same environment. To study this problem, we introduce MaRiO, a benchmark for evaluating multi-agent collaborative reasoning, and evaluate 23 state-of-the-art MLLMs on it. Our analysis reveals three key findings: (1) tasks governed by static perception are generally easier than tasks that require transformational event sequences; (2) most models struggle to resolve cross-agent correspondences, with errors increasing with scene complexity; and (3) models show limited ability to leverage geometric cues present in the scene. As an initial step toward addressing these challenges, we explore geometry-guided synthetic view augmentation that generates intermediate transition frames between agent observations, providing auxiliary spatial context that improves performance in certain categories. Overall, our results highlight that multi-agent collaborative reasoning remains an open challenge and an important direction for multimodal embodied systems. The code is available on the project page: https://omarftt.github.io/MaRiO-page/.

0.33

S2HAND: Accuracy-Preservation Trade-Offs for Hand Specialization in Whole-Body Pose Models

cs.CV ·2026-11-07 ·Singyu Muk, Xuanren Song, Kalin Stefanov
code/data
Abstract · 39 pages, including appendices. Code and models: https://github.com/ronggong000/S2HAND

Unified whole-body pose estimators jointly predict body, face, and hand landmarks, but often degrade on small, blurred, or occluded hands. Attaching a standalone hand specialist introduces extra detection overhead, coordinate alignment, and wrist-stitching artifacts. We introduce S2HAND, a hand-specialized Sapiens2-5B framework that preserves the native 308-keypoint output interface. Specifically, S2HAND-MT adapts a lightweight 0.87M-parameter pose branch to route 40 finger heatmaps while strictly preserving the shared Transformer, auxiliary task heads, and the other 268 body, foot, and face heatmaps. S2HAND-Pose further fine-tunes three Fisher-selected Transformer blocks (267.13M parameters), trading exact preservation for stronger hand recovery: it lifts macro AUC from 57.83% to 65.73% across HInt TEST, COCO WholeBody VAL, and FreiHAND EVAL, and surpasses HInt-adapted Hand5/HRNet and stitched Hand5 systems on all five Native Pipeline benchmarks. On COCO, S2HAND-Pose improves hand AP by 12.89 points and body-plus-foot AP by 0.19 points. Across a 512-image multi-task audit, it maintains 99.60% foreground segmentation agreement with the public checkpoint, with quantified normal and pointmap drift. Its 2D coordinates also lower 3D mesh error in frozen Pose2Mesh transfer, while our optional TailGuard re-cropping module cuts S2HAND-MT's severe failure rate on NewDays from 13.45% to 10.43%. S2HAND thus resolves the whole-body hand bottleneck with either exact non-hand preservation (MT) or empirically audited trade-offs (Pose).

0.33

From User Behaviors to Satisfaction Judgments: Reasoning about User Satisfaction with AI Overviews

cs.IR ·2026-11-07 ·Xingzhu Wang, Yitong Xu, Yue Li, Daiting Shi, Jingzhou He +1 more
Abstract

The growing presence of AI overviews in everyday search makes their effective evaluation essential. Response quality has been the primary focus of existing evaluation research, with only a small number of studies examining the user perspective. These evaluations rely on queries and response content, leaving the information conveyed by user behavior insufficiently utilized. This work investigates satisfaction prediction for AI overviews from a user-centered perspective and incorporates behavioral signals into the prediction process. Applying existing behavior-based satisfaction prediction methods from traditional information retrieval to this setting, however, presents limitations. We address these limitations with STaRS, an LLM-based approach designed specifically for predicting satisfaction with AI overviews. At its core is a structured reasoning procedure: individual behavioral signals are interpreted first, and their implications then inform the prediction of user satisfaction. Reinforcement learning further optimizes this procedure, while a reward shaping mechanism alleviates reward sparsity during training. Evaluation on two real-world datasets demonstrates that STaRS outperforms the comparison methods in satisfaction prediction, providing search providers with a practical means of accurately estimating users' satisfaction with AI overviews.

0.33

Coding or Cheating? An Investigation of Frontier Coding LLMs

cs.CL cs.AI ·2026-11-07 ·Yuntao Gui
code/data
Abstract

Frontier agentic LLMs are trained on verifiable rewards in reinforcement learning (RL) environments built from real software repositories. A passing test suite is treated as the reward for solving the task. We show that, on hard problems, frontier models often satisfy these rewards by obtaining the answer instead of deriving it, known as reward hacking. Prior reports document this behavior on public benchmarks. We test its persistence on unseen problems. In a hardened environment where the answer cannot leak, nearly every model we tested attempted to recover the answer, escalating through tiers of increasingly specialized exploitation: mining the repository history for the upstream fix, inspecting the git object store directly, and sweeping the filesystem for a possible newer copy of the source or hidden tests. These attempts were concentrated on the hardest problems, and the models differed in whether, how often, and how deeply they escalated. Our position is that this behavior is a learned shortcut, acquired in RL environments that leak the answer, not a sign of intelligence. We release agent traces for these models to support further study at https://huggingface.co/datasets/logitly-ai/reward-hacking-raw-v1.

0.32

Machine Hearing: Can Audio LLMs Generalize Beyond Human Auditory Perception?

cs.SD cs.CL ·2026-11-07 ·Liwei Liu, Jun Tao, Xingyuan Li, Xuenan Xu, Chen Zhang +6 more
benchmarkcode/data
Abstract · 22 pages, 7 figures, 5 tables

Multimodal LLMs have made remarkable progress in perceiving the physical world, while their perceptual scope remains largely human-centric, focusing on what humans can directly see or hear. This progress raises the prospect of perception beyond human limits, such as extending hearing to sounds that are inaudible to humans. More broadly, the physical world is full of waves, vibrations, and oscillations, only a small fraction of which are directly audible to humans. This invites the extension of multimodal LLMs toward a broader form of machine hearing, which remains largely underexplored. To investigate this direction, we introduce the Machine Hearing Benchmark (MHB), comprising 6,128 instances across 13 tasks and seven domains. MHB spans highly heterogeneous input regimes, with sampling rates from approximately $10^{-5}$ to $10^6$ Hz, sequence lengths from $10^1$ to $10^6$ samples, and both single- and multi-channel inputs. Evaluations of current open-source and proprietary Audio LLMs (ALLMs) reveal limited zero-shot transfer to machine hearing. Supervised adaptation on machine-hearing data alone does not close this gap, suggesting that the challenge extends beyond data exposure to a systematic mismatch between existing audio front ends and machine-hearing inputs. To address this mismatch, we introduce MachEar, a lightweight encoder extension that complements the pretrained Audio LLM for machine-hearing inputs. The results show that MachEar equips a pretrained Audio LLM with stronger machine-hearing capabilities while largely preserving its existing audio understanding performance.

0.32

Scaling Optimization-Oriented Hypernetworks for Implicit Neural Representations

cs.CV ·2026-11-07 ·Lulu Cai, Hanxiang Ren, Siyan Dong, Li Sun, Jiefeng Wu +3 more
code/data
Abstract · 33 pages, including appendices. Code: https://github.com/lulucai435/hyper-inrs

We present a hypernetwork that generates the parameters of an implicit neural representation (INR) for a given signal in a single forward pass, without test-time optimization. Existing hypernetworks for this problem either ignore the chain-rule structure of the test-time optimization that produces an INR or, when they preserve it, instantiate the structure with MLPs whose parameter count grows quartically in target-network width. Our central contribution is to show that this parameter-count cost is structural to the MLP instantiation, not to the optimization-oriented principle of preserving chain-rule structure, and to resolve it by replacing the MLP body with attention. The resulting hypernetwork combines three design choices: an output-neuron-centric tokenization that aligns the token axis with each layer's output-neuron axis; an information-flow consistency rule that fixes the role of every cross-attention module from the representation its output should occupy; and a token-space iterative rollout that realizes the five quantities one optimization step on the target INR computes as five cross-attention modules with parameter count independent of target-network width. Across five 2D and 3D INR generation benchmarks, our method outperforms the state-of-the-art single-pass and optimization-oriented baselines; scales to target-network widths at which the MLP-based optimization-oriented paradigm exhausts GPU memory; and produces the highest result on NeRF generation without ground-truth INR weights supervision.

0.32

Hibana: Fast Agentic Inference on Unified-Memory Devices

cs.CL cs.AI ·2026-11-07 ·Yahya Emara, Chi-Chih Chang, Anthony Fei, Ahmed F. AbouElhamayed, Chong-Yan Chen +3 more
code/data
Abstract

Local AI promises capable personal assistants that run on users' own devices, keeping data and computation under their control. Open-weight models that fit on local devices can now handle practical tasks such as coding. Yet current inference engines fall short. Popular engines such as vLLM and SGLang target datacenter GPUs with dedicated host memory, while local engines underuse the memory hierarchy of unified-memory devices. We present Hibana, an inference engine for unified-memory devices that co-optimizes model representation, memory management, and decoding strategy. Model surgery, the process of editing a trained model's architecture to suit its hardware, changes how the model computes, shrinking its memory footprint and data movement. Unified-memory management preserves each agent's state across turns, so long-running agents can resume instantly instead of recomputing their state. Adaptive speculation generates several tokens per step. Hibana codesigns these features around the shared memory capacity and limited bandwidth of unified-memory systems-on-chip to support local multi-agent workloads. On a single DGX Spark serving Qwen and Nemotron models of up to 125B parameters, Hibana completes up to 3.4$\times$ more agentic tasks per hour and delivers up to 3.9$\times$ higher decoding throughput than other established frameworks such as vLLM, NVIDIA Flare, llama.cpp, and TensorFold. Code is available at https://github.com/sharar-project/hibana.

0.32

Semantic Synthesis over Brute-Force Search: LLM-Guided Transformation Design for Efficient Learning

cs.CV cs.AI ·2026-11-07 ·Usha Shrestha, Radu Timofte, Dmitry Ignatov
Abstract

Data transformation pipelines shape the training distribution in computer vision, but designing them is a computationally expensive problem. We investigate whether large language models can enable efficient discovery by synthesizing executable transformation programs outside predefined augmentation spaces. We compare nine strategies under a unified training protocol, including standard augmentation policies, brute-force search, direct code generation with two LLMs, and performance-guided generation using LLMs fine-tuned on accuracy-ordered transformation pairs. When the best candidates are retrained for 20 epochs, LLM-generated transformation programs lead on every dataset, reaching 0.8877 accuracy on CIFAR-10, 0.5825 on CIFAR-100, and 0.9588 on SVHN, surpassing brute-force search over 6,000 candidates. The best transform on CIFAR-10 and SVHN comes from a fine-tuned generator using 270 candidates, while the best on CIFAR-100 comes from an off-the-shelf generator using only 10. This needs 22x fewer candidate evaluations in the generation step and 600x fewer for off-the-shelf generations, relative to brute-force search. Beyond transformation discovery, we analyze the reliability of the reduced-budget evaluation commonly used for augmentation search. Correlating one-epoch selection scores with 20-epoch outcomes reveals strong, weak, and absent predictive relationships depending on the dataset and architecture, with correlations ranging from 0.756 to near zero. A positive control indicates that range restriction alone is unlikely to explain the low correlations. Our results indicate that semantic synthesis with code LLMs provides an efficient alternative to search-based transformation discovery and highlight the importance of validating evaluation fidelity when optimizing data transformations.

0.32

Evaluating Physically Grounded Procedural Reasoning in VLMs through Real-World Execution: Crochet as a Testbed

cs.CL cs.CV ·2026-11-07 ·Maximiliano Wardle
Abstract

Vision-Language Models (VLMs) can generate procedures that look convincing on paper, but do they work in practice? In this paper, we propose crochet as an ideal testbed for physically grounded procedural reasoning in VLMs. Crochet cannot be reliably simulated or automated: the only way to know what a pattern produces is to make it by hand, yet verification is tangible and cheap, requiring nothing but a hook and a ball of yarn. We introduce FoundationChain, a deceptively simple test of procedural reasoning: given an image of a basic crochet object, can VLMs infer how it was made and generate a pattern that produces the same result? We crochet 40 patterns generated by four VLMs by hand. Even at this scale, physical execution not only reveals failures that automated proxies cannot detect, but also enables a more precise diagnosis of their causes and consequences: while models write fluent, convincing patterns, only four are free of errors, and our analysis exposes deficiencies in visual, geometric and procedural reasoning.

0.31

Split and Bridge: Multimodal Generation via Diffusion Bridging

cs.CV cs.AIcs.LG ·2026-11-07 ·Ahmad Arrabi, Xiaohan Zhang, Xingyu Li, Safwan Wshah
code/data
Abstract

Visual data are often expressed through multiple complementary modalities (e.g., images and segmentation masks), each capturing different aspects of the same underlying structure. Most existing multimodal methods adopt a top-down paradigm, learning a shared latent variable through joint training from scratch, which limits scalability. We propose Split and Bridge (SnB), a bottom-up framework for multimodal generation that couples pretrained unimodal diffusion models at sampling time. Rather than learning a unified latent space, SnB generates coherent multimodal samples by following a shared diffusion trajectory up to a split point and then branching into modality-specific trajectories guided by a bridge module. This design enables plug-and-play reuse of powerful pretrained models without joint training, making it scalable and easily adaptable. Across standard benchmarks (PolyMNIST, CelebAMask-HQ) and more complex settings (PIE-Bench++, ROCO), SnB achieves a strong balance between sample quality and coherence. We show that SnB naturally extends to downstream zero-shot tasks such as interleaved co-generation, image editing, and style coupling/decoupling. Our results demonstrate that bottom-up generation provides a practical framework that complements existing top-down multimodal approaches. Code is available at https://github.com/AhmadArrabi/SnB

0.31

BAPLe++: Cross-Model Transferability and Prompt-Level Defense of Backdoors in Medical Foundation Models

cs.CV cs.AIcs.LG ·2026-11-07 ·Ufaq Khan, Umair Nawaz, Satyajit Tourani, Awais Rauf, Fahad Shamshad +4 more
code/data
Abstract · 13 pages, 4 figures, 3 tables

Medical foundation models (Med-FMs) are increasingly adapted to clinical tasks through prompt learning, which tunes a small set of prompt tokens while the backbone remains frozen. Recent work has shown that this efficiency introduces a vulnerability: jointly optimizing learnable prompts and an imperceptible noise trigger during few-shot adaptation implants a backdoor that leaves clean performance intact but steers triggered inputs to an attacker-chosen class. Since prompts are compact and readily shared, this raises two questions: whether a backdoored prompt remains effective on a different Med-FM, and whether it can be repaired without retraining the backbone. We address both in BAPLe++. First, we present the first study of cross-model transferability of prompt-based backdoors, quantified by two proposed metrics, Clean Accuracy Difference (CAD) and Backdoor Accuracy Difference (BAD). Second, we propose DePLe (Defense via Prompt Learning), which uses the infected prompt as a fixed anchor and learns a new prompt on a small clean set that preserves the clean task while diverging from the compromised embedding region. Experiments on four Med-FMs and six chest X-ray and histopathology datasets show that backdoors transfer across models, but asymmetrically and in a dataset-dependent manner. Compared with four training-free input transformations, DePLe reduces backdoor accuracy more consistently while preserving, and in several settings improving, clean accuracy. Our code can be accessed on https://github.com/Ufaqkhan/Baple-Plus-Plus

0.31

Cascading Circle Loss: Order-Preserving Contrastive Learning for Graded Legal Retrieval

cs.CL ·2026-11-07 ·Rayane Kimouche, Ahmed Touila
Abstract

This paper introduces Cascading Circle Loss (CasCL), a contrastive learning framework for legal retrieval tasks with graded relevance. CasCL combines an InfoNCE objective for positive-negative separation with a cascade of ranking boundaries that model ordered relevance levels between queries and legal clauses. The framework also includes a negative-aware self-paced optimization mechanism and a relevance gain weighting strategy to improve ranking quality. Experiments on legal passage retrieval benchmarks show that CasCL consistently outperforms existing methods, including InfoNCE, Circle Loss, MNRL and Margin MSE, across both binary retrieval and graded ranking metrics.

0.31

WorldFoundry: Towards Unified Infrastructure for World Intelligence

cs.CV cs.AI ·2026-11-07 ·Juanxi Tian, Jiadong Hou, Baiang Li, Xin Lin, Zhengbo Jiao +11 more
benchmarkcode/datacommunity pickpopular code
Abstract · 39 pages, 10 figures. This technical report will be updated periodically

Visual generative models increasingly blur the line between synthesizing observations and simulating environments. Yet visual plausibility alone does not constitute successful world modeling: compelling outputs may lack persistent state, physical dynamics, or causally consistent responses to interactions. We define a generative world model as a predictive dynamical system that maintains a persistent, history-conditioned representation of an environment, updates that representation in response to actions or interventions, and generates temporally coherent trajectories that reflect outcomes under observed and counterfactual interventions. From this definition, we derive an intelligence-oriented framework spanning perceptual grounding, persistent state modeling, controllable transformation, dynamics-aware prediction, and closed-loop interaction. We present WorldFoundry, open-source infrastructure for reproducible inference and evaluation across video, 3D and 4D, embodied action, and interactive world systems. WorldFoundry unifies model and benchmark discovery, runtime and asset preparation, execution orchestration, artifact management, and interactive visualization, while modality-specific adapters preserve native inputs, execution semantics, output representations, and evaluation protocols. Explicit manifests, durable artifact contracts, and evidence-aware scorecards separate execution readiness from evaluative claims, making each claim traceable to its supporting artifacts. The catalog contains 270 model entries, 74 benchmarks, and 62 registered metric entries, with readiness and execution evidence recorded separately. WorldFoundry enables reproducible comparison of persistent, controllable dynamics across heterogeneous systems, advancing the field from perceptual synthesis toward measurable world intelligence.

0.31

LIFT: Local Metric Alignment and Foundation-guided Propagation for Few-shot Depth Completion

cs.CV ·2026-11-07 ·Xiang Li, Xiaojin Gong
code/data
Abstract · 8 pages, 5 figures

Few-shot depth completion aims to recover dense metric depth from sparse measurements using only a small number of target-domain annotations, offering a favorable balance between annotation cost and model adaptation. We propose LIFT, a framework that harnesses both depth and feature priors from a frozen pretrained monocular depth estimation (MDE) foundation model for few-shot depth completion. Specifically, LIFT first estimates local scale and shift parameters on a coarse grid and interpolates them into smooth, dense parameter fields, providing a spatially adaptive metric depth initialization. It then progressively refines this initialization through multi-scale depth propagation guided by foundation features. Finally, a lightweight residual decoder combines multi-level foundation features with propagation correction to refine the metric depth. Extensive experiments on NYUv2 and KITTI-DC demonstrate that our proposed method achieves state-of-the-art few-shot performance. The source code will soon be released.

0.31

RoadState: A Domain-Prior-Enhanced Vision-Language System for Industrial-Grade Lane-Level Map Updating

cs.CV cs.AI ·2026-11-07 ·Deguo Xia, Zihan Li, Jiaqi Wang, Dong Xie, Xiyan Liu +9 more
published
Abstract · 12 pages, 5 figures, including appendices. Accepted to AAAI-IAAI 2027, poster presentation

Lane-level map updating requires temporally consistent, physically valid, and auditable road-scene semantics despite occlusions, ambiguous lane markings, temporary obstacles, and viewpoint changes. We present RoadState, a production-deployed vision-language system that transforms sequential road-scene observations into structured lane-level semantics for Baidu Maps' map-updating pipeline. Its core model, RoadState-VLM, formulates road-scene understanding as autoregressive state estimation, predicting the current structured road state from the previous observation and symbolic state together with the current observation. We introduce Geometric Chain-of-Thought (Geo-CoT), which uses a directed acyclic graph of semantic dependencies as a domain prior to improve semantic consistency and reasoning fidelity across interdependent road-state dimensions. Extensive experiments on a real-world dataset show that RoadState-VLM achieves 94.33% mean accuracy across six road-state dimensions, outperforming the strongest fine-tuned VLM baseline. Ablation studies further demonstrate the complementary effectiveness of Geo-CoT and temporal transition context. A cascaded routing pipeline directs routine cases to lightweight models while reserving complex cases for RoadState-VLM. Since its deployment in March 2025, RoadState has supported lane-level map updating across more than 360 cities and reduced overall production costs by 30%. Together with human-in-the-loop auditing, the system maintains robust performance under stress-test conditions. These results demonstrate how domain-prior-enhanced reasoning, structured state interfaces, selective routing, and human quality control can bridge the gap between VLM-based road-scene understanding and large-scale lane-level map updating.

0.30

How reliable is an embedding of the seabed? Repeat-look measurement of frozen encoders on synthetic aperture sonar

cs.CV cs.LGstat.AP ·2026-11-07 ·Harold Horsley, Oscar Pizarro, Hayat Rajani, Morten Omholt Alver
publishedposition
Abstract · Accepted at the Maritime Computer Vision Workshop (MaCVi) at WACV 2027

Pretrained encoders from self-supervised representation learning are increasingly applied to sonar imagery of the seafloor, where they are evaluated with protocols that take one embedding per target. A target patch of seabed, by contrast, is re-imaged from an alternate look on overlapping survey passes, so each embedding is one measurement of that target. Standard protocols cannot say how consistent those measurements are, or how much of their spread belongs to the patch and how much to the acquisition geometry. We treat this as a measurement-reliability problem. Using a synthetic-aperture sonar (SAS) survey, we measure the repeat-look reliability of embeddings from a frozen natural-image encoder (DINOv3 ViT-S/16). Standard single-look metrics, effective rank, clustering and a linear probe for coral, are all stable to the look, and the probe performs well (AUC 0.93). Yet the repeat-look reliability of the same embeddings is poor, with a distance-based intraclass correlation (dbICC) of 0.20 and nearest neighbours and clusters that barely agree across looks. This contradiction resolves as we find the probe learns a stable direction, whose ICC of 0.60 is 3.0 times that of the full embedding. Variance components, calculated from generalizability theory, attribute under a tenth of the within-cell spread to look acquisition geometry. The rest is cell-specific and cannot be separated without repeat passes at matched geometry. A sonar-pretrained encoder (BenthicDINO) and a classical SAS texture descriptor show the same pattern. For this dataset and the tested representations, uses that depend on patch identity fail, such as cross-pass comparison and retrieval, while uses at the level of habitat type (e.g. coral classification) do not. Because the standard metrics did not detect this, we argue that repeat-look reliability should accompany any evaluation of representations of the seafloor.

0.30

RoutingTrie: Metadata Routing for Constrained Generative Retrieval

cs.IR cs.LG ·2026-11-07 ·Darshan Fofadiya, Yang Liu, Shuying Sun, Kyumin Park, Juhana Kangaspunta
benchmark
Abstract

Advances in large language models and semantic IDs (SIDs) have brought generative retrieval (GR) into industrial applications. These systems must return items that satisfy request metadata constraints as well as relevance requirements. A catalog trie ensures SID validity, but filtering metadata only after generation wastes computation on rejected candidates and may leave too few eligible results. We propose RoutingTrie to incorporate metadata constraints before beam selection. RoutingTrie maintains one shared SID trie and builds a compact index for each metadata field using Dense Bitset or Sparse List. We derive a storage cost formula to guide the choice between these index representations for different field distributions. During decoding, field probes determine whether requested values occur below candidate edges and produce a combined mask that guides beam selection. Experiments on ShopTrie, our configurable synthetic benchmark, demonstrate compact storage and efficient CPU and GPU probing across diverse metadata distributions. In a production replay, RoutingTrie raises Hit@100 from 67.5\% to 96.1\%, a gain of 28.6 percentage points, with only 1.1 ms latency increase. We will release the ShopTrie generator, benchmark scripts, and RoutingTrie CPU and Triton GPU implementations to support reproducibility.

0.30

TimeTraveler: Temporal Strategy Planning with Time Dictionary for Streaming Video Understanding

cs.CV ·2026-11-07 ·Minkuk Kim, Junho Kim, Seong Tae Kim
top venuecode/data
Abstract · Accepted at NeurIPS 2026. Project page: https://geppa.github.io/TimeTraveler

Streaming video understanding (SVU) requires models to answer user queries over continuously evolving video streams. Unlike offline video QA, SVU must handle queries whose evidence may lie in the past, appear in the current scene, or emerge only after future observations. Existing streaming video systems often rely on compact visual memories or fixed evidence-access schemes, which can lose fine-grained details and fail to adapt the answering process to the temporal intent of each query. In this paper, we propose TimeTraveler, a streaming VQA framework that performs temporal strategy planning with a Time Dictionary. Inspired by the need to search across different temporal regions of a stream, TimeTraveler stores each observed moment as a timestamp-indexed structured caption and uses this dictionary as an explicit source of queryable evidence. Given a query, TimeTraveler determines whether to recall past evidence, read the present context, or wait for future observations, and then applies a strategy-specific evidence acquisition process. Comparative experiments on streaming video QA benchmarks show that TimeTraveler improves streaming question answering by preserving fine-grained temporal evidence and selecting the appropriate temporal strategy for each query. The project page is available at https://geppa.github.io/TimeTraveler.

0.30

Behavioral Evaluation of Large Language Models Under Cognitive Biases in HAZOP Studies

cs.CL cs.AI ·2026-11-07 ·Vinicius M. Müller, Vitor A. O. Silva, Sávio S. V. Vianna, Flávio V. Silva
code/data
Abstract · 8 pages, 1 figure, 3 tables. Submitted to IFAC for possible publication. Supplementary material available at https://github.com/ViniMelloeMu…

Hazard and Operability (HAZOP) studies safeguard industrial processes against catastrophic accidents. Large language models (LLMs) increasingly assist HAZOP drafting, yet evaluations rely on lexical overlap. Because LLMs internalize human heuristics, prompts may trigger biased judgments. Whether frontier LLMs withstand cognitive perturbations in safety-critical workflows remains unknown. Here we show across 420 trials that three frontier LLMs resisted 96.8% of cognitive manipulations, rendering overall susceptibility statistically inconclusive. Contrary to assumed broad fragility, models rejected confirmation and availability cues. However, safety-critical risk downgrades emerged under low anchors and gain frames. Thus, automated hazard analysis requires disciplined human-in-the-loop validation.

0.30

The RenAIssance: Caption-Guided Diffusion Models for Historical-Art Visual Media

cs.CV cs.AIcs.LGcs.MM ·2026-11-07 ·Zhanyu Tuo
Abstract · Accepted at ACMMM26 RichMediaGAI. This work is co-funded by the European Union's Horizon Europe research and innovation programme under MSCA…

Generative AI can expand cultural collections into prompt-responsive visual media, but historical artworks rarely provide the paired descriptions needed to adapt modern image generators. We present The RenAIssance, a scalable pipeline that turns a collection of historical European artworks into a reusable text-to-image media resource. The pipeline generates scene-level captions with BLIP and uses the resulting approximately 15,000 image--text pairs from DEArt to train a compact LoRA adapter for Stable Diffusion v1.5. The adapter shifts generated images toward the historical-art distribution while retaining the base model's ability to render novel concepts. On historical and contemporary prompts of different lengths, it improves FID to the DEArt reference distribution in every setting with only a modest decrease in CLIP text--image similarity. Qualitative results show that the same adapter can render both period-appropriate subjects and contemporary concepts in a coherent historical visual language. By decoupling a reusable style representation from the base generator, the approach offers a practical building block for controllable historical-art media creation, interactive authoring, and rich-media experiences.

0.29

Writer-independent handedness identification from offline Bangla handwriting under controlled acquisition

cs.CV cs.CYcs.LG ·2026-11-07 ·Tanzim Hasan Prappo, Sayeed Hassan, Sayem Ali, Sukhendra Nath Dhar, M. Shahidur Rahman
unusual
Abstract · 8 pages, 7 figures, 3 tables

Handedness is a soft-biometric cue for forensic document examination and writer profiling, yet reported accuracies from offline handwriting rarely keep writers, texts and image acquisition apart, so models may recognise the writer or capture conditions rather than the writing hand. Such controlled evidence is missing for Bangla. Using our BanglaHand dataset of 42 writers (21 left- and 21 right-handed), who each wrote six shared prompts and two own texts under controlled collection and photography, we evaluate handedness identification with all images of each test writer held out, whole-writer permutation tests and photometric controls. An ink-gradient descriptor identified the handedness of unseen writers with 77.4% balanced accuracy (95% writer-bootstrap interval 67.7-86.5%; p=0.001) and frozen ImageNet embeddings with 69.5-72.0%. Part of this signal is stroke darkness: in a post hoc analysis, right-handed writers' strokes were darker at their centres for equal stroke width, consistent with higher pen pressure and not explained by lighting drift. With ink density and brightness regressed out, the descriptor still reached 64.5% (p=0.027), whereas writing-only shape features were weaker (60.8%; p=0.090). Handedness is thus identifiable from offline Bangla handwriting under controlled acquisition, and photometric controls are essential to interpret such cues.

0.29

Same Measurements, Different Velocity Risk: Complementary Rolling-Shutter Readouts under Clock Uncertainty

cs.CV ·2026-11-07 ·Lunshuo Tian, Dongheng Liu, Mingwei Ma
theoryunusual
Abstract · 36 pages, including appendices. Lunshuo Tian and Dongheng Liu contributed equally

Camera rigs with identical measurement budgets can have different velocity estimation rates when exposure timing is uncertain. We establish this distinction for rolling-shutter reconstruction with two calibrated moving anchors, unknown scene points, and independently unknown auxiliary poses and clocks. On the same fixed local physical class near a separated lateral rig, repeated readouts have worst-case squared velocity risk at least of order $σρ_{\mathrm{eff}}$, whereas complementary readouts have minimax risk of order $σ^2$, for a fixed admissible clock displacement $ρ_{\mathrm{eff}}>0$ as the noise standard deviation $σ$ tends to zero. A finite physical compensation path gives the lower bound; a nonlinear velocity inverse proves complementary stability. All motion, structure, pose, and clock nuisances remain unknown. We identify mounting configurations in which both designs become regular and quantify the risk floor induced by metric uncertainty. We quantify full-parameter domains with independent 6-ms clocks and test in-domain reconstruction under paired noise and a depth-offset intervention. Raster studies examine image formation and calibration sensitivity.

0.29

Vision Language Models for Primary Structure Threat Identification in Wildland Urban Interface Fires

cs.CV ·2026-11-07 ·Zhangding Liu, Neda Mohammadi, John E. Taylor
published
Abstract · 6 pages, 3 figures, 1 table. Accepted at the 2026 IEEE International Smart Cities Conference (ISC2), Porto, Portugal

Wildland urban interface (WUI) fires increasingly threaten communities and infrastructure, motivating the need for rapid visual assessment. Recent research has focused primarily on detecting fire and assessing post-event damage. However, structure-level fire threat identification during active WUI fire events remains underexplored. This study evaluates Qwen3-VL-8B-Instruct for this active-fire decision using two separately scored tracks: 1,824 synthetic images from 491 designed WUI scenarios and 45 frames from a controlled-burn experiment conducted in collaboration with Columbus Fire & EMS Department. Under zero-shot prompting, the model returned valid target fields for 1,808 synthetic images and 39 controlled-burn frames. Accuracy among valid outputs was 94.0% for the synthetic track and 69.2% for the controlled-burn track, compared with a 16.0% random-selection baseline for the latter. On the synthetic track, the model generally identified the intended target across varied designed WUI scenarios when the relevant fire-spread cues were visually distinct. Controlled-burn performance was well above random selection, but errors showed difficulty interpreting spatial relations in complex scenes. The dual-track protocol provides scalable stress testing in designed scenarios and a controlled-burn reference for assessing model behavior under physical fire conditions. Future work can use the synthetic data for VLM post-training and then evaluate the adapted models on larger and more diverse real WUI fire datasets.

0.28

MGE: A Membership Guided Data Extraction Attack against Diffusion Models

cs.CV ·2026-11-07 ·Yuchang Mei, Longqiang Wang, Haoyang Jiang, Mingyang Yi, Shaolei Zhang +3 more
Abstract

Recently, diffusion models have achieved remarkable performance in image generation. However, their inherent memory properties introduce a potential safety risk of training data leakage, i.e., data extraction attack. To illustrate such risk, we propose a membership guided extraction (MGE) framework to effectively extract training data from a pre-trained diffusion model. Our core idea is to formulate the data extraction as a conditional generation, where the condition signal is from a membership inference classifier. To improve the success rate and efficiency of MGE, we propose adversarial gradient and parallel branch decoding. Our experiments on image generation benchmarks demonstrate that our MGE can extract hundreds of images, which shows that more attention should be paid to privacy-preserving of diffusion model.

0.28

CompositionEvidence: Controlling Geometric Evidence for Cinematic Composition Reasoning with Frozen Vision-Language Models

cs.CV ·2026-11-07 ·Yichen He, Xin Huang, Yuxi Liu
published
Abstract · Accepted at the 33rd International Conference on MultiMedia Modeling (MMM 2027)

Cinematic composition questions depend on in-frame geometry that vision-language models (VLMs) use inconsistently. External perception tools can expose this geometry, but naive prompt injection can overwrite useful visual predictions. On ShotBench composition, pre-question geometry reduces Qwen2.5-VL-7B from 49.5% to 46.3%; corrupted measurements lower it to 33.6%. CompositionEvidence derives nine reproducible features from frozen detection, depth, and line experts and supplies their shared 24-dimensional representation through two interfaces. A training-free channel appends gated neutral evidence after the question, reaching 56.2% (+6.7, exact McNemar p=0.0018) and 48.4% under corrupted evidence. A minimal-training channel maps the same representation to four soft tokens with a 7.37M-parameter projector while leaving the VLM frozen; it reaches 61.0% (+11.5). Across nine VLMs, including five commercial models queried through official provider APIs, zero-shot accuracy ranges from 43.8% to 63.0%, and the text channel changes it by +6.7 to -6.9 points; the highest baseline is among the regressions. The effect of explicit geometry depends on measurement reliability and on how the receiving model integrates it.

0.28

Beyond Real or Fake: A Dual-Channel Authenticity and Reasoning Protocol for Photographic Assessment

cs.CV ·2026-11-07 ·Xiaoxiao Li, Ruinan Jin, Lili Meng, Miaosen Wang, Athula Balachandran +1 more
top venuebenchmarkposition
Abstract · Accepted at NeurIPS 2026

Forensic verification has converged on a binary "real vs. fake" label that confates fully synthetic, tampered, and AI-retouched images despite their very different consequences. We argue that GenAI manipulations decompose along two complementary channels: a camera channel of sensor fingerprints (destroyed by neural processing), and a semantic channel of scene-level coherence (disrupted by content edits). Each manipulation type leaves a distinctive two-channel signature. We instantiate this taxonomy in 2CAP (2-Channel Authenticity Protocol), pairing a contrastively-trained camera encoder with a frozen semantic encoder to serve two applications on a shared backbone: (i) four-class authenticity classification via cross-attention fusion with learnable reliability weights; and (ii) evidence generation, where an authentic image serves as a reference for a query and the two encoders produce per-channel alignment scores together with a patch-level saliency map. These signals are supplied as privileged context to a frozen Vision-Language Model (VLM) through an Observe-Generate-Refine loop, yielding explanations with localization without forensic fine-tuning. On a multi-source benchmark, 2CAP attains the best classification performance, including strong retouching detection where existing methods fail, and improves explanation quality.

0.28

Formalizing Patient-Set Semantics for OMOP CDM-Based Clinical Trial Feasibility Assessment:A Neuro-Symbolic Methodology for Traceable Site-Feasibility Patient Counts

cs.IR ·2026-11-07 ·Sohyeon Jeon
Abstract

This study examines the interpretation and execution of natural-language eligibility criteria for clinical trial site-feasibility patient-count estimation using the OMOP Common Data Model (CDM). It treats each count as the size of a patient set produced under explicitly declared choices about clinical concepts, temporal conditions, exclusions, proxies, and available data. The study presents NS-CoT, a neuro-symbolic pipeline that uses an LLM to interpret and structure criteria, followed by symbolic processing and deterministic SQL compilation. Intermediate representations and execution records document how criteria are translated into patient sets and counts. Evaluation uses synthetic OMOP CDM data, controlled benchmarks, and a multi-site simulation. Comparisons with a direct-SQL baseline examine execution success, count agreement, patient-set agreement, and preservation of reference-defined site-feasibility labels and rankings. Performance varies across evaluation settings. Layer-wise inspection also supports the identification of implementation and policy errors, followed by correction and re-evaluation. The work investigates the reviewability and fidelity of cohort computation relative to declared reference definitions. It does not establish that these definitions capture a unique clinical ground truth or demonstrate improved recruitment, site selection, or trial outcomes in practice.

0.28

DeepFinance: Can AI Diligence a $100MM Deal?

cs.CL cs.AIq-fin.GN ·2026-11-07 ·Drew Hahn, Tyler Shibata, Jackson Grove, Rahul Sonwalkar
benchmark
Abstract · 12 pages, 3 figures, 6 tables. Leaderboard: https://www.juliusresearch.ai

Existing evaluations of AI models on high-finance work focus on retrieving answers from financial documents and producing numerically correct outputs. Real private equity (PE) and investment banking (IB) work also depends on choosing between conflicting sources, making sound assumptions, and keeping every step auditable. We introduce DeepFinance, an ultra long-horizon benchmark of 21 end-to-end PE and IB tasks that take a human professional an average of 27.1 hours. Each task pairs a request with a data room averaging 19 files and is graded against an average of 84 expert-written rubric criteria covering output, workflow, and judgment. We evaluate 22 model configurations, each in its native agent harness. The best configuration, GPT-6 Astra with max reasoning, satisfies 57.6% of rubric criteria, and the top four configurations are not statistically distinguishable. Frontier models often calculate correctly from the wrong inputs and find decision-relevant evidence without acting on it.

0.27

SepLRM: A Foundation Model Paradigm for Recommendation

cs.IR ·2026-11-07 ·Di Chen, Xijun Xiao, Qin Ren, Weijiang Lai, Zhijian Pan +21 more
Abstract

Large recommendation models (LRMs) improve recommendation performance by scaling behavior sequence length, model capacity, and training compute. However, the return on investment (ROI) of continued scaling deteriorates as marginal gains diminish and long-sequence encoding costs rise, especially when this computation is repeatedly executed by distinct models for diverse recommendation stages and business scenarios. To address this issue, we propose SepLRM, a recommendation foundation model (RFM) paradigm that identifies reusable sequence encoding and reorganizes multiple end-to-end (E2E) LRMs into one RFM and multiple independently optimized vertical models (VMs). The RFM adopts an encoder-decoder architecture and exports its final-layer encoder state as the shared key-value (KV) cache that is independent of downstream task-offered candidates. Unlike candidate-specific logits or aggregated user embeddings, the KV cache preserves fine-grained behavioral information. Separation at this native KV interface also allows each VM to construct candidate-conditioned queries while retaining the same query-KV interaction as its E2E counterpart, enabling high-fidelity capability transfer. To ensure both transfer effectiveness and system efficiency, we co-design cache compression, the training and serving workflows, and LG-Store, a distributed storage system for KV caches. Experiments on three large-scale datasets show that SepLRM achieves state-of-the-art offline performance with a transfer rate of approximately 100%. It also demonstrates strong transferability across model iterations, recommendation stages, and business scenarios. Deployment in two large-scale industrial scenarios substantially improves the ROI of sequence scaling and delivers consistent gains in online A/B tests at ByteDance.

0.27

ROLECUE: Diagnosing Cross-Lingual Role-Binding Failures in Vision-Language Models

cs.CV ·2026-11-07 ·Niaz Mohaiman Abtahi, Evan Maksud
published
Abstract · Accepted to WACV 2027, Evaluations and Datasets Track

Vision-language models (VLMs) can recognize events in images and parse sentences, but it remains unclear when they bind visual Agent-Patient structure to linguistic semantic roles across languages whose role cues depend on morphology, word order or construction. We introduce ROLECUE, a controlled diagnostic that pairs 192 validated two-person action images with German, Turkish, Japanese, English, Korean and Bangla sentences in canonical and patient-first orders. The evaluation separates image-only role recognition, text-only role parsing, direct image-sentence matching, response-compliance behavior and an external modular composition of the model's own component predictions. Across eight recent open-weight VLMs, ROLECUE reveals a structured spectrum: some models fail visual or linguistic component tests, some largely solve the diagnostic, some collapse to constant semantic assent and several show component-joint gaps in which reliable image-only and text-only role predictions do not survive direct joint inference. The largest non-degenerate gaps reach 44.9 percentage points, with final bootstrap intervals confirming that the effect is not a point-estimate artifact. These results reveal multilingual component-joint performance gaps and acceptance biases hidden by aggregate VLM benchmarks and show why component diagnostics are necessary before interpreting joint multimodal errors.

0.27

Gendered Misclassification from Surname-Based Race Imputation

cs.CL stat.AP ·2026-11-07 ·Noah Dasanaike
Abstract

Researchers increasingly estimate political behavior based on race and gender from voter files, where race is imputed from surnames and geography. This only works where surnames suggest group membership, an assumption violated where surnames follow norms around marriage. Imputation is therefore biased against women who adopt their husband's surname, especially those who racially inter-marry, an outcome itself endogenous to partisanship. Using six US states recording self-reported race, I show that imputation is less accurate for Asian and Hispanic women, particularly those sharing the surname of a partner of another race, who differ politically from other women of their race. This violates assumptions behind downstream inference approaches, which nearly double the gender gap in partisanship among Asian voters. I derive an expression showing the bias is largest where a group is small. Using maiden rather than marriage-adopted surnames in North Carolina raises accuracy up to fourfold for women married across racial lines. Recovering maiden surnames, or omitting the married surname, removes much of the bias. In a national replication of a study of gender gaps in turnout, the correction more than triples the gap among Asian voters. In Lebanon, where wives register in their husband's locality, the error enters through geography.

0.27

EndoSCOP-V: A Multi-Turn Video Understanding Evaluation Framework for Multimodal Models in Endoscopy Reporting and Clinical Reasoning

cs.CV cs.AI ·2026-11-07 ·Abhishek Tyagi, Tanmay Jain, Vilas Biradar, Krithi Krishna Koduri, Misba Hul Bashir Masoodi +13 more
published
Abstract · NeurIPS 2026

Endoscopic diagnosis unfolds over native video, where a single positive finding during traversal of the gastrointestinal tract sets off a multi-minute chain of reasoning over location, morphology, severity and intervention. Existing endoscopy multimodal benchmarks test only static images or single-disease videos, which miss the temporal and clinical reasoning that real-world reporting demands. We introduce EndoSCOP-V, a video-grounded multi-turn evaluation framework for structured endoscopy reporting and clinical reasoning. It contains 1,307 real-world endoscopy and colonoscopy videos (63.2 h; 97% 1080p) averaging about 174 s. Its 20,274 question-answer turns are instantiated deterministically from clinician-filled structured endoscopy-reporting EHRs, so no per-case annotation is synthetic or LLM-rephrased. They come from 1,033 clinician-reviewed templates, about 18 per disease, that span a clinically aligned 9-task x 30-sub-task reporting taxonomy over 54 World Endoscopy Organization (WEO) MST 3.0 diseases. 22% of procedures carry co-existing pathologies, up to six per case. A Markovian context-pruned oracle-forcing protocol evaluates the dialogue turn by turn, injecting the gold answer before each subsequent turn so that per-turn reasoning is isolated from cascading dialogue errors. A panel of 15 board-certified endoscopists sets a 78% strict-accuracy baseline on 1,500 questions. Across 14 open-weight LMMs, the best 27B-class model reaches 59.8% strict accuracy, falling to 53.4% once format-trivial disease-rejection items are excluded. On a between-subjects spot-check, the best of 3 frontier proprietary models trails the clinician panel by 23 pp on strict NDR accuracy. Sub-task analysis reveals consistent gaps in quantitative measurement, multi-variable grading and intervention reasoning on long-horizon video. Data: Hugging Face (EndoSCOP-V). Code: GitHub (abhi25t/EndoSCOP-V).

0.26

Can General-Purpose Generative AI Supply Training Data for Medical Vision Models? A Multimodal Retrospective Study

cs.CV cs.AI ·2026-11-07 ·Jiachuan Peng, Kyle Lam, Jierui Qu, Lin Li, Haining Zheng +8 more
Abstract

Background: Well-annotated, diverse medical imaging datasets remain a major constraint for vision-based medical AI. We evaluated Nano Banana, a general-purpose image generation and editing model, as a source of synthetic medical images for model training, subgroup augmentation, and model auditing. Methods: In this retrospective study across four imaging modalities, chest radiography, fundus photography, and CT models were trained only on synthetic images and externally validated on real-world data against real-data-trained models. Carotid ultrasound models were trained on a private Human Phenotype Project cohort (aged 40-75 years) with and without synthetic augmentation. Clinicians rated image plausibility. Evaluation used AUROC, AUPR, calibration, and decision curve analysis. Findings: Synthetic-only models achieved AUROCs of 0.877 (95% CI 0.843-0.910) for pneumonia (n=14863), 0.963 (0.948-0.975) for retinal detachment (n=1149), and 0.711 (0.600-0.823) for renal calculus (n=4122), comparable to real-data-trained models, but their calibration and net benefit were not consistently more favorable. With mixed synthetic-real training, a carotid ultrasound model produced an ordinal vascular age index that stratified 633 external subjects into subgroups differing in BMI and glucose (both p<0.01), linking vascular aging to cardiometabolic health. To stress-test shortcut learning, editing a moderate tessellated pattern into fundus photographs reduced the true-class probability for central serous chorioretinopathy from 0.996 to 0.822. Clinician-guided editing raised low-quality CT plausibility from 3.10/5 to 4.30/5. Interpretation: Synthetic data can complement real data in model training, subgroup representation, and model testing via controlled editing, but its clinical value should be assessed with calibration, decision curve analysis, and clinician review, not discrimination alone.

0.26

DM-Call: Decision-Model-Guided Function Calling with Cumulative Verification Feedback

cs.IR cs.AIcs.CLcs.LG+1 ·2026-11-07 ·Minhyeong Yu, Wonduk Seo, Seunghyun Lee
Abstract · 4 tables, 2 figures

Reliable function calling requires Large Language Models (LLMs) to select appropriate functions and generate arguments consistent with user intent. However, syntactically valid calls may still be semantically incorrect due to inappropriate function selection or mismatched arguments. In this paper, we propose Decision-Model-Guided Function Calling (DM-Call), a framework that combines local schema validation with separate decision-model verification of function selection and argument correctness. For rejected candidates, DM-Call accumulates verification probabilities, binary judgments, and schema errors across attempts. A hybrid feedback representation further combines numerical probabilities with verbal correctness labels to guide iterative refinement without executing external tools. Extensive evaluation across BFCL and API-Bank (first-call) demonstrates DM-Call boosts function-calling accuracy versus direct generation, at modest additional cost.

0.26

Tiresias: Towards Language-Conditioned World Models

cs.CV ·2026-11-07 ·Fengyi Wu, Yifei Dong, Yetong Sha, Lingdong Kong, Qiyu Hu +4 more
benchmarkcode/data
Abstract · 33 pages, including appendix and references. Code: https://github.com/UWMILab/Tiresias

World models enable planning through internal simulation, while natural language provides intuitive task specification. Yet, language-conditioned world modeling remains fragmented across domains with disparate action spaces, datasets, and evaluation protocols. We introduce LCWM, a shared task that maps an initial agent-view observation and a language instruction to an action sequence that fulfills the instruction. Its accompanying benchmark, LCWM-Bench, integrates 15 datasets across navigation, robotics, 3D gaming, and driving under a shared action interface and evaluation protocol, pairing 142,987 episodes with 428,961 instructions in three styles. We further propose Tiresias, an agent that plans via selective imagination: a language-conditioned diffusion policy generates candidate trajectories, a trajectory ranker shortlists them, and a video-pretrained world model imagines their visual consequences. A visual ranker evaluates these rollouts, complementing trajectory scores to select the final plan. Tiresias achieves competitive planning and imagination performance on LCWM-Bench. Ablations demonstrate the complementary benefits of trajectory ranking and visual verification, while experiments with alternative proposal methods show that the trained rankers remain effective beyond our diffusion policy. Together, LCWM-Bench and Tiresias provide a unified evaluation setting and an effective starting point for studying language-guided planning through world models.

0.26

Global Context Guidance for Diffusion Large Language Models

cs.CL ·2026-11-07 ·Jeimin Jeon, Bumsub Ham
top venue
Abstract · Accepted to NeurIPS 2026

Diffusion large language models (LLMs) have recently emerged as a promising alternative to autoregressive LLMs, enabling parallel generation through iterative denoising. Recent guidance methods improve diffusion LLMs by introducing a weak model to guide the prediction of the base model during denoising. However, the influence of the weak model during denoising remains less clear, and computing its prediction requires an additional forward pass at each denoising step. We present in this paper an in-depth analysis of how diffusion LLMs use contextual information during denoising. Our analysis reveals that diffusion LLMs rely more heavily on local context while often underutilizing global context. Based on this observation, we introduce Global Context Guidance (GCG), a feature-space guidance method that enhances global-context utilization without constructing an explicit weak model. Extensive experiments on LLaDA-8B, Dream-7B, and LLaDA-1.5 across diverse benchmarks demonstrate that GCG consistently outperforms existing guidance methods, while incurring much lower inference overhead.

0.26

BAmbiDrop: Array-Agnostic Speech Enhancement with Zero-Shot binaural Reproduction using Multi-center Ambisonics Encoding

eess.AS ·2026-11-07 ·Michael Tatarjitzky, Vladimir Tourbabin, Boaz Rafaely
Abstract · Submitted for ICASSP 2027

Array-agnostic enhancement networks generalize across microphone geometries but produce a single channel, while binaural enhancement methods preserve interaural cues only for one fixed device geometry. We propose BAmbiDrop, a framework that extends array-agnostic enhancement to binaural reproduction without retraining. Exploiting the fact that the Ambisonics origin is an arbitrary point in space, a frozen single-channel AmbiDrop network estimates a complex mask at multiple centers across the array; these masks are fused by a complex mean into a single shared mask, applied to the microphone signals and rendered by fixed BSM filters. Across 21 free-field and rigid-sphere arrays, the fused mask improves enhancement and interaural cue reproduction simultaneously, whereas masking each microphone independently degrades the latter without a consistent gain in the former.

0.25

When Safety Becomes An Outlier: Understanding the Retention of LLM Safety Behaviors

cs.CL ·2026-11-07 ·Binchi Zhang, Hadi Abdullah, Yiwei Cai, Jundong Li
top venue
Abstract · Accepted at NeurIPS 2026

Safety-aligned large language models often lose their safety behaviors after fine-tuning, even when safety data are included. While this fragility is well documented, its underlying cause remains unclear. In this paper, we propose that formulaic refusal strings can function as outlier modes: narrow, weakly grounded response patterns that are easy to fit but vulnerable to subsequent updates. Controlled experiments support this account: fixed refusals are acquired much like arbitrary constant strings, in contrast to prompt-grounded natural responses. Motivated by these findings, we introduce Naturalness-Enhanced Safety Targets (NEST), a safety alignment framework that improves safety retention by constructing natural, context-aware safety targets that keep safety responses within the model's existing output distribution and grounded in semantics. We instantiate NEST in both supervised and preference-based alignment settings, using perplexity under the original model as a proxy for distributional support. Across multiple models and alignment regimes, natural safety targets improve both immediate safety and its retention under continued fine-tuning without sacrificing helpfulness. These findings suggest a practical principle for more durable safety alignment: construct safety targets that draw on the model's existing language capabilities and remain grounded in the meaning of each request.

0.25

Agent2BIM: A Systematic Evaluation of Reality-Capture Configurations and Prior Models for As-Built Modeling

cs.CV ·2026-11-07 ·Boyu Wang, Borja García de Soto
Abstract

AI agents can generate as-built Building Information Models (BIMs) from reality-capture data, but the evidence required for accurate reconstruction remains unclear. This paper presents Agent2BIM, an agent-based Scan-to-BIM framework, and evaluates seven capture configurations in a research laboratory and a mechanical, electrical, and plumbing (MEP) scene. Inputs include 360° panoramas, smartphone video, smartphone LiDAR, terrestrial laser scanning (TLS), TLS with posed panoramas, and imagery supplemented by prior IFC models. Outputs are compared against TLS-verified reference models for component detection, geometric accuracy, and MEP connectivity. In the laboratory, component recall ranged from 84% to 100%, but only 9% of components reconstructed from panoramas were within 20 cm of their reference positions, compared with 86% using smartphone LiDAR. In the MEP scene, scale-free inputs recovered 11 - 43% of components, versus 75 - 93% for metric inputs. Prior IFC models improved MEP reconstruction but could propagate inaccuracies. The results highlight the complementary roles of visual and metric evidence in agent-based Scan-to-BIM.

0.25

FG-DSA: Frequency-Guided Depthwise Separable Attention for Efficient Dense Prediction

cs.CV ·2026-11-07 ·Yiyao Yang
Abstract · 20 pages, 8 figures, 9 tables

Dense prediction and generative modeling in resource-constrained scenarios remain challenging, as repeated downsampling substantially reduces spatial detail and degrades boundaries. Although recent attention mechanisms have alleviated the problem of semantic ambiguity, they often fail to preserve high-frequency features or do so at the expense of a very high computational cost. In this paper, we propose the Frequency Guided Depthwise Separable Attention for Efficient Dense Prediction (FG-DSA), a novel framework for efficient dense visual understanding and generation. FG-DSA employs a 2D FFT high-pass filtering module to explicitly inject frequency domain priors into the spatial attention mechanism, allowing lightweight networks to recover sharp boundary transitions with linear complexity. A range of experiments show that FG-DSA is effective at recovering fine object boundaries and leads to a consistent performance gain in the task of semantic segmentation. Importantly, we show that FGA-DSA can be used as a module in diffusion-based generative U-Nets, leading to a significant improvement in the detail of the generated images with very little additional parameter and Floating Point Operation (FLOP) cost.

0.25

TransMem: Transition-Aware Retrieval for Evolving Personal Memory

cs.CL ·2026-11-07 ·Shigeng Chen, Linhao Luo, Changlong Shi, Zhangchi Qiu, Shirui Pan
top venue
Abstract · Accepted at NeurIPS 2026

Long-term personalized agents are increasingly expected to answer queries over user histories in which preferences, goals, and constraints evolve across interleaved conversations. Existing retrieval-based and structured-memory systems can surface relevant facts, but often return isolated memory units that do not explain which earlier preference was refined, contradicted, or superseded by later evidence. We introduce TransMem, a transition-aware memory retrieval framework that stores preference updates as explicit retrieval units alongside ordinary factual memories. During memory construction, TransMem routes user interaction history into topic-structured memory graphs, detects shift events, and grounds them as transition records. At inference time, TransMem first retrieves factual anchors and then adaptively expands over transition records, producing a compact context of factual anchors and accepted transition records. Experiments on PersonaMem and PersonaBench across two Qwen backbones show that TransMem outperforms representative retrieval-based and structured-memory baselines. Controlled analyses further show that transition records and adaptive expansion contribute beyond factual anchoring, chronological expansion, flat transition retrieval, and fixed-hop traversal. These results suggest that preference-update records are useful retrieval units for long-term personalization.

0.25

Simple Test-Time Refinement for Plot-to-Code Generation via Visual-Code Diagnostics

cs.CL cs.CV ·2026-11-07 ·Peiming Guo, Jinpeng Huang, Yangbin Sun, Gaoxiong Cao, Ensheng Shi +5 more
top venue
Abstract · Accepted by NeurIPS 2026

Plot-to-code generation has been greatly advanced by recent multimodal large language models (MLLMs). However, existing studies mainly focus on one-pass generation or iterative sampling guided by verifier-based filtering, which does not align with the typical human programming workflow of generate first, then iteratively error-location and fix. This naturally motivates a test-time refinement paradigm for plot-to-code generation. In this work, we present the first study on plot-to-code generation via iterative test-time refinement. We propose a Visual-Code Diagnostics framework that enables refinement through visual feedback from source code. Specifically, we first categorize generation errors into four coarse-grained types, and then train lightweight discriminators to recognize and localize these errors based on the rendered visual outputs and corresponding code. Unlike traditional compiler- or debugger-based verification, our approach performs iterative refinement guided by one diagnostics system, enabling more effective semantic and visual correction during test time. Extensive experiments demonstrate that our method consistently improves generation quality across diverse generators and significantly outperforms strong test-time scaling baselines.

0.25

Steer Unified Audio Editing Toward Reliable Source Selection

cs.SD ·2026-11-07 ·Youquan Fu, Dongnuan Cai, Yu Pan, Henghui Du, Yujia Zhang +6 more
benchmarkcode/data
Abstract

Unified audio editors support diverse operations, yet often struggle to modify the intended source while preserving non-target content. We show that source selection in audio editing is shaped by early denoising states, while incorrect early selections can hardly be corrected during subsequent denoising. Motivated by these findings, we propose Source-selection Training via Early-trajectory Emphasis and Recovery (STEER), a two-stage post-training algorithm for reliable selective audio editing. STEER-Emphasis prioritizes training on task-critical early states to reduce source-selection errors, while STEER-Recovery learns corrective denoising from self-generated rollout states to recover from early deviations. To systematically assess source-selection reliability, we introduce the Selective Audio Editing Benchmark, enabling large-scale automatic evaluation of target manipulation and non-target preservation. Applying STEER to our pretrained multitask base editor gives SteerSound, a unified audio editor with leading selective editing performance among open-source audio editors. Experiments on our base editor, UNISON, and SAM Audio demonstrate that STEER consistently improves both selective editing capability and overall editing performance. Early-trajectory emphasis further brings substantial improvements in transcript-conditioned target speaker removal and separation, highlighting the broader applicability of this training principle.

0.24

Yuanbao-Reranker: Document-Grounded Synthesis and Preference Distillation for Text Reranking

cs.IR ·2026-11-07 ·Hongyuan Xu, Dezhi Ye, Jiahao Cao, Jie Liu, Haijin Liang +2 more
Abstract · 23 pages, 6 figures, 7 tables

High-precision text reranking is a critical component in modern search, recommendation, and agentic retrieval-augmented generation (RAG) systems. We present Yuanbao-Reranker, a family of general-purpose dense rerankers spanning 0.6B, 4B, and 8B parameter scales built upon the Qwen3 backbone. To construct diverse and reliable ranking supervision at scale, we pair answer-first and query-type-guided document-grounded synthesis with a three-stage curation funnel that verifies positive support, eliminates semantic redundancy, and enforces balanced source-by-length quotas. For each curated pair, we screen real retrieval pools with a teacher score margin to filter out potential false negatives and sample Hard, Medium, and Moderate negative tiers, yielding 2.75M $1{:}3$ ranking groups. We train the models by jointly optimizing pointwise relevance cross-entropy and a bounded minimum-gap hinge objective over both positive--negative and negative--negative pairs. This formulation transfers fine-grained teacher preferences across candidate tiers without over-constraining well-separated predictions, supporting both cross-scale and same-scale distillation followed by layerwise adaptive model merging. Across five comprehensive benchmarks spanning Chinese (CMTEB-R), multilingual (MMTEB-R), English (BEIR), long-document (MLDR), and code (MTEB-Code) retrieval, Yuanbao-Reranker consistently leads every parameter tier: the 0.6B variant reaches 70.96 Avg. to surpass 1B--4B active-parameter baselines, the 4B variant reaches 73.63 Avg. to outperform the 8B teacher (72.68), and the 8B variant achieves a state-of-the-art 74.21 Avg., with substantial gains generalizing across distinct first-stage retrievers.

0.24

Unified Framework with Convex-Concave Decomposition and Sparse Topological Learning for Skeleton-Based Clinical Assessment

cs.CV cs.AI ·2026-11-07 ·Quang-Anh N. D., Duc Pham Minh, Minh-Anh Nguyen, Hien Quang Kha, Manh Hung Ha +2 more
Abstract

Skeleton-based rehabilitation assessment faces fundamental limitations of fixed topologies which cannot model compensatory patterns, uniform joint treatments overlook clinically important deviations, and lack of geometric understand- ing. We propose Concave-Convex Decomposition Sparse Topological Graph Convolutional Network (CC2-STG), a multi-stream model that combines geo- metric decomposition, sparse feature learning, and adaptive topology. Our framework decomposes joint relationships into complementary convex-concave distance functions to model both local and global movement characteristics, with sparse regularization with proximal optimization selecting clinically relevant geo- metric features, while multi-scale topological analysis captures non-anatomical joint dependencies in pathological movements. Experiments on the KiMoRe and UIPRMD datasets validate the effectiveness of our approach, achieving state-of-the-art performance on clinical pathological movement assessment.

0.23

Quadratic Bezier Splatting with Structural Initialization for Efficient Painting Rendering

cs.CV cs.GR ·2026-11-07 ·Jinfan Liu, Xuhan Zhan, Mengqin Zhao, Guangyi Deng, Wuze Zhang +1 more
Abstract

Stroke-based rendering is often optimized as a pixel-level reconstruction task. This over-reliance on global error reduction not only degrades strokes into fragmented curves due to a lack of structured planning, but also allows locally defective primitives to be concealed by overlapping layers. To address these issues, we introduce a model-guided framework for stroke-audited rendering. Our framework uses Quadratic Bezier Splatting to analytically model continuous curved strokes with efficient differentiable alpha compositing. Built on this renderer, a feedforward multi-scale stroke decoder provides a structural initialization with learned stroke-scale allocation, while test-time stroke optimization supplies the reconstruction capacity needed for high-fidelity refinement. We further propose a single-stroke audit mechanism that explicitly evaluates local stroke quality, preventing defective primitives from surviving optimization through layer-wise compensation. Compared with existing methods, our approach achieves coherent layouts, improves PSNR by up to 5.46 dB, yields a 16% more compact representation, and accelerates combined renderer forward and backward computation by a factor of 220.3.

0.23

From Reflection to Co-Evolution: Self-Evolving Agents with Locally Verified Memory

cs.CV ·2026-11-07 ·Xinxiang Yin, Xinglang Zhang, Haodong Chen, Zhiqing Cui, Weidong Tang +5 more
Abstract · 31 pages, 9 figures, 6 tables

LLM agents have made impressive progress in coding and mathematical reasoning, motivating growing interest in learning from long-horizon interaction. Memory supports this process, but plausible lessons can be redundant or reinforce inappropriate actions. Can a memory's source decision provide a practical test for whether to retain it? We localize consequential trajectory steps, extract candidate memories, and replay each source decision with and without its candidate under the same context. This reveals two properties: most candidates barely shift the probed distribution, and both successful and failed source actions can be reinforced or suppressed. These observations motivate Anchor, which tests shift magnitude and source-role alignment before retention. Admitted memories guide later trajectories, enabling experience-driven improvement under a frozen policy. On held-out ALFWorld and WebShop tasks, Anchor improves success rate and mean score by up to 19.9% and 17.2%, respectively, relative to the strongest baseline for each backbone. With Qwen3.5-27B, it retains 41% fewer ALFWorld cards than unfiltered retention and outperforms all four WebShop cost-comparison baselines while using fewer tokens.

0.23

Heterogeneous Asymmetric Contrastive Distillation

cs.CV ·2026-11-07 ·Qi Wang, Jinjia Zhou
positioncode/data
Abstract · 13pages, 7 figures

Knowledge distillation (KD) between architectural families, e.g., from a Transformer teacher to a CNN student, is difficult because the two networks learn different representations. Logits share a label space, but heterogeneous teachers differ in logit offset, scale, and sharpness, and their non-target classes are less semantically organized. Conventional logit distillation matches these values sample by sample, asking the student to imitate how another architecture expresses its predictions. We argue that a heterogeneous teacher should pass on the relational structure of its predictions, i.e., how it arranges samples relative to one another, and propose Heterogeneous Asymmetric Contrastive Distillation (HACD) to transfer this structure. HACD maps teacher and student logits to centered, temperature-softened probability directions on a shared unit hypersphere, which removes architecture-dependent offsets and lets the temperature control how sharply peaked outputs enter the representation while keeping informative non-target classes. Each student output must then identify its own teacher target within the batch. Because every teacher target also acts as a negative for the others, HACD computes the targets on a moderately augmented view and contrasts both a moderate and a strong student view against them. HACD adds no learnable parameters and needs one teacher pass per sample. It achieves the best or second-best accuracy on 11 of 12 heterogeneous pairs on CIFAR-100, without ground-truth labels, and on 14 of 15 pairs on ImageNet-1K, improving over the reported vanilla KD results by 8.82 and 1.33 percentage points on average. Checkpoints are available at https://github.com/BTAEYANG/HACD-Heterogeneous-Asymmetric-Contrastive-Distillation; the code will be released upon publication.

0.23

Understanding the Effect of Visual Attention in CNNs: A Decision-Space Alignment Perspective

cs.CV cs.AI ·2026-11-07 ·Niladri B. Puhan, K. Kamala Priya
Abstract

Visual Attention enhances convolutional neural networks (CNNs) by modulating intermediate features, yet their effects on downstream decisions remain less understood. We present a decision-space framework that models multiplicative attention as an intermediate feature perturbation and derives a first-order prediction of its effect on the target-versus-competitor classification margin. We further introduce the Decision Alignment Score (DAS) to quantify the directional alignment between the perturbation and the local margin gradient. We evaluate the framework on CIFAR-100 using SimAM and CBAM across five popular CNN backbones, ResNet-18, ResNet-20, ResNet-50, ResNet-56, and MobileNetV2. The analysis examines prediction agreement, DAS margin correlations, and downstream activation-state changes. Across most final-attention configurations, predicted and actual margin changes show good agreement, while DAS exhibits positive correlations with actual margin changes. Single-attention experiments further show that prediction agreement and DAS margin correlations depend on attention placement, with stronger associations at later stages.

0.22

The Truth Beneath the Tokens: Prompt-Space Recovery of Knowledge Edits in Language Models

cs.CL cs.AI ·2026-11-07 ·Manit Baser, Rachel Yeo Hui Min, Aditya Nawal, Dinil Mon Divakaran, Mohan Gurusamy
Abstract

Model editing provides an efficient way to update factual knowledge in transformer models. Recent approaches reverse edits by modifying, masking, or decomposing model weights, but make recovery dependent on weight-space interventions. We instead show that pre-edit knowledge can be recovered through input-space interventions alone, without modifying the edited model's weights. We introduce Memento, a prompt-space mechanism for auditing residual knowledge that learns a reusable recovery signal to steer an edited model toward its pre-edit response. Memento supports both continuous recovery prefixes and discrete textual prompts for standard text interfaces. Across three LLMs and five editing techniques, Memento recovers 80.07% of held-out edited facts, achieving a 9.58% relative improvement over baselines, suggesting that editing may suppress the facts without fully eliminating their recoverability. We further investigate whether prompt-space recovery extends beyond model editing. On pristine, unedited LLMs, Memento improves average factual recall by 53.9% relative to standard inference. These results demonstrate the potential of prompt-space recovery for auditing residual knowledge in edited models and investigating the gap between factual knowledge and its accessibility during inference. We will release our source code.

0.22

Track or Follow? A Jev-Compatible Multi-Agent Trading System When Web News Lags the Market

cs.CL cs.AIcs.CRcs.IR+1 ·2026-11-07 ·Yebo Feng, Guangyan Gan, Liangrui Pan, Ruichao Liang, Cong Wu +2 more
Abstract

LLM agents increasingly trade on the Web news they read, yet much Web news about listed firms is published after, and about, the price move it reports, so a reader can agree with the market by following it instead of reading the event. We ask whether LLM judgments of Web news track the market's reaction to the reported event or follow it. To answer, we built a multi-agent trading system whose agents share only point-in-time stores and answer typed questions through the interface of Jev, a hosted typed decision model; code sets every position under a proved bound on how far text can move it. In a hash-committed test of a Jev-compatible reader built from a pinned LLM's option probabilities, item judgments track announcement-window abnormal returns (weekly Spearman 0.119), also after the LLM's release, and event types agree with the issuers' 8-K filings (AUC 0.833); Jev's judgments of news published after its own release track the reaction as well (descriptive). Exploratory analyses show where the association comes from: 72% of earnings news items appear after the issuer's filing, contributor-written items often report the price move, and from the first trade after publication the association falls to 0.023. At a weekly horizon, the confirmatory test detects no rank information over code-only features. Under attack, decision cards traced 88% of successful attacks to the planted item. We propose that market-graded tests of Web-text readers report the criterion from publication, not only from the event.

0.22

ZARA: Zero-shot Animatable and Relightable 3D Avatar reconstruction from a single image

cs.CV ·2026-11-07 ·Yixiao Chen, Dongdong Weng, Chengfeng Zhao, Tao Gong, Zhihe Zhao +3 more
Abstract

Recent large human reconstruction models create remarkable animatable 3D avatars from a single image, but their appearance remains entangled with the illumination and cannot be relighted. In contrast, existing 3D relightable-avatar reconstruction methods typically require person-specific optimization. We present ZARA, a zero-shot feed-forward framework that reconstructs an animatable 3D Gaussian avatar together with its intrinsic attributes from a single image. ZARA couples a pretrained large-scale human reconstruction model with a Material Transformer to inherit geometry priors while learning material-aware Gaussian representations. To preserve localized details, we introduce Gaussian-aligned material feature fusion strategy, which injects multi-scale material features onto individual Gaussians and combines them with geometry tokens. As most human-avatar datasets lack ground-truth material maps, we use pseudo-material labels as conditioning inputs and supervision targets during training. We further introduce MatFace regularization to preserve facial structure and color in albedo prediction under unreliable facial pseudo-material supervision. Extensive experiments demonstrate that ZARA reconstructs animatable and relightable humans within tens of seconds without person-specific optimization or post-processing, outperforming existing baselines in both relighting quality and materials generalization ability. As the ability of recovering the physically-based-rendering attributes of the human, ZARA also enables direct material editing of the avatar's appearance.

0.22

KeyASRBench: Measuring Whole-Phrase Keyword Preservation in Real-World ASR

cs.CL cs.AI ·2026-11-07 ·Guanzheng Wei, Ziyi Zhao
Abstract · 8 pages, 3 figures, 4 tables. Bilingual ASR evaluation with a 10-hour model-conditioned challenge set and a disjoint 30-minute model-blind c…

Public automatic speech recognition (ASR) benchmarks increasingly report low sequence error rates, yet production systems still use hotword lists, contextual biasing, retrieval, and post-correction. This motivates a complementary question: does a transcript preserve selected names, products, technical terms, versions, and relational phrases? We present KeyASRBench, a bilingual pilot that operationalizes occurrence-based Keyword Error Rate (KER) over frozen whole-phrase annotations. The 10-hour model-conditioned challenge set contains 1,501 Chinese and English clips with 4,723 keyword occurrences. Across five locally inferred checkpoints and two official API systems, language-specific MER/WER ranges from 5.31% to 12.91%, while KER ranges from 13.61% to 33.81%. A disjoint 30-minute model-blind control yields KER of 10.65%-27.31%. The control confirms that selected-phrase errors extend beyond challenge-set selection.

0.22

Open Problems in Chatbots and Mental Health

cs.CL cs.CY ·2026-11-07 ·Jared Moore, Jacy Reese Anthis, William Agnew, Andrea Mock, Kevin Klyman +6 more
positionhot direction
Abstract

General-purpose chatbots are increasingly used for emotional and mental health support. They can be useful, but issues such as sycophancy, reinforcement of delusional beliefs, and poor crisis response create risks that are difficult for researchers to study and evaluate. In this paper, we review the most critical open problems for researchers and policymakers in the use of general-purpose chatbots and their effects on mental health. First, there are tractable problems in the controlled evaluation of chatbots: establishing the construct and ecological validity of evaluations, evaluating multi-turn interactions, and connecting evaluations to improvements in chatbot behavior. For example, conventional benchmarks that primarily rely on static datasets cannot establish chatbot safety, and novel simulation methods need validation. Second, researchers must connect their studies of chatbots and mental health to real-world evidence. This includes analyzing the chatbot products that AI companies and mental health practitioners have deployed, including their user interfaces and human review and escalation practices for user welfare concerns. However, clinical analysis opportunities afforded by real chat log data sharing are complicated by significant privacy concerns. Finally, we argue that researchers should take an active role in the construction and refinement of evidence-based policy through corporate and governmental decisions such as user protections, standards of care, protected access to user chat logs, data sharing between institutions, incident reporting, and post-deployment monitoring. Such behavioral evaluations, studies of real-world outcomes, and data governance depend on each other and must be addressed in tandem. We believe a coordinated research effort would have substantial real-world impact.

0.22

PoseBench: Probing Pose Awareness in Visual Foundation Models

cs.CV cs.AI ·2026-11-07 ·Most Nilufa Yeasmin, Draix Wyatt, Ruyi Lian
benchmark
Abstract

Pose awareness characterizes the extent to which a learned feature space aligns with the underlying object-pose manifold, such that observations with similar poses exhibit greater feature similarity than those with increasingly different poses. While such structure is desirable for pose-sensitive 3D vision tasks, visual foundation models pre-trained on individual 2D images typically lack explicit 3D or pose supervision. This raises a fundamental question: Do 2D visual foundation models exhibit pose awareness after large-scale pretraining? To answer this question, we introduce PoseBench, a systematic benchmark to probe pose awareness in pretrained visual features. Our evaluation follows a progressive design: held-out rendered views probe intrinsic pose-aware structure under controlled conditions, while real-image queries evaluate how robustly this structure remains accessible under real-world variations. We further introduce pose-graph-based refinement as a complementary probe of compatibility between feature neighborhoods and the underlying pose manifold. Across 31 publicly available checkpoints from 10 pretraining families, PoseBench reveals that pose awareness is an implicit but non-uniform property of 2D visual foundation models, and consistent gains from pose-graph refinement suggest that feature spaces retain structure compatible with the underlying pose manifold. PoseBench therefore provides a systematic basis for understanding and selecting pretrained features for pose-sensitive 3D vision tasks.

0.21

Sensitivity-Gated Accumulator Caching for On-Device Learning of Low-Bit Hand Networks

cs.CV cs.LG ·2026-11-07 ·Soroor Ghandali, Kanika Mamgain, Manjot Kaur, Akanksha Shukla, Shashira Guntuka
theory
Abstract · 28 pages, 7 figures, 15 tables

Small hand-tracking networks are deployed as 4- and 8-bit codes, and they should keep adapting to a user's camera after deployment, on the device and within roughly the memory of the model itself. Quantization-aware training cannot do this cheaply, because it keeps a 32-bit master copy of every weight together with its optimizer state, about 96 bits per weight. Training the codes directly with stochastic rounding needs no copy but injects rounding noise at every step, and the methods that remove this noise keep a residual buffer for every weight. We introduce sensitivity-gated accumulator caching (SGAC), which keeps residual state for only a small and changing set of weight groups. A set-associative cache of 8-bit accumulators collects the sub-step updates of the groups with the largest Fisher-weighted squared code step and writes them into the codes as whole steps; an evicted line is flushed by stochastic rounding, and a group without a line does not move until it is admitted. We prove that the write-back is unbiased and show, under a local quadratic model, that rounding a sub-step update costs loss of the same order as the update's own gain, so that rounding pays only where the expected gradient is large relative to the curvature. On a 383k-parameter hand-keypoint network, SGAC trains with 8.1 bits of state per weight. Scored on the 4/8-bit network that the device runs after adaptation, it has the highest mean accuracy of nine deployed low-bit networks on two simulated cameras. On twelve ImageNet-C corruptions it has the highest average of corrupted and clean accuracy of seven deployed methods, ahead of the PEQA-style, ECO-style and direct low-bit baselines with confidence intervals that exclude zero, and it matches an 8-bit accumulator kept for every weight with 56% of that state.

0.21

Think or Remember: The Exchange Rate of Agentic Inference

cs.CL cs.AI ·2026-11-07 ·Juan Wisznia, Santiago Luis Groba Alonso, Matías Gómez Taddeo, Gonzalo Zabaleta, Joaquín Cullen +1 more
Abstract

Agents built on reasoning models resend their growing transcript at every call. Two transacting parties, not the model, decide what that costs: the client keeps or edits messages; the server decides whether earlier reasoning returns in the prompt and what its cache reuses. We define a cost model whose counts match an instrumented engine exactly; the architecture charges $c$ per token processed and $β$ per pair attended. Their ratio $P^*=c/β$ is an exchange rate between thinking and remembering: per token, processing dominates below this context length, a thinking regime, and attention above it, a memory regime. A policy change has a mechanical cost, incurred even if the agent writes the same tokens, and a behavioral one: it writes different tokens, changing cost and outcome. Serving alone can flip which policy is cheaper: on recorded Qwen3-8B ScienceWorld runs, dropping earlier reasoning costs 1.74 times keeping it when the server reuses generated states and 0.34 times when it reuses nothing. In live reruns, dropping it cuts correct HotpotQA answers by 9-23 percentage points on four models, at lower point-estimate cost. Comparing context policies means comparing systems: architecture and serving set the cost of the history kept and the work it induces.

0.21

Frequency-Structure Complementary Modeling for Hyperspectral Image Fusion

cs.CV cs.AI ·2026-11-07 ·Song Qiya, Tian Siyu, Li Shutao, Dian Renwei, Sun Long +1 more
hot direction
Abstract

Existing methods mainly rely on convolutional aggregation or attention-based spatial-spectral interaction. However, conventional feature-domain attention does not explicitly characterize variation patterns along spatial-token and learned latent-channel coordinates, while spatially shared local operators lack the flexibility to adapt neighborhood aggregation to heterogeneous structures. To address these limitations, we propose a Frequency-Structure Complementary Modeling Network (FSCM-Net), which couples frequency-aware dependency learning with structure-adaptive local refinement. Specifically, a Spatial-Spectral Frequency Calibration Attention (SSFCA) module constructs spatial-token affinities from Fourier-interacted query-key representations and builds latent-channel affinities after input-conditioned frequency filtering along learned channel coordinates. The latent-channel response is then calibrated by a learned shared gain and combined with the spatial-token response. Complementarily, a Structure-Coherent Dynamic Feature Fusion (SC-DFF) module combines regularized structure-tensor cues from MSI features with context-modulated HSI detail representations to generate channel- and position-dependent aggregation kernels. This design makes local HSI refinement responsive to smooth regions, edges, and textures while maintaining spectral fidelity. The two modules are progressively integrated within a multiscale encoder-decoder to coordinate representation-level interaction and structure-adaptive reconstruction. Experiments on two simulated benchmarks and one real remote-sensing dataset demonstrate favorable performance against representative fusion methods in spatial detail reconstruction and spectral preservation.

0.20

SWEET: Mitigating Spurious Feature Reliance in LLM Reasoning via Weight Editing

cs.CL cs.LG ·2026-11-07 ·Zhizhang Fu, Lucas Zhao, Michalis Korakakis, Mengru Ding, Lukasz Szpruch +1 more
Abstract

Reasoning in large language models can be disrupted by simple input perturbations (e.g. adding irrelevant distractors), suggesting reliance on spurious correlations. Identifying and suppressing the spurious features involved in long chain-of-thought reasoning remains challenging. We introduce Spurious-feature WEight EdiTing (SWEET), a training-free method that combines counterfactual interventions with mechanistic interpretability to identify and suppress spurious-features. SWEET extracts token- and sequence-level behavioral directions from activation contrasts between original problems and their counterfactual variants - capturing local and global patterns of spurious reasoning. It then localizes the responsible MLP neurons and attention heads and suppresses their contributions along these directions through weight edits. Using only 20 original-counterfactual pairs for direction extraction, SWEET improves accuracy by an average of 11.8 percentage points on perturbed inputs and 2 points on original inputs across five open-weight reasoning models and five reasoning benchmarks. It also reduces the relative gap between original and perturbed accuracies from 17.6% to 2.5%. Analyses identify two recurring forms of spurious reasoning, i.e. over-integrating irrelevant conditions and overthinking, and provide mechanistic evidence that SWEET improves reasoning robustness by suppressing both behavioral directions.

0.19

Pandemonium: Longform Target-Speaker Speech Recognition with Cross-Speaker Interactions

eess.AS cs.CLcs.SD ·2026-11-07 ·Alexander Polok, Dominik Klement, Ivan Medennikov, Antoine Laurent, Petr Pálka +7 more
code/data
Abstract · Submitted to ICASSP 2027

Diarization-conditioned target-speaker ASR (TS-ASR), in particular Diarization-conditioned Whisper (DiCoW), is an accurate and data-efficient approach to multi-talker transcription. It has, however, three structural limitations: the model transcribes each speaker independently, operates on short (30 s) windows, and optimizes its diarization and TS-ASR components separately. We address the first two. We integrate the diarization conditioning into a longform Parakeet backbone (DiCoP) and add cross-speaker attention, which lets the model attend across the speaker dimension so that speakers are no longer decoded in isolation, and an STNO mask refiner that re-estimates the diarization-derived conditioning. As conversational data with genuine multi-speaker interaction is scarce, we scale training with a pseudo-labeling pipeline that mines 5.4k hours of in-the-wild multi-speaker audio from YODAS2. Combining these with synthetic and real data yields state-of-the-art results across several meeting benchmarks. We open-source the resulting models, DiCoP and Pandemonium, and the pseudo-labeled data.

0.18

Remembering to Reason: Structured Memory for Scene and Workflow Understanding in Robot-Assisted Surgery

cs.CV ·2026-11-07 ·Zhiran Zhang, Hu Wang, Yichen Wu, Yutong Xie
Abstract

Surgical video understanding is an important foundation for intelligent assistance in robot-assisted surgery. Similar surgical views can reflect different actions or procedural stages. Interpreting these views requires access to earlier observations, yet frame-centric prompting leaves this history unavailable. We introduce MemSurg, a framework that remembers surgical observations before reasoning over them. Its structured memory graph stores timestamped entity categories, locations, motion cues, and interaction summaries. Chronological links preserve recent context, while entity-associated links connect recurring observations across the procedure. Task-adaptive retrieval then supplies evidence-grounded prompts for scene and workflow reasoning, supporting both frozen inference and memory-conditioned supervised fine-tuning. Evaluation on robot-assisted prostatectomy and two complementary surgical benchmarks shows a 10.43-percentage-point gain over SurgRAW with GPT-4o. Memory conditioning adds 6.52 points under matched Qwen3-VL-8B fine-tuning, and an inference-time plug-in adds 7.04 points to EndoChat. These results support structured memory as a reusable source of procedural evidence for surgical video understanding. Our code will be publicly released.

0.18

Evaluating Jev on Arabic: Comparing Decision Models and LLMs Across Arabic Tasks and Dialects

cs.CL cs.AI ·2026-11-07 ·Hossam Elsafty, Tobias Deußer, Lorenz Sparrenberg, Rajkumar Ramamurthy, Rafet Sifa
Abstract · 28 pages, including references and appendices

How does Jev compare with other systems on Arabic tasks, and how does its performance change between Modern Standard Arabic (MSA) and dialects? We evaluate 67 systems, including 47 decision models and 20 large language models (LLMs), on 43 tasks across 9 task families using a shared zero-shot protocol. The Arabic-focused frozen suite includes English controls in three parallel resources, which we also remove in a separate descriptive comparison. Jev 1.14.0 achieves a mean task-primary metric of 69.24 on a scale multiplied by 100, compared with 68.77 for Jev 1.13.0 and 71.81 for Gemma-4-31B, the highest-scoring system on this aggregate. Removing the recorded English controls, repeated cross-task items or both leaves the full system order unchanged. Relative performance and gains between Jev versions vary across task families. In content-matched comparisons within Jev 1.14.0, pooled dialect-minus-MSA accuracy gaps range from -7.94 to -0.35 percentage points across four datasets; dialect-specific results show larger variation, and SIB-200's source-clustered interval includes zero. Paired contrasts with Jev 1.13.0, Gemma-4-31B and Jais-2-8B show a larger Palestinian banking penalty for Jev 1.14.0 than for either LLM under simultaneous uncertainty; the other gap contrasts remain inconclusive. These descriptive findings show why aggregate scores should be read alongside task-family results, coverage and matched language comparisons. They do not establish a general model ranking for dialect robustness or explain the causes of the observed gaps.

0.18

YRC-Bench: Benchmarking Culturally Aware Machine Translation for Historical Heritage via Tri-Level Semantic Bridging

cs.CL ·2026-11-07 ·Yan Huang, Aobo Li, Jiayi Wang, Huidong Zhu
benchmark
Abstract

Translating historical cultural heritage poses profound challenges for machine translation systems due to the acute semantic asymmetry across eras and cultures, where terse classical syntax frequently obscures domain-specific cultural knowledge. While Large Language Models have demonstrated remarkable prowess in generic translation benchmarks, they struggle to negotiate Culturally-Specific Items and diachronic shifts in domain-rich historical texts, such as Yellow River cultural records. To bridge this critical gap, we introduce YRC-Bench, the first human-curated, multi-parallel benchmark dedicated to historical heritage translation across Classical Chinese, Modern Chinese, and nine target foreign languages. YRC-Bench establishes a tri-level parallel corpus incorporating modern explanatory glosses as explicit semantic bridges and operationalizes translation strategies through paired conserved versus substituted CSI variants categorized by cultural translatability. Systematic evaluations of nine leading LLMs across bidirectional translation directions reveal that leveraging modern Chinese glosses - especially human-curated ones - dramatically elevates both lexical fidelity and semantic adequacy. Furthermore, multidimensional human preference audits by native experts unveil distinct cross-cultural preferences between Sinographic-tradition and non-Sinographic language groups.

0.17

RefineSplat: Reconstruction-Guided Object-Centric Gaussian Splatting for Open-Vocabulary Querying

cs.CV ·2026-11-07 ·Yiren Lu, Disheng Liu, Sreya Srinidhi, Yunlai Zhou, Yiran Qiao +5 more
benchmark
Abstract

Open-vocabulary 3D scene understanding aims to segment and retrieve arbitrary objects in 3D scenes using natural language queries. Most existing methods typically follow a splat-then-segment paradigm, where 3D Gaussian Splatting is first reconstructed and then augmented with labels or language features for subsequent segmentation. However, this pipeline often suffers from view inconsistency, ambiguous object boundaries, and weak object-level correspondence, especially in complex static and dynamic scenes. Building on our prior Segment-then-Splat framework, we present RefineSplat, a reconstruction-guided extension that retains its object-centric paradigm: it first tracks objects in 2D and then reconstructs each object with a dedicated set of 3D Gaussians, preserving explicit object-Gaussian correspondence for unified open-vocabulary segmentation and retrieval. To improve robustness, we further close the loop between 2D tracking and 3D reconstruction. Since off-the-shelf video trackers can produce unreliable masks that propagate errors into object reconstruction, we introduce a reconstruction-guided feedback mechanism that refines object tracking using rendered masks and multi-view consistency. Moreover, instead of assigning each object a single language embedding, we propose a viewpoint-aware embedding assignment strategy, where each object maintains multiple view-clustered embeddings to capture view-dependent semantic cues. We also introduce Blender-OVS, a synthetic benchmark of static and dynamic sequences with exact object meshes and point clouds, enabling direct evaluation of extracted geometry. Experiments on static and dynamic benchmarks show improved rendered-view segmentation and lower object-level Chamfer distances, with competitive scene reconstruction quality.

0.17

Enhancing Mathematical Reasoning in Turkish Large Language Models with Search-Based Inference and Model Merging

cs.CL cs.AI ·2026-11-07 ·Ahmed Zeer, Osama Shbib, Muzaffer Kaan Yuce, Himmet Toprak Kesgin, Mehmet Fatih Amasyali
Abstract

We improve Turkish mathematical reasoning in Large Language Models by combining LoRA fine-tuning, weight-space model merging, and search-based inference. Using Turkish-Llama 8B-DPO with Turkish translations of MetaMath, NuminaMath, and TigerLab, we find data quality beats scale (Mix-30k best among fine-tuned; large MetaMath underperforms). Linear merging delivers the largest gains (GSM8K up to 0.63), while Ties is modest and Breadcrumbs/Slerp degrade. Search with MCTS and Tree of Thoughts improves mean LLM-judge scores for reasoning quality on a 0-100 scale. Limitations include machine translation, limited scale, and search stochasticity.

0.17

Test-Time Scaling for Diffusion Language Models with Frequency-Aware Remasking

cs.CL cs.AI ·2026-11-07 ·Bowen Zuo, Yue Yu, Dongruo Zhou, Yinglun Zhu
Abstract

Diffusion language models (DLMs) generate text through iterative denoising, where decisions about which tokens to commit or remask strongly affect output quality. Existing certainty-aware strategies rely on instantaneous confidence or entropy, but these local signals can be misleading on difficult reasoning problems. We formulate DLM test-time scaling as adaptive allocation of both inference compute and decoding strategies. Our two-stage approach uses a fast base sampler to solve easy questions, then applies frequency-aware remasking to the remaining hard questions. This strategy tracks how often each token's prediction changes across denoising steps and preferentially commits frequently changing tokens, encouraging samples to explore alternative reasoning paths rather than collapse onto the model's preferred trajectory. Experiments with LLaDA-8B-Instruct and Dream-7B-Instruct show consistent improvements over a strong adaptive allocation baseline, with absolute accuracy gains of up to 3.73 percentage points (LLaDA) and 7.33 percentage points (Dream) on MATH-500, and absolute coverage gains of up to 11.39 percentage points (LLaDA) and 10.98 percentage points (Dream) on HumanEval. On questions unsolved by the base certainty-aware strategy after 100 samples, frequency-aware remasking recovers solutions for 3.75% to 18.75% of examples.

0.17

Who Remembers What? Multi-Granularity Memory Extraction for Multi-User, Multi-Agent Interaction

cs.CL cs.AIcs.LGcs.MA ·2026-11-07 ·Yunzhe Wang, Mesut Erhan Unal, Sheng Zhang, Shabnam Ghadar, Peng Tang
benchmark
Abstract

Self-evolving agents improve their own skills, memory, and harness code from experience without retraining the base LLM, and group-evolving agents go further by sharing these lessons across agents. Yet most such systems overlook that sharing across users and heterogeneous agents requires deciding who can benefit from each lesson and what information must remain private. We present MixUA, a memory system that distills independent user-agent trajectories into lessons and assigns each lesson to a user-agent pair, user, agent, or global scope according to whom it benefits and what it may disclose. For evaluation, we introduce CrossLearnBench, a mixed-user-agent benchmark that pairs established agent tasks with user preferences and privacy probes that test whether private facts leak into memory other users can read. With five users and seven agents, MixUA raises pass rates by up to 16 points over no memory and 5 over a single global memory, even on tasks previously tried only by other users on other agents. It retains up to 90% of previously stated preferences, and at most 0.02% of privacy probes find a planted private fact in a forbidden scope. We hope MixUA offers a scalable path toward a global system where users and agents co-evolve.

0.17

ESTRA: Enabling Extreme Attention Sparsity for Training-Free Video Diffusion Acceleration

cs.CV ·2026-11-07 ·Enhuai Liu, Yunke Wang, Yutong Wang, Changming Sun
Abstract

Video diffusion transformers generate high-quality videos, but attention over long spatiotemporal sequences makes inference computationally expensive. Sparse attention reduces this cost by computing only a subset of query-key interactions exactly, and recent training-free methods further approximate the remaining interactions. These methods, however, degrade quickly at high sparsity, which limits the achievable speedup. We present ESTRA (Extreme Sparsity Through Refined Attention), a training-free acceleration method that preserves generation quality even at extreme sparsity. ESTRA concentrates exact attention computation on a small set of important key-value tokens and approximates the remaining contributions with cluster representatives. To identify these tokens efficiently, it groups similar queries and uses their representatives to score individual keys. A variance-based correction further refines the attention weights of the cluster representatives, improving the accuracy of the approximation. Experiments on Wan2.1-14B and HunyuanVideo-13B show that ESTRA establishes a new quality-efficiency frontier for training-free sparse attention, achieving $3.26\times$ and $4.71\times$ end-to-end speedups at 5% attention density while surpassing Sol-Attn and SVG-EAR in fidelity to dense generation and on most VBench dimensions. Even at 1% density, ESTRA remains more faithful to dense generation than Sol-Attn at 12.5%, while reaching up to $5.39\times$ speedup.

0.16

Guided Co-Attention with Adaptive Gating for Multimodal Medical Data Fusion

cs.CV cs.LG ·2026-11-07 ·Muntaqim Ahmed Raju, Priyanka Siddappa, Md Shifat Haider Al Amin, Ruizhe Ma, Md Ripon Ahammed
Abstract · IEEE BIBM 2026

Multimodal fusion requires not only exchanging information across modalities but also determining which representations should participate in the interaction and how complementary cross-modal views should be combined. We introduce Guided Co-Attention with Adaptive Gating, a fusion framework that integrates token-wise guidance, bidirectional cross-attention, and sample-specific fusion. The proposed method constructs two directional representations: an image-conditioned text representation, obtained by allowing textual tokens to query visual features, and a text-conditioned image representation formed in the reverse direction. Token-wise guidance modulates the representations before cross-attention, while an adaptive gate learns the relative contribution of the two resulting directional representations for each sample. We evaluate the approach using paired chest radiographs and radiology reports as an image-text test bed for binary (normal vs. abnormal) classification. The two cross-attention directions exhibit distinct performance characteristics, demonstrating that reversing the query and source modalities does not produce interchangeable representations. Ablation experiments further show that bidirectional interaction, adaptive gating, and token guidance provide complementary benefits. The proposed model achieves 96.05% accuracy, 96.10% F1-score, 98.37% AUC, 95.60% Sens@95Spec, and 96.51% Spec@95Sens, yielding the highest mean performance across all evaluated metrics. These findings support explicitly modeling interaction direction, token relevance, and sample-dependent integration when constructing multimodal representations.

0.16

Efficient 3D plant segmentation through multiscale graph learning and optimized geometric processing

cs.CV ·2026-11-07 ·Jesus Ivan Ruiz-Martinez, Andres Mendez-Vazquez, Eduardo Rodriguez-Tello
Abstract

Plant organ segmentation on compact computing platforms requires accurate predictions and affordable geometric preparation. We develop and evaluate a multiscale edge based graph attention model that combines local shape descriptors, two neighborhood scales and a residual stem and leaf correction in approximately 78 thousand parameters. The engineering application is plant point cloud segmentation for precision agriculture, assessed through grouped accuracy experiments and physical Raspberry Pi 5 inference. Broad comparisons with hierarchical, graph and transformer baselines establish PointNeXt as the strongest external accuracy reference among the tested local recipes. A nested four variant ablation on a local maize collection separates geometric information from the combined multiscale configuration across seven outer groups and three seeds. The complete model achieves the highest mean intersection over union, 86.36%, although uncertainty and boundary errors qualify the contribution of each component. We then optimize full cloud principal component analysis, sampling and neighbor queries without changing learned weights or descriptor definitions. All 1,180 paired desktop comparisons on PlantNet and TomatoWUR preserve exact logits and predictions. Multiscale pipeline latency falls from 330.33 to 96.82 milliseconds on PlantNet and from 1,054.52 to 463.10 milliseconds on TomatoWUR. PointNeXt also benefits from optimized sampling and remains faster over the complete pipeline. These results support compact graph learning with reduced geometric overhead, while physical board measurements establish inference feasibility rather than energy efficiency or readiness for autonomous flight.

0.15

LongCapHarness: Capability-Guided Auto-Harnessing for Long-Context Reasoning

cs.CL ·2026-11-07 ·Tao Zhang, Yuhang He
Abstract

Existing methods enhancing an LLM's long-context reasoning capability often rely on data synthesis and follow-up LLM post-training. However, the data synthesis strategy and the post-training recipes are empirically driven and computationally intensive. In this work, we propose a new framework, dubbed LongCapHarness, to instead harness the underlying capabilities for long-context reasoning. The harnessing process recursively revises and updates capability primitive implementation, capability state and control logic through recursive self-improvement on the development set. With neither data synthesis nor LLM post-training, LongCapHarness is applied to Llama3-8B and transferred to Qwen2.5-3B. LongCapHarness substantially outperforms recent data-synthesis-based baselines on HELMET and RULER benchmarks; its estimated development cost is less than 50% of the estimated training costs of NExtLong and PolicyLong. We hope these results motivate further research into harnessing the latent capabilities of LLMs to solve tasks beyond their explicit training.

0.14

PrivMark: Pixel-Space Steering against Privacy Leakage in Vision-Language Models

cs.CV ·2026-11-07 ·Ruofan Xing, Arman Akbari, Weikai Lin, Yuhao Zhu, Xuan Zhang
Abstract · 33 pages, 12 figures

Vision-language models (VLMs) routinely disclose soft-biometric attributes such as age, gender, and race in their text outputs, exposing demographic information from uploaded images to any downstream API user. Existing privacy defenses do not fit this setting: face de-identification leaves distributed soft-biometric cues intact, classifier-targeted adversarial perturbations do not transfer to open-ended generation, and stronger VLM-specific methods require retraining or runtime activation access unavailable to downstream APIs. We propose PrivMark, a framework that suppresses biometric leakage at the image level. By analyzing VLM activations, we identify two weakly aligned steering vector directions in hidden-state space, each tied to a distinct leakage behavior --- one corresponding to incidental attribute disclosure under open-ended descriptions, the other to compliance with direct biometric probes instead of refusal --- and encode both directions into a single bounded pixel perturbation, so the protected image alone enforces privacy with no invasive access to model internals at deployment. To support evaluation, we assemble BioleakBench, $1{,}600$ VQA pairs covering biometric privacy attributes under both free-form description and direct-probe protocols. Across three white-box source models, PrivMark achieves up to $85.9\%$ biometric suppression while retaining $75\%$ of clean OK-VQA utility, up to $1.4\times$ the privacy-utility tradeoff of the strongest pixel-only baseline. The same one-shot perturbations transfer consistently to seven unseen VLMs including closed-source APIs.

0.14

Physically Guided TeX Generation for Low-Light Visual Tasks

cs.CV ·2026-11-07 ·Daoheng Li, Jianjun Qian, Mengkai Yan, Jian Yang
Abstract

Low-light conditions reduce visible contrast and obscure details, while infrared images lack color and texture, motivating the fusion of infrared and visible information for visual tasks. Existing fusion methods primarily optimize image appearance or task-specific features without explicitly disentangling the coupled factors underlying thermal observations, limiting their physical interpretation under thermal degeneracy. Hyperspectral thermal measurements enable physical decomposition, but they require specialized imaging hardware, while pixel-level physical annotations for supervision or evaluation incur additional costs. We propose TeX-Gen, a physically guided framework for TeX generation from paired visible and infrared images in low-light visual tasks. TeX Generation constructs complementary multi-view features and employs independent branch-specific gates to route cues into the Temperature, Emissivity, and Texture maps. TeXMetric integrates radiometric reconstruction, channel structure, and material consistency to assess the generated maps without requiring pixel-level physical labels. TeX Tune improves performance through task-guided adjustments, jointly optimizing the adjustment modules and task network while keeping the generator frozen. Experiments show that TeX-Gen achieves competitive task performance in low-light environments on the FLIR, LLVIP, M3FD, and MSRS datasets. Experiments on HADAR also examine the association between TeXMetric scores and downstream detection performance.

0.14

AI in the Clinical Documentation Loop: Effects of LLM-Induced Note Bloat and Diagnostic Overcall on Downstream LLMs

cs.CL cs.AI ·2026-11-07 ·Jinghui Liu, Anthony Nguyen, Taylan Gurgenci
Abstract · ML4H 2026 - Findings

Large language models (LLMs) are increasingly used to process medical records for healthcare tasks such as clinical documentation. Because these tools also generate documentation, the relationship is now bidirectional: models both read from and write to the record. Understanding how LLMs behave when reading records that already contain LLM-authored content is therefore critical for their safe deployment. In this study, we examine how LLM-authored content in the medical record affects downstream LLMs that read the same record. We consider two types of LLM-induced artifacts: bloat, in which LLM-generated summaries accumulate in the record and change the length and provenance of its content over time; and distortion, in which LLMs subtly alter clinical findings rather than fabricating facts outright. As a case of distortion, we study diagnostic overcall (documenting an unconfirmed differential diagnosis in the record) and test whether it propagates to the reader LLM's output. Across seven reader models from four families, we find evidence of both effects. First, LLM-generated summary notes modestly but consistently degraded downstream hospital-course generation compared with copy-forward and templated text of matched length. Second, injected overcalls reappeared in downstream outputs at rates that varied by reader model, without detectable changes in standard quality metrics. These findings show the need to monitor AI-generated content in medical records and to account for it when those records are later processed by AI.

0.13

Tiny Evidence Region Localization with Selective Contrastive On-Policy Self-Distillation

cs.CV cs.AI ·2026-11-07 ·Yujuan Pang, Wenhao Xu, Yinglei Teng
Abstract

Multimodal large language models have made progress in general visual tasks. However, limitations remain in fine-grained visual tasks requiring evidence from minute image regions. Existing methods primarily train models using evidence-related regions while overlooking the influence of extensive background regions, resulting in a limited ability to resist interference from irrelevant regions. We propose selective positive-negative on-policy self-distillation (SPN-OPSD), which considers both evidence and background regions. Supervision from both views improves the model's ability to resist interference from irrelevant regions. Furthermore, selective token-level distillation provides more informative supervision, improving the model's ability to attend to evidence regions in Fine-Grained Visual Tasks. On Qwen3.5-2B, SPN-OPSD improves accuracy on V* Bench, ZoomBench, and MME-RealWorld-Lite from 76.44%, 45.92%, and 57.43% to 84.82%, 54.56%, and 59.56%, respectively. Similar gains are also observed on Qwen3.5-0.8B.

0.12

Reason First, Retrieve What Matters: Hypothesis-Guided Keyframe Selection for Long Video Understanding

cs.CV cs.AI ·2026-11-07 ·Yisong Hu, Junbo Wang, Yuke Li, Jiangbin Zheng
Abstract · 14 pages, 4 figures, 3 tables

Long video understanding requires models to identify and reason over sparse, temporally distributed evidence within limited frame budgets. Existing keyframe selection methods typically retrieve query-relevant frames from a densely sampled candidate set before inference. However, global image-text matching incurs unnecessary computation when a small set of uniformly sampled frames already provides sufficient evidence. Query relevance alone may also overlook evidence that distinguishes competing answers, while one-shot selection offers no opportunity to correct earlier selection errors. We propose HypoKFS, a training-free, plug-and-play framework that couples keyframe selection with iterative inference by multimodal large language models (MLLMs). HypoKFS first reasons over sparsely sampled frames to construct an initial hypothesis space. The resulting answer distribution determines whether additional frame selection is needed. When needed, the retained hypotheses guide the selection of discriminative frames, followed by renewed inference and hypothesis refinement. This iterative process directs additional visual computation toward unresolved hypotheses while avoiding unnecessary global frame matching. Experiments on Video-MME, LongVideoBench, and MLVU demonstrate consistent improvements across multiple MLLM backbones. HypoKFS improves accuracy by up to 6.7 percentage points and achieves 72.9% accuracy on Video-MME at only 1.3% of the reference frame-hypothesis matching cost.

0.12

Latent-WAM++: Shaping World and Action Representations for End-to-End Autonomous Driving

cs.CV ·2026-11-07 ·Yichen Zhang, Yupeng Zheng, Linbo Wang, Pengxuan Yang, Zebin Xing +8 more
Abstract · 16 pages, including supplementary material

We introduce Latent-WAM++, a representation-centered World Action Model (WAM) for end-to-end autonomous driving. Existing WAMs are primarily trained to predict future states or final actions, leaving the intermediate world and action representations that connect prediction to planning only indirectly constrained. Latent-WAM++ matches the supervision of each representation to its role in planning. World Representation Construction combines scene-ego future prediction, SIGReg-based distribution stabilization, and frozen geometric priors to learn a temporally predictive, non-degenerate, and geometry-aware world representation without an EMA target encoder. Trajectory MeanFlow Planner directly predicts multimodal trajectory proposals while being optimized with an Improved MeanFlow velocity-space objective, enabling one-step sampling. Visual-Action Representation Alignment further aligns the resulting pre-output action features with a stop-gradient target derived from current visual semantics. Latent-WAM++ achieves state-of-the-art performance on NAVSIM v2 and under zero-shot evaluation on HUGSIM, reaching 90.1 EPDMS on NAVSIM v2 and 47.0 average RC and 30.1 average HD-Score on HUGSIM. Training the planner end to end costs 190 GPU-hours for Latent-WAM++ versus 1,600 for Latent-WAM, a reduction of 88.1%.

0.12

Disentangled Stylized Image Generation via IP-Cross-Attention Rearrangement

cs.CV ·2026-11-07 ·Beibei Dong, Yueming Lyu, Yue Jiang, Jing Dong
Abstract · 12 pages, 11 figures

With a solid theoretical foundation and pretraining on large-scale datasets, diffusion-based methods have achieved controllable generation leveraging text and image conditions, and generating excellent stylized images is possible using diffusion models. However, previous methods face significant challenges. First, because diffusion generation is iterative, the common approach of learning implicit style representations via cycle-consistency strategies is too costly and not well-suited to diffusion, even for LoRA-based methods. Second, decoupling style is difficult in the latent space of diffusion models. Representations of attributes are coupled in the latent space of diffusion models; thus, extracting style attributes with diffusion models usually suffers from the issue of content leakage. Furthermore, because style is hard to represent in natural language, capturing style using text prompts or textual inversion is deficient. In this paper, we propose a method for achieving disentangled stylized image generation through IP-cross-attention rearrangement, based on our experimental observations. We also propose negative content guidance, inspired by classifier-free guidance, to reduce content leakage in stylized image generation.

0.11

Dynamics-Informed Action Learning for Goal-Conditioned Latent Planning

cs.CV cs.RO ·2026-11-07 ·Yinuo Jiang, Guangzhi Wang, Yiran Wang, Junfu Pu, Cheng Cheng +3 more
Abstract

Latent world models enable visual control by planning action sequences in compact representation spaces, but a gap remains between how latent dynamics are learned and how they are used for planning. First, teacher-forced training does not expose the dynamics model to the recursively predicted latent states encountered during planning. Second, sampling-based planners typically initialize action search from goal-agnostic proposals, without exploiting the action-conditioned dynamics learned by the world model. We introduce Dynamics-Informed Action Learning (DIAL) to bridge these two gaps. During training, rollout alignment exposes the forward model to its own autoregressive latent predictions, while multi-step inverse supervision trains a lightweight inverse dynamics head to infer actions from latent-state transitions. We further apply the inverse head to model-generated transitions with its parameters held fixed, encouraging predicted rollouts to retain action-relevant information. At test time, DIAL autoregressively couples the inverse and forward dynamics to construct a dynamics-informed action sequence toward the goal, which initializes the mean of the Cross-Entropy Method (CEM) proposal distribution. The subsequent CEM search then refines this initialization using the standard sampling and updating procedure. Experiments across four visual control benchmarks demonstrate that DIAL outperforms existing latent world models and planning baselines in both standard and long-horizon scenarios.

0.09

DocFS: File Systems as Agentic Reasoning Environments for Complex Multimodal Document Question Answering

cs.IR ·2026-11-07 ·Bingdian Zhang, Zixuan Li, Long Bai, Xiaolong Jin, Jiafeng Guo +1 more
Abstract

VLM-based agents solve complex tasks by iteratively interacting with environments, and their effectiveness depends heavily on how the environment exposes information and provides tools for acting on it. However, constructing an effective environment for complex multimodal document question answering remains challenging: documents usually span many pages, with multimodal contents distributed across hierarchical sections and linked by cross-modal associations, while existing methods organize documents into structured representations that address only the information side of the environment and lack tools aligned with the structure, limiting agents' ability to fully exploit it. We observe that file systems naturally combine structure organization with aligned tools: directories and files encode document structure, while file operations provide rich tools to access file contents. Building on this insight, we propose DocFS, a file-system-based agentic framework that reorganizes raw documents as file systems, where directories mirror the section hierarchy and explicit links connect text with figures, tables, and equations. Based on well-established file system tools such as navigation, search, and reading, agents can iteratively explore the document structure, locate relevant evidence, and understand fine-grained local information. Paired with Qwen3.5-Plus, DocFS achieves state-of-the-art performance on MMLongBench-Doc and FinRAGBench-V, outperforming the strongest baselines by 2.9 and 4.3 percentage points, respectively, and is up to 10.3x faster with up to 10.1x fewer tokens than the strongest baseline during reasoning.

0.08

Interactive Person Retrieval with On-the-Fly Visual Feedback

cs.CV cs.AI ·2026-11-07 ·Lyucheng Qian, John Yuehan Zhang, Pingyu Wang
dataset
Abstract

Traditional text-based person retrieval (TPR) assumes that a witness provides a complete description at once, whereas interactive person retrieval (Inter-PR) refines a partial description through question-answer exchanges. In this setting, the questioner inspects retrieved candidates to decide what evidence to request next, while the answerer sees only the witness-side memory and cannot report candidate-specific visual differences. Existing methods mainly learn offline question orderings, leaving this visual-feedback loop underexplored. To address this limitation, we introduce visual-feedback interactive person retrieval (VFI-PR), a new interactive retrieval paradigm in which retrieved candidates produce visual feedback that becomes evidence for subsequent questioning and re-ranking. To support training and evaluation in this setting, we construct VFI-PEDES, a dataset and evaluation protocol for visual-feedback interaction, and propose VFI-ReID, a framework that learns to use visual feedback during interactive person retrieval. Through the Dynamic Accumulated-evidence Retriever (DAR) and retrieval-aware question optimization, VFI-ReID adapts retrieval and question generation to evolving evidence states, turning visual feedback into complementary evidence for subsequent retrieval. Experiments show that VFI-ReID consistently improves retrieval across VFI-PR interaction stages, demonstrating the utility of visual feedback for interactive person retrieval and outperforming existing Inter-PR baselines.

0.07

VisProve: Learning Typed Visual Provenance for Scientific Document Question Answering

cs.CV cs.AI ·2026-11-07 ·Jingxuan Wei, Zhengyang Wang, Zheng Guo, Xiaoming Wu, Fuqiang Wang +4 more
dataset
Abstract

Visual citations show where an answer is supported, but not how the cited content contributes to it. In scientific document question answering, a model may read a value directly, compress several observations, or combine evidence to infer a conclusion. We study typed visual provenance, where a system generates an answer together with source regions and explicit support relations. We introduce ReViE, a dataset of 3,941 questions over complete scientific documents, with page, section, and region annotations labeled as Quotation, Compression, or Inference. We further propose VisProve, which augments joint answer and provenance generation with differentiable Spatial Evidence Reading and Evidence-Answer Relation Feedback. VisProve conditions next-token prediction on localized visual evidence and its estimated support relation through a gated norm-bounded residual. On the document-disjoint ReViE test set, VisProve improves Typed-F1 from 3.52 to 5.48 and Section-F1 from 45.99 to 53.24 over joint supervised fine-tuning with the same backbone, while Page-F1 changes only marginally. These results suggest that coupling visual grounding with support relation modeling mainly improves fine-grained provenance, while the low absolute Typed-F1 reveals a substantial gap between answer generation and precise visual evidence attribution.

0.07

Security & Privacy 15

Governed Self-Improvement for a Local-First Security Analyzer: Baseline Measurements and a Preregistered Evaluation Protocol for Tantular Cyber

cs.CR ·2026-11-07 ·Raditio Ghifiardi, Rifny Rachman
benchmarkcode/data
Abstract · 7 pages, 2 figures, 5 tables. Code and benchmark: https://github.com/ghifiardi/tantular-cyber

Local-first security tools let organizations analyze code without sending it to third-party services, but the small language models they can run locally are weaker than hosted ones and are hard to improve without breaching the privacy guarantee that motivates them. We present Tantular Cyber, an open-source, read-only repository security analyzer that keeps deterministic scanner findings authoritative and confines a locally served language model to advisory explanations in Bahasa Indonesia. We report its measured v0.3.0 baseline on a synthetic, ground-truth benchmark: combining Semgrep with an interprocedural call-graph scanner yields precision 0.843 and recall 0.915 (Semgrep alone: 0.818 and 0.766). The advisory layer is the weak point: 57% of explanations met an evidence-grounding check, against 95% for a different 8B model under identical prompts. A forensic re-scoring of retained outputs shows that every grounding failure cites a code identifier absent from the evidence shown, and that a prompt change which appeared to raise grounding from 0.56 to 0.72 did so partly by citing nothing. We document two evaluation-validity hazards we encountered: silent model-file migration during a run and non-reproducible published figures. Building on these lessons and on constraint-aware evaluation from security operations research, we propose a governed self-improvement loop (code-free feedback, curated data, a cheapest-first intervention ladder from prompts to LoRA and weight-space edits, hard gates, and human release approval) and preregister the protocol by which its first cycle will be judged. We report no results for the loop itself.

0.39

Validating AI-Discovered Cloud Threats on Live Infrastructure: A Multi-Model, Multi-Cloud Study

cs.CR ·2026-11-07 ·Nicholas Pecka, Lotfi Ben Othmane, Renee Bryce
Abstract · 8 pages, 3 figures, CSTA 2026

Cloud infrastructure threat modeling lacks empirical validation of AI-discovered threats. We present a three-phase methodology of multi-agent threat discovery, automated exploit synthesis, and non-destructive live validation, and apply it to three commercial foundation models on two cloud architectures, validating each threat natively on the cloud (AWS, GCP, or Azure) it was discovered against. Of 233 AI-discovered threats, 103 were confirmed and 34 refuted (75.2\% precision, 95\% CI [67.3, 81.7]); 44 were conditional (the named mechanism refuted but an equivalent impact shown through another path) and 52 indeterminate. Multi-model discovery helped through breadth, not agreement: the three models together found 2.2$\times$ the confirmed threats of the best single model at comparable precision, while threats reported by several models were confirmed no more often than single-model ones. False positives were not architecture-specific: they asserted access or system state the deployment did not have. Threats coded in advance as observable configuration resolved cleanly more often than multi-step behavioral threats (72.5\% vs.\ 51.6\%), which were more often conditional, suggesting that models more often misidentify the exploitation mechanism than the underlying weakness. AI-assisted discovery is a practical input to security workflows when paired with empirical validation.

0.32

Design and Evaluation of Secure Distributed Chaotic Image Encryption on ESP32-S3

cs.CR ·2026-11-07 ·Mohammed Almozaini, Fazal Noor, Sohail Alhazmi
unusual
Abstract · 12 pages, 10 figures, 5 tables. Corrected spelling of an author's name; research content unchanged

The limited computational resources of Internet of Things (IoT) devices present challenges for secure image processing. This paper presents a distributed chaotic image encryption system using independent Wi-Fi-connected ESP32-S3 microcontrollers. A master-worker architecture distributes image blocks through master-coordinated chaotic sequence management, maintaining one logically continuous chaotic sequence by assigning block-specific starting states independently of worker assignments. The implemented system uses reversible chaotic substitution (SUBSTITUTE), while Elliptic Curve Digital Signature Algorithm (ECDSA)-signed discovery messages and Transport Layer Security (TLS) 1.3 with mutual authentication secure device communication. Experimental evaluation using one, two, and four workers demonstrates a reduction in execution time from 43.215 to 15.878 seconds, achieving a speed-up of 2.72 times. The encrypted image achieves a Shannon entropy of 7.999940 bits/byte and near-zero adjacent-pixel correlation. Exact image recovery is verified through matching Secure Hash Algorithm 256-bit (SHA-256) hashes and a Structural Similarity Index (SSIM) value of 1.0. Additional experiments demonstrate successful encryption and decryption using different worker counts and block sizes, automatic task reassignment following worker failure, and rejection of a mismatched private key during TLS authentication. These findings demonstrate the feasibility of secure, flexible, and distributed chaotic image processing across independent resource-constrained IoT devices.

0.31

An Agentic Multi-Agent Framework for System Log Anomaly Detection

cs.CR ·2026-11-07 ·Ngoc-Son Le, Huy-Hoang Mac Nguyen, Ngoc Hong Tran
Abstract · 14th International Conference on Multidisciplinary Research (ICMR 2026), ISBN 978-974-474-120-2, Thaksin University Publisher, Bangkok, Thai…

As enterprise systems grow in scale and complexity, manual log analysis and conventional SIEM tools struggle to keep pace with the volume and diversity of incidents. To address this challenge, this paper proposed a multi-agent AI framework supporting in analyzing the system log to trace the attack forensic in the system more effectively and efficiently. The framework is designed using n8n platform and Gemini large language model. The framework consists of 7 AI agent layers. Each agent work on a specific task. When the task is done, the agent will forward the result to the output-associated agent of the next layer. The multi agent model improves the analysis velocity, so it is able to support numerous log analysis requirement at the same time. We deployed 16 test cases to test operation of the multi agent workflow, and evaluate the accuracy per test case and the operative performance. The result findings are very promising.

0.31

Agents of Chaos: A Systematization of Knowledge (SoK) on Malicious Attacks on Agentic AI Systems

cs.CR ·2026-11-07 ·Thanh M. Truong-Le, Amber Bennoui, Hoang-Ly Nguyen, Gia-Thang Ho, Thai-Minh Truong +1 more
publishedsurvey
Abstract · Accepted at the Workshop on Agentic AI in Offensive and Defensive Cyber Operations (AIDC 2026), co-located with ACSAC 2026

Agentic AI systems extend large language models with tools, memory, retrieval, browser or computer-use interfaces, code execution, external services, and multi-agent coordination. Although these capabilities enable agents to act beyond text generation and perform tasks, they also create attack paths through which adversaries can exploit untrusted content, deceptive metadata, poisoned state, compromised skills, mutable dependencies, or malicious code. Such attacks can manipulate planning and tool use, corrupt memory, abuse code-execution environments, and trigger harmful downstream effects. In this paper, we present a systematic review of malicious attacks targeting agentic AI systems. We screened 186 candidate sources and retained 98. Following independent coding and resolver adjudication, we identified 284 atomic attacks, 452 atomic safeguards, and 660 explicit attack--safeguard relationships. We organize attacks into eight causal-mechanism classes while retaining insertion points, activation conditions, abused capabilities, objectives, impacts, persistence, and evidence as orthogonal dimensions. Our analysis shows that tool, MCP, and action-layer attacks account for the largest share of atomic attacks, while the prompt-injection and action-hijacking class has the broadest paper coverage. We identify distinct class-specific activation patterns and separation between insertion and later activation. Safeguard coverage remains uneven: only 278 of 660 relationships are directly assessed, and strong coverage is uncommon. These findings support lifecycle-aware threat modeling, agent evaluation, tool governance, and defense-in-depth security for agentic AI systems.

0.30

When Experience Becomes a Backdoor: Poisoning Self-Evolving LLM Agents via Failure-Aware Task Optimization

cs.CR cs.AI ·2026-11-07 ·Minhua Lin, Zongyu Wu, Hanqing Lu, Xianfeng Tang, Xiang Zhang +2 more
code/data
Abstract

Large language model (LLM) agents increasingly improve themselves through self-evolution, distilling execution experience into persistent, reusable skills. However, this persistence creates a new attack surface: malicious rules implanted into skills can compromise subsequent agent behavior. Existing skill backdoor attacks typically assume direct manipulation of the victim's skills, which is impractical when skills are constructed internally by a trusted evolver. In this paper, we study a novel problem of task-level backdoor attacks, where the attacker can only inject poisoned evolution tasks, inducing the agent to implant backdoors into its own skills. Since the attacker controls neither skill induction nor downstream execution, a natural approach is to iteratively optimize poisoned tasks based on attack success. However, our preliminary analysis reveals two key challenges: (i) failures at different stages of skill evolution and downstream execution can produce indistinguishable outcomes, providing little guidance for targeted poisoned-task updates; and (ii) unsuccessful updates may undermine earlier progress, while simply discarding them may lead to repeated ineffective updates. To address these challenges, we propose BadEvolve, a failure-aware task-poisoning framework against self-evolving LLM agents. BadEvolve diagnoses attack failures across four stages, i.e., Induction, Activation, Following, and Effect, and directs updates toward the failing stage. It further employs rollback to preserve earlier progress and failure memory to avoid repeating ineffective updates. Extensive experiments across diverse self-evolution frameworks, benchmarks, and backbone LLMs demonstrate the effectiveness of BadEvolve. Our code is available at https://github.com/ventr1c/BadEvolve.

0.27

The Devil Wears Markdown: Detecting Malicious Intent Hidden in AI Agent Skill Instructions through Hybrid Static, Dynamic, and Agentic Analysis

cs.CR ·2026-11-07 ·Thanh M. Truong-Le, Amber Bennoui, Hoang-Ly Nguyen, Gia-Thang Ho, Thai-Minh Truong +1 more
publishedcode/data
Abstract · Accepted at the Workshop on Agentic AI in Offensive and Defensive Cyber Operations (AIDC 2026), co-located with ACSAC 2026

Agentic AI ecosystems increasingly rely on reusable agent skills that combine executable code, natural-language instructions, configuration artifacts, and external tool integrations. These hybrid artifacts create software supply-chain risks because malicious intent may be distributed across instructions, code, metadata, and runtime behavior, making isolated alerts difficult to interpret. We evaluate three open-source scanners on a held-out test set of 140 skills comprising 80 malicious and 60 benign samples, including Cisco Skill Scanner with and without its optional LLM analyzer. The default Cisco configuration achieves a recall of 0.91 with 33 false positives among the 60 benign skills, while the LLM-enabled configuration using Qwen3:8B achieves a recall of 0.85 with 40 false positives. We then present a framework that combines static findings, safety-constrained sandbox evidence, and LLM-based reasoning to assess malicious intent. The framework tracks the origins of findings and relates evidence across analysis layers. In a within-system evidence-source ablation using the same reasoning backend, adding dynamic evidence to skill-content reasoning increases F1 from 0.70 to 0.85. Combining static and dynamic evidence further increases F1 to 0.89, with five false positives among 60 benign skills. Dynamic evidence provides the largest observed improvement. The Full Pipeline detects 68 of 80 malicious skills, achieving 0.88 accuracy, 0.93 precision, and 0.85 recall. It matches the observed recall of Cisco Skill Scanner with LLM while reducing false positives from 40 to five. This system-level comparison does not isolate model or architecture effects.

0.26

A Case Study of a Phishing Cluster Targeting German Universities

cs.CR ·2026-11-07 ·Raphael Springer, Alexander Schmitz, Joel Taddey, Gerhard Senkowski, Artur Leinweber +1 more
survey
Abstract

In light of recent ransomware attacks, this study analyzes and measures the impact of cyberattacks against academic institutions in German-speaking areas. First, we provide an overview of 25 confirmed cyberattacks against universities in German-speaking areas between February 2022 and February 2023. In this time frame, we find that 5% of all German universities have experienced a confirmed cyberattack with a measurable impact. Second, we document victimology and Tactics, Techniques, and Procedures (TTP) of a phishing activity cluster spanning more than three years and showing a particular focus on universities in German-speaking areas. Of the 93 targeted universities, 58 (62%) are German, representing a significant fraction with approximately 14% of the total number of universities (423) in Germany. A notable technique is that the phishing emails are sent via university infrastructure, using valid, compromised credentials of university-associated personnel and involving university Virtual Private Network (VPN) tunnels.

0.26

The Limits of Information-Flow Control Enforcement for Agentic Systems

cs.CR cs.AI ·2026-11-07 ·Jonathan Gregory, Kyle Domico, Kunyang Li, Jean-Charles Noirot Ferrand, Patrick McDaniel
Abstract

Information-flow control (IFC) is a promising approach to enforce policy in agentic AI systems. However, IFC suffers from label creep, a phenomenon where monotonically increasing security labels propagate through a system and progressively limit its utility. Label creep in agentic AI systems can be mitigated using two major strategies: (1) improved system design and (2) label declassification. In this paper, we characterize how far system design alone can mitigate label creep in agentic systems. Specifically, we measure how system topology, scale, and agent composition affect utility loss and security enforcement cost proxies for conservative IFC enforcement in a stateful agentic system simulator. We develop a pipeline to systematically vary these features and measure their impacts on label propagation. Our results demonstrate strong associations between structural graph features and our IFC utility loss metric (up to a Spearman correlation $|ρ|$ of $0.96$) and show that system size and tool allocation substantially affect how labels propagate. These findings identify design choices to reduce IFC utility loss, but they also demonstrate that system design alone cannot fully prevent utility loss due to monotonic label propagation. This motivates the development of label declassification mechanisms that aid these design choices and diminish IFC utility cost.

0.25

QUASI: Post-Quantum Quorum Attestation of Pre-Validated Statements

cs.CR ·2026-11-07 ·Maksym Ioffe, Dahlia Malkhi
Abstract

We present QUASI, a newly conceived class of protocols that enable a group of parties to produce a publicly verifiable attestation about a pre-validated statement. Designed for quorum-based systems, QUASI enables parties to attest to external entities that the statement is valid. Surprisingly, each party need not sign the entire statement: authenticating only a small subset of symbols suffices for the quorum to attest to the whole. A simple deterministic construction illustrates this possibility. Built purely from hashing and error correction, QUASI achieves small per-party contributions and low attestation generation and verification costs, without relying on SNARKs for aggregation. Our main contribution is a randomized construction achieving reductions of more than $10\times$ in per-party contribution size and hash computations for generation and verification, compared with full hash-based signatures. QUASI is a generic primitive supporting diverse applications. Two principal use cases are external attestation for signature-free consensus, enabling clients to verify statements about committed state; and preprocessing for SNARK-based aggregation, compounding QUASI's savings with succinct-proof compression.

0.23

CAROUSEL: GPU-Accelerated Private Vector Search via Homomorphic Sketching

cs.CR ·2026-11-07 ·Sohaib, Divyakant Agrawal, Amr El Abbadi, Soamar Homsi
Abstract

Vector search on untrusted infrastructure exposes the \emph{query embedding}, a faithful summary of the user's intent. Fully homomorphic encryption can hide the query, but existing approaches still require a full scan followed by ranking of encrypted scores. Scalable private systems avoid encrypted ranking by either sending scores from a single cluster to the client or moving the search to the client, which must first stream the entire vector index for preprocessing. We present \textsc{Carousel}, which instead scores the full corpus under encryption and compresses the results for client-side ranking. The server scores every vector under encryption, raises the scores to a power that preserves their order by magnitude while suppressing all but a handful, and compresses the resulting sparse vector into a logarithmic number of ciphertexts using homomorphic additions alone. Because every vector is scored, \textsc{Carousel} incurs no recall loss from discarding candidates. Since clients hold no database-dependent state, insertions and deletions take effect immediately at the cost of a single vector write. \textsc{Carousel} searches $1$ billion vectors of BIGANN SIFT1B across multiple GPUs and achieves recall@100 of $0.9715$. At $100$ million SIFT vectors, it achieves recall@100 of $0.9845$ with a query latency of $3.735$ seconds on a single GPU.

0.20

Towards Lightweight IoT Intrusion Detection: Feature Optimization and Explainable Machine Learning for Resource-Constrained Environment

cs.CR ·2026-11-07 ·Shaker Nawasra, Munther Abualkibash
published
Abstract

Machine learning-based intrusion detection systems (IDSs) can improve IoT security but often rely on large feature sets that increase computational overhead and limit deployment on resource-constrained devices. This paper presents a resource-efficient and explainable feature optimization framework for lightweight IoT intrusion detection using the CICIoT2023 dataset. A Random Forest model was used to rank feature importance and identify compact feature subsets. Experimental results show that a three-feature model achieved an MCC of 0.9622 while reducing the feature space by 92.1%, whereas a six feature model achieved an MCC of 0.9675 with an 84.2% reduction. Furthermore, a fifteen-feature model achieved an MCC of 0.9744, preserving 99.97% of the performance of the full 38-feature baseline model while reducing the feature space by 60.5%. SHAP-based explainability analysis validated the selected features and provided interpretable insights into model decisions. The results demonstrate that lightweight, explainable, and high-performance IoT intrusion detection can be achieved using only a small subset of carefully selected features, significantly reducing resource requirements with minimal loss of detection effectiveness.

0.20

PQC-Corpus: A Dataset of Linked Development Histories for Post-Quantum Cryptography

cs.CR cs.SE ·2026-11-07 ·Souhardya Saha Dip, Moshood A. Fakorede, Umar Farooq
datasetcode/data
Abstract · 5 pages, 4 figures, 2 tables

As post-quantum cryptography (PQC) standards evolve, developers must adapt implementations, tests, and protocols. Studying this work requires connecting code changes to problem reports and developer reviews. We present PQC-Corpus, a dataset of linked development histories from 179 public GitHub repositories, spanning cryptographic libraries and general-purpose projects. It connects issues and closing pull requests to commits and full patches, with comments and reviews. We combine Git histories with GitHub threads, reconstruct 23 pull-request commit lists truncated at 250 entries, and check file integrity, schemas, and links. Two annotators independently assessed sampled commits and issues ($κ=0.86$ and $0.93$), labeling about seven in ten commits and two in three issues PQC-related within these samples. We provide full and filtered views with removal ledgers, and quantify how filtering reduces links from issues to commits. A patch-free view supports inspection before downloading patches, while traversal scripts follow issues to code changes. We make the dataset and collector publicly available for studies of specification changes, hybrid interoperability, cryptographic review, and migration assistance.

0.19

Mirage: Provable Defense Against LLM KV Cache Timing Side Channels with Topology-Aware Isolation

cs.CR cs.AI ·2026-11-07 ·Xiaobei Yan, Tianwei Zhang
Abstract

Large Language Models (LLMs) are increasingly deployed as shared multi-tenant services, relying on advanced optimizations like KV cache and prefix sharing to improve throughput and reduce latency. However, these optimizations introduce a critical vulnerability: attackers can exploit time-to-first-token (TTFT) variations between KV cache hits and misses to infer sensitive prompt tokens via prefix-matching side-channel attacks. Existing defenses indiscriminately restrict cross-tenant cache sharing, which incurs prohibitive performance overhead and negates the fundamental efficiency benefits of the KV cache. We propose Mirage, a topology-aware emission obfuscation framework that provably bounds timing leakage without sacrificing the performance gains of cache sharing. Instead of stalling the GPU or forcing redundant recomputations, Mirage delays token emission at the gateway so the attacker observes a natural cache miss distribution, while the GPU seamlessly continues serving other requests at full speed. Mirage uses a Privacy Odometer that persistently tracks the topological novelty of prefix fan-out expansions to distinguish aggressive probing from benign traffic. Once a prefix exceeds a safety threshold, Mirage engages a two-phase policy that uses a dynamic estimator to pad the TTFT to a calibrated cache-miss baseline. On a 500-sample public corpus across three models, Mirage suppresses the timing side-channel to near the random-guessing floor, while maintaining a significantly lower TTFT than defense baselines. We further show that Mirage withstands a strong adaptive adversary and the same principle carries over to semantic caching layers.

0.18

CyberBinGym: Can AI Agents Turn Binary Patches into Real Exploits?

cs.CR cs.AIcs.LG ·2026-11-07 ·Lambang Akbar Wijayadi, Zhaoqi Xiao, Zhun Wang, Hongwei Li, Wenbo Guo +3 more
benchmark
Abstract

Security fixes usually reach users as compiled updates, but deployment delays leave unpatched systems exposed. During this window, comparing patched and unpatched binaries can reveal how to exploit a vulnerability, turning a defensive update into actionable information for attackers. Existing benchmarks do not measure this capability: they expose source code, rely on human-crafted challenges, or evaluate binary comprehension without requiring exploitation. We introduce CyberBinGym, a benchmark containing 100 tasks derived from real vulnerabilities in 53 widely used projects, spanning six vulnerability families and three levels of analysis complexity. Agents receive stripped vulnerable and patched binaries without source code or vulnerability information, and must combine static and dynamic analysis to recover and exploit the vulnerability. Evaluation separately measures vulnerability recovery through differential crash testing and end-to-end exploitation through flag retrieval, with trajectory audits to reject reward hacking. The strongest evaluated agent recovers the target vulnerability in 88% of tasks and exploits it in 20%. Providing the patched binary raises its exploit rate from 10% to 25% on a matched 20-task subset; with standard binary hardening enabled, it still exploits two of its 20 previously successful tasks. These results demonstrate that current agents can turn released binary patches into working exploits in a controlled setting, although exploitation remains unreliable. These capabilities lower the barriers to targeting unpatched software and compress defenders' effective response window, increasing the urgency of timely patch deployment.

0.13

Systems & Networking 16

Bit-level Perceptron Prediction for Indirect Branches

cs.AR ·2026-11-07 ·Elba Garza, Samira Mirbagher-Ajorpaz, Tahsin Ahmad Khan, Daniel A. Jiménez
top venuehighly cited
Abstract · Author manuscript of Proceedings of the 46th International Symposium on Computer Architecture (ISCA '19), pp. 27-38 (2019). 12 manuscript pa…

Modern software uses indirect branches for various purposes including, but not limited to, virtual method dispatch and implementation of switch statements. Because an indirect branch's target address cannot be determined prior to execution, high-performance processors depend on highly-accurate indirect branch prediction techniques to mitigate control hazards. This paper proposes a new indirect branch prediction scheme that predicts target addresses at the bit level. Using a series of perceptron-based predictors, our predictor predicts individual branch target address bits based on correlations within branch history. Our evaluations show this new branch target predictor is competitive with state-of-the-art branch target predictors at an equivalent hardware budget. For instance, over a set of workloads including SPEC and mobile applications, our predictor achieves a misprediction rate of 0.183 mispredictions per 1000 instructions, compared with 0.193 for the state-of-the-art ITTAGE predictor and 0.29 for a VPC-based indirect predictor.

0.64

Measuring Tabular Data Formats Cold: A Replicated Cold-Cache Benchmark of Serialization, Scan, and Index Performance on Real Datasets

cs.DB cs.PFcs.SE ·2026-11-07 ·Asra Fathima, Kristoffer Roberts
benchmarkcode/data
Abstract · Submitted to IEEE Open Journal of the Computer Society

Choosing a file format for tabular data is a routine engineering decision, yet published comparisons often time reads served from the operating-system page cache, use one session, or rely on a single synthetic dataset. This article benchmarks fifteen format and access-path configurations (CSV, JSON Lines, Pickle, Feather, Parquet, ORC, DuckDB, and SQLite, with several access paths) on two real public datasets, New York City taxi trips and United States airline on-time records, at 50,000 to 3 million rows. Every measurement uses an explicitly cold cache, randomized execution order, and three independent sessions (with restarts on the laptop), and is analyzed with bootstrap intervals and Holm-adjusted permutation tests. Parquet answered a filtered aggregation 2.6 to 11 times faster than CSV, while JSON Lines was 10 to 27 times slower. A secondary index made SQLite queries 2.6 to 4.8 times slower on randomly ordered data and helped only on data sorted by the filtered column, at low selectivity. A row filter inside the Parquet reader was significantly faster than a projected scan on flight data from one million rows and on both datasets at three million, and row-group skipping needs several row groups, which the default writer produced only at 3 million rows. A 4 GB Linux host ran out of memory on configurations that a 24 GB laptop completed and reversed some orderings, such as Pickle versus Parquet. Code, derived data, and raw timings are released.

0.42

Beyond Host-Side Control: HBM Controller Partitioning Across Host and HBM Base Dies

cs.AR ·2026-11-07 ·Tharunya Yogananda, Wantong Li
published
Abstract · Accepted to IEEE International 3D Systems Integration Conference (3DIC)

Recent generations of high bandwidth memory (HBM), such as HBM4, allow the base die of the memory stack to be manufactured using a logic process that can accommodate more complex controller functions next to the DRAM. We present the first systematic quantitative study of function-level controller placement between the host and the HBM4 base die on a silicon interposer. We divide the controller into its main functions and evaluate four placements, from a complete host controller to a complete base die controller, with intermediate placements that move maintenance or scheduling to the base die. We extend Ramulator 2 with HBM4 timing and maintenance, model a die-to-die (D2D) link with timing checks at the DRAM interface, and compose area, energy, and temperature from synthesized controller blocks, modeled storage, and interface data. A provisioned D2D link preserves per-stack bandwidth and adds about twice its one-way delay to unloaded latency. Maintenance moves to the base die with no bandwidth cost and frees host command slots, and the freed share grows with stack height. Replacing the wide interface with a D2D link saves 61% of host shoreline and 16.85 mm2 of host area per stack. The D2D link dominates the added base die area and power, and the base die runs cooler under streaming with link energy below 0.4 pJ/bit. Shared control delivers up to 1.72x the throughput of static partitioning between two requestors.

0.36

Firepool: Brokering Warm GPU Capacity for Agentic and RL Workloads at LinkedIn

cs.DC cs.LG ·2026-11-07 ·Ankit Goyal, Santosh Jha, Zhuo Zhi, Youmin Han, Cong Gu +2 more
Abstract · 18 pages, 7 figures, 8 tables

AI agents and reinforcement-learning (RL) trainers run thousands of short, isolated GPU jobs: tool calls, verifiers, and rollouts. A job often finishes in seconds, but starting a GPU pod for it takes 20-35 s even when the container image is already on the node, and minutes when it is not. A warm pod removes the wait only if it can serve the next job, and keeping pods warm must neither waste GPUs nor leak state between callers. Firepool, the sandbox and execution broker behind agentic and RL workloads at LinkedIn, meets both requirements. It makes pods interchangeable by building them from a shared base image per hardware profile and binding each job's code, extra dependencies, and cluster topology when the job starts. It keeps them warm cheaply and safely by reusing a pod only for the caller that used it last, staging zero-VRAM standby pods in fractional GPU shares alongside active pods, and capping the GPUs that warm pools may hold via queue-driven budgets. Over a sampled peak period in production, Firepool serves 94.7% of 3,206 pod requests from warm or standby capacity. In paired benchmarks with fixed-duration jobs on production H200 nodes, agent and RL loops finish 13x and 10.8x faster than with a fresh pod per job; against a fresh reuse tag per job, warm reuse cuts p90 start latency from 59.1 s to 3.5 s. When a job cannot reuse the previous pod, standby promotion shortens the handover wait from 51 s to 0.94 s. In local and simulated runs, a keep-alive matched to the loop's pauses keeps the latency benefit while holding about half the capacity of an always-on pool.

0.33

RTLREX: LLM-Guided Framework for Exact High-Level RTL Recovery

cs.AR cs.AI ·2026-11-07 ·Wuqian Tang, Fangzhou Liu, Ruijie Wang, TingTing Hwang, Bei Yu +1 more
Abstract · 14 pages, 6 figures, 9 tables

Hardware designs may be available only as gate-level netlists or input-output descriptions when the original register-transfer-level (RTL) design and documentation are not accessible. Low-level representations obscure word-level computation and its finite-width behavior. Recovering high-level RTL from such representations therefore helps engineers understand a circuit's functionality and enables RTL rewriting and resynthesis. A plausible function model is also insufficient unless the generated RTL is equivalent to the original circuit for all input combinations. In this paper, we present RTLREX, a large language model (LLM)-guided framework that coordinates domain-specific tools for function recovery, RTL generation, independent verification, and cost-guided rewriting. Independent verification establishes equivalence between the recovered RTL and the original circuit. The LLM uses tool feedback to select further analyses, revise candidate RTL, and propose syntax-cost reductions. Experiments on 30 benchmarks from Problem A of the ICCAD 2022 CAD Contest show that RTLREX recovers equivalent high-level RTL for all benchmarks, surpassing the state-of-the-art method WolFEx in recovery coverage. On benchmarks recovered by both methods, RTLREX reduces mean syntax cost by 20.5% relative to WolFEx. On complete truth-table benchmarks from the IWLS 2026 Contest, our framework reduces mean syntax cost by 95.7% relative to an LLM agent baseline on the benchmarks validated by both methods.

0.31

DICOMPUTE: An Economic and Privacy-Aware Framework for Distributed Artificial Intelligence Inference

cs.DC ·2026-11-07 ·Atul Khekade, Ritesh Kakkad, Anil Chinchawale, Omkar Mestry, Behnam Mohammadkhani
Abstract · 27 pages, 7 figures, 6 tables. Project websites: https://dicompute.ai and https://dicompute.si

The prevailing design of large-scale artificial intelligence inference concentrates accelerators, model weights, and request processing in centralized computing facilities. Centralization offers important benefits in accelerator utilization, high-bandwidth interconnects, operational control, and serving efficiency. Yet centralized infrastructure is not always the least-cost or most governance-compatible choice for every inference workload. This paper develops a research framework for DICOMPUTE, a heterogeneous distributed inference architecture in which policy-constrained scheduling can allocate suitable requests across privately controlled, regional, and independently operated computing resources. We derive a full-cost comparison that accounts for capital amortization, energy, network transfer, orchestration, utilization, failures, and trust-related expenses; characterize conditions under which distributed inference can outperform centralized alternatives; and specify a privacy model separating data residency, transport confidentiality, execution confidentiality, and verifiable accounting. We propose falsifiable hypotheses and a reproducible experimental design, without presenting simulated examples as measured performance. Our central conclusion is conditional: distributed inference is most promising for divisible, latency-tolerant or locality-sensitive workloads on underutilized compatible hardware, while centralized clusters retain structural advantages for tightly coupled, memory-intensive, high-throughput inference. The framework is intended to support rigorous evaluation rather than to assert universal superiority.

0.30

Augur: An Online Learning-based Hardware Data Prefetcher using Reservoir Computing

cs.AR ·2026-11-07 ·Satvik Maurya, David Trilla, Swamit Tannu, Alper Buyuktosunoglu
Abstract

Machine-learning-based hardware prefetchers can learn complex access patterns, but practical deployment demands online adaptation, low-latency inference, and modest hardware cost. We present Augur, a hardware delta prefetcher based on next-generation reservoir computing (NGRC) that meets all three requirements. Augur formulates delta prediction as a history-based regression problem. It constructs a fixed-size reservoir from hashed delta features over a short context window and predicts the next delta using a single dot product. The model is updated online via one-step SGD and requires no offline pre-training, no workload-specific tuning, and no floating-point arithmetic. A key technical insight makes this practical: replacing the integer multiplications traditionally used to generate nonlinear reservoir features with bitwise AND operations removes all feature-generation multipliers while preserving the joint context encoded in hash-derived features. Because Augur regresses a continuous delta rather than classifying into a fixed action set, it can predict any delta in [-64, 63]. Augur pairs the reservoir with a simple Markov-based delta predictor for simple delta patterns and employs multi-layered filtering and adaptive lookahead to suppress wasteful prefetches. Evaluated on 169 traces spanning GAP, Ligra, PARSEC, SPEC CPU 2006, SPEC CPU 2017, and AI workloads, Augur outperforms the state-of-the-art ML-based delta prefetcher by 2.9%, 4.5%, and 5.3% in geomean speedup on 1-, 4-, and 8-core systems, respectively, while achieving ~2x higher prefetch accuracy and 37.8% fewer overpredictions. Augur is practical to implement at 35 KB storage, 0.32 sq. mm area, and 32.7 mW in 12 nm.

0.30

QAOA Inside ADMM: Subproblem Wins Do Not Become Decomposition Wins at Matched Budgets

cs.DC quant-ph ·2026-11-07 ·Mohammad Arif Hossain, Sanim Yousuf Fahim
benchmark
Abstract

We benchmark the quantum approximate optimization algorithm (QAOA) as the binary subproblem solver in a two-block alternating direction method of multipliers (ADMM) for mixed-binary problems against classical solvers under matched budgets. With 23 optimizer evaluations per subproblem, scoring determines the comparison: on its final readout, the optimized QAOA loses to random search at $n=12$ and 16 binary variables (inconclusive at 20; exploratory and post hoc tests). When credited with its best sample, it wins. In a noiseless 12-qubit simulation, transferred-angle depth-2 QAOA finds the exact optimum of 97.3\% of planted-family subproblems, against 50.6\%--67.4\% for the four strongest local heuristics (100\% against 90.4\%--94.4\% crediting the four best bitstrings) and a not significantly different 96.5\% for semidefinite rounding with local search (SDP+LS). Under our evaluation-counting convention, its time to solution there is a factor of three to five below the local heuristics' but grows with a larger exponent; it is ten times that of SDP+LS and, in seconds, 54 times that of tabu search with restarts. For the benchmark class, an unconstrained binary quadratic problem in disguise, ADMM's binary step is a proximal step on the reduced problem: at unit penalty ADMM re-solves it, and for larger penalties exact solutions increasingly stall. On two families that do not reduce to one, less accurate solvers found the problem optimum more often at unit and larger penalties. No family gave QAOA a better ADMM outcome than the best classical solver and penalty. The benchmark includes a correctness gate and a log of its hypothesis tests.

0.27

Distributed Intelligent Ad Hoc Routing Agents for Communication at Protests During Internet Shutdowns

cs.NI ·2026-11-07 ·Matthew Schmidheiser, Milena Radenkovic
unusual
Abstract

Governments around the world are increasingly using internet shutdowns to block digital communication. Internet shutdowns are particularly dangerous during political protests, as they inhibit protesters' ability to organize and communicate critical, potentially life-saving information and may be used to prevent awareness of state-based violence. Recent prior works have explored secure, private messaging systems for use during political protests and internet shutdowns but have encountered long delivery delays or low delivery rates under challenging conditions. In this work, we explore the use of distributed, adaptive, social- and congestion-aware heuristics and efficient communication patterns to create an intelligent ad hoc routing protocol in which network agents learn and leverage the network social environment and monitor and respond to dynamic network congestion behaviour to route messages in a time- and resource-efficient manner. Through this exploration, we contribute an intelligent algorithm and protocol which overall achieves comparable or higher delivery rates relative to the leading dynamic clique routing protocol which it extends while also transmitting messages through notably fewer carriers and, when network devices have larger amounts of buffer storage, transmitting fewer message-related packets. We additionally contribute a further understanding of routing limitations and challenges and conceptualizations of extensions of intelligent dynamic clique routing to overcome them.

0.26

Performance Benchmarking of Federated Learning Frameworks for Medical Image Classification

cs.DC cs.PF ·2026-11-07 ·Vithor Pileco, Luan Mantegazine, Claudio Geyer
Abstract · 9 pages, 7 figures

Selecting a Federated Learning (FL) framework requires understanding the resource demands of its training and evaluation workflow. This work compares Flower, NVIDIA FLARE, and FedBioMed using two-dimensional MedNIST classification with ResNet-10 in a containerized, single-host environment. Following a dated review of repository activity and documented medical applications, we evaluated nine configurations per framework, with three random seeds, for a total of 81 executions. The configurations varied federation size, communication rounds, and batch size. We monitored execution time, CPU and RAM consumption, host-level GPU utilization, server network traffic rates, and model accuracy at the end of each execution. The implementations shared nominal learning settings while retaining differences in data splitting and federated evaluation. Configuration-blocked non-parametric analysis identified differences in execution time, client and server memory, GPU utilization, and network traffic rates. NVIDIA FLARE exhibited the lowest server-memory consumption and the highest host-level GPU utilization, Flower exhibited the highest average server network traffic rate, and FedBioMed required more client memory and longer execution time. No statistically significant difference in final accuracy was detected within the evaluated short training schedules; this does not establish convergence or equivalent predictive performance. The results characterize the configured workflows under shared hardware resources. They provide a concrete case study of system behavior, with conclusions bounded by workflow asymmetries, local networking, and the lightweight MedNIST workload.

0.26

AutoC2V: Synthesis-Guided LLM Training for QoR-Optimized C++-to-Verilog Generation

cs.AR cs.LG ·2026-11-07 ·Guanqi Lyu, Fei Zhao, Zhaojun Lu, Jiaji He, Yongqiang Lyu
Abstract · 14 pages, 7 figures, 4 tables

High-level synthesis (HLS) can translate C/C++ kernels into register-transfer level (RTL) designs, yet generating high-quality hardware still relies heavily on manual pragma tuning and repeated synthesis evaluation. We propose AutoC2V, a C++-to-Verilog generation framework driven by real synthesis feedback that enables large language models (LLMs) to directly generate RTL with improved quality of results (QoR) from synthesizable C++ kernels. We construct training data from real synthesis results, including synthesizable C++ source code, the corresponding Verilog RTL, QoR metrics, and optimization objectives. Using this dataset, we align the model toward synthesis-aware RTL generation through Supervised Fine-Tuning (SFT) and Group Relative Policy Optimization (GRPO). To reduce the high cost of repeatedly invoking real synthesis during GRPO, we design the QoR Prediction Reward Model (QoR RM), which predicts latency and resource metrics by encoding RTL into long-context semantic representations, thereby providing low-cost and continuous QoR reward signals for policy optimization. Experimental results demonstrate that AutoC2V not only substantially improves RTL functional pass rates, but also achieves superior latency and resource optimization compared to the unfine-tuned base model, representative HLS optimization approaches (HLSPilot, LLM-DSE, and AutoDSE), and advanced general-purpose commercial models (Claude Opus 4.8, GPT-5.5, Gemini 3.1 Pro, DeepSeek v4 Pro, and GLM-5.2). These results indicate that combining real synthesis data with efficient QoR reward modeling can further improve the ability of LLMs to generate high-quality RTL.

0.24

From Radio Measurements to Sensing Services: A 3GPP Perspective on Multi-UAV Tracking with Networked ISAC

cs.NI ·2026-11-07 ·Xingqin Lin
Abstract · 7 pages, 6 figures

Integrated sensing and communication (ISAC) can turn radio measurements into networked sensing services, but accurate position estimates alone do not ensure useful tracking. Missed targets, late acquisition, track interruptions and aging observations also shape the information available to an application. This article examines tracking of noncooperative unmanned aerial vehicles (UAVs) from a Third Generation Partnership Project (3GPP) perspective. It connects sensing-service requirements and network architecture to three information interfaces: receiver search regions, measurement context and track validity. A five-UAV case study in a 3GPP urban scenario and a ray-traced San Francisco scene illustrates the tradeoff between target coverage and extra outputs, the distinction between acquisition and ongoing continuity, and the value of testing a static mission envelope before adding adaptation. An observation-age example explains how a state predicted to the present can still rely on old evidence. The resulting lessons are to align receiver search with requested coverage, preserve measurement time and uncertainty, and assess coverage, ambiguity and timeliness together. We conclude by pointing out fruitful avenues for further work in sensing design, calibration and service validation.

0.21

PENSA: A Spatial Accelerator for BitNet b1.58 Inference on an HBM-equipped AMD Alveo FPGA

cs.AR eess.SP ·2026-11-07 ·Karthikeyan B. Kalyanasundaram, Ivan S. Maksymov, Philip H. W. Leong
Abstract

Ternary large language models (tLLMs) replace general multiplication in dominant linear layers with sign-controlled accumulation, shifting hardware bottlenecks toward packed-weight delivery, reduction, and operator occupancy. PENSA is an end-to-end FPGA accelerator for the BitNet b1.58 2B4T model on an AMD Alveo U50. Two row-partitioned ternary matrix engines consume packed TQ1_0 weights from eight independent high-bandwidth memory (HBM) pseudo-channels, each processing 1,280 ternary weights per cycle through sign selection and integer reduction. The sequencer, attention, and vocabulary-projection units remain resident on the FPGA, while two-stream interleaved scheduling overlaps sequences across the persistent pipeline to form the Penta-Unit. At 104.1~MHz, PENSA achieves 58.09 tok/s at context 128 and 24.60 tok/s at context 2048, corresponding to 2.14-3.14$\times$ the throughput of a same-model Xeon E-2356G baseline. The resulting system demonstrates end-to-end FPGA execution of 2B-scale BitNet topology without modifying its model structure.

0.21

Extending DBLog-Based Incremental Snapshots to Handle Orphan Rows in CDC Pipelines

cs.DB cs.DCcs.LO ·2026-11-07 ·Andreas Andreakis
Abstract · 15 pages, 4 figures, 2 tables. Isabelle/HOL, Lean 4 and TLA+ development: https://doi.org/10.5281/zenodo.24035424

Applications use heterogeneous datastores where a source feeds sinks such as search indexes or data warehouses via change data capture (CDC). As transaction logs have limited retention, CDC alone cannot bootstrap sinks, and traditional approaches either stall source writes or hold back log events. DBLog-based incremental snapshots, adopted by Debezium and Flink CDC, read tables in chunks interleaved with the log without blocking writes or long log stalls and resume where they stopped. After the initial bootstrap, a sink can drift when losing change events or receiving unintended writes. Even a repeated bootstrap can leave orphan rows that exist in the sink and not at the source. A common remedy is to populate a new empty sink, but repairing an existing sink avoids parallel pipelines, a switch and serving traffic from a drifted sink until the new sink is ready. In DBLog's original design, each chunk only emits the latest state of the rows the table still contains, so it does not handle orphan rows. We extend DBLog to delete them using what a chunk read already knows, namely the keys that the range select did not return, emitting a new kind of event that sinks apply as deletes. We also add a repair request that an external process calls with orphan candidates, which the connector checks with its own read, emitting these events for keys confirmed absent. This enables three modes. In the fully integrated mode, every chunk pass removes the orphans it finds during full or periodic repairs. In the second mode, an external drift process hands suspected orphans to the connector. The third mode combines both and is what we recommend. Together, they turn DBLog into a complete data synchronization engine whether a sink catches up on live changes or missing rows, holds wrong column values, or contains orphan rows. Their safety and convergence are formally verified in Isabelle/HOL and Lean 4.

0.19

Network Contention State Aware Dynamic Optimizer for Accelerator Collective Communication Libraries

cs.DC cs.LGcs.MA ·2026-11-07 ·Rajat Bisht, Tevfik Kosar, Anthony Skjellum, Gene Cooperman, MD S Q Zulkar Nine
Abstract

Collective Communication Libraries (CCL) are critical for large-scale distributed AI/ML training, yet their configurations are determined by static heuristics that do not account for real-time network conditions. Our empirical study shows that these heuristics perform near-optimally in contention-free environments, but degrade substantially under background traffic. To address this, we present a cooperative multi-agent run-time optimizer for accelerator collective communication. Our approach employs a two-agent architecture: (i) detector- a Long Short Term Memory(LSTM) recurring neural network fed classifier head with a sequential probability ratio test that distinguishes sustained contention from transient spikes; and (ii) selector- a configuration selector built on a Dueling Double DQN (D3QN) with Conservative Q-Learning that adapts configuration parameters in response. We implemented this dynamic tuner over NCCL, a widely used CCL developed by NVIDIA. When evaluated on Chameleon Cloud's isolated bare-metal cluster, this dynamic tuning approach reduces AllReduce latency by 36% over NCCL and 32% over a fully converged AutoCCL. We further evaluated it across six real AI workloads spanning convolutional networks, transformers, and recommendation architectures. Our multi-agent tuner approach delivers up to 65% speedups on Delta cluster with NVIDIA A100 GPUs and Slingshot-11 fabric internode connectivity.

0.19

OneReel: Efficient Multi-Scene Video Generation with Scene-Structured Attention

cs.DC ·2026-11-07 ·Wei Zhang, Yuchen Wang, Tianrui Feng, Hongrui Wu, Chenfeng Xu +1 more
Abstract

Multi-scene video generation must follow scene-specific prompts while preserving consistency across scenes. Dense joint generation enables cross-scene information exchange but incurs quadratic attention cost, whereas separate generation avoids this cost at the expense of cross-scene consistency. We present OneReel, a training-free system that exploits scene structure to enable efficient multi-scene generation. Our key insight is that multi-scene generation can preserve complete intra-scene attention while restricting cross-scene information exchange to a bounded shared memory. OneReel uses scene prompts and durations to establish scene-specific attention scopes and dynamically selects a bounded set of representative latent frames for cross-scene sharing. This reduces video attention complexity from quadratic to linear in the number of scenes. To translate these savings into execution speedups, OneReel packs scene-local and shared K/V into dense variable-length attention and employs work-balanced multi-GPU partitioning with communication-aware selection between K/V gathering and Ulysses-style parallelism. Our evaluations across four video generation models show that OneReel achieves 1.7-4.9x end-to-end speedup over dense attention and 1.2-2.2x over existing sparse-attention advances, while improving text-following accuracy by 34-37 percentage points and maintaining cross-scene consistency.

0.17

Software & Programming 10

What extraction cannot slim, and whether function injection can: a pre-registered study of 580 Java repositories

cs.SE ·2026-11-07 ·Daniel Sagenschneider
publishedcode/dataunusual
Abstract · 19 pages, 1 figure, 10 tables. Pre-registered; every number regenerated by script. Replication package in the ancillary files and at https:/…

A method of 26 code lines or more reads as a Long Method, and Extract Method is the usual fix. Some long methods have no single-exit fix: a return cannot leave an extracted piece, a piece hands back one value, and some methods are knotted so that every cut is blocked. Function injection, as the OfficeFloor framework does it, cuts differently: each piece becomes a function the framework injects and sequences, an early return ends a function, and a function hands on several values. The framework's author pre-registered a test of whether it takes apart what extraction cannot. A stratified random sample of 580 open-source Java repositories, each pinned before AI coding assistants were in general use, was scanned by released tools whose cost models were fixed before the sample was drawn: 3,026,902 methods, 168,466 of them long. Single-exit extraction cannot slim 22.8% of the long methods. The registered hypothesis, that at least a quarter of those stuck methods are ones only function injection takes apart, is refuted: 22.2% (95% CI 20.3 to 24.3). Nearly half of the stuck methods are early-return methods the plan counts as conventionally fixable, and the split takes three in four of those apart as they stand. Where extraction would do, the split chooses injection and costs less for 29.0% against extraction wrapped as one function and 17.2% against plain extraction, almost always as a straight sequence of functions. An exploratory reading of the quarter of the stuck methods that neither takes apart finds a third of it tables and literals, long for their data rather than their logic, and most of the rest blocked by a jump inside a loop, by a nested statement too large to cut, or by values too many for any cut. Three registered studies are proposed: the classes as rules in the tools, a refinement per class, and the same measures on code written with AI assistants.

0.56

Quadratic Bendixson criterion for delay systems: a numerical implementation

cs.MS math.DSmath.OC ·2026-11-07 ·Mikhail Anikushin, Andrey Romanov
unusual
Abstract

We present a software implementation (as a Python package) of approximation schemes for verifying frequency inequalities arising from the quadratic Bendixson criterion for systems of delay equations (with a single delay in the linear part). In applications to dissipative nonlinear autonomous systems, such inequalities ensure global stability. The theoretical foundations of this approach were developed in our related work. In particular, it was shown that the transfer operators involved into the frequency inequalities can be represented as integral operators with bounded kernels of the Hardy--Volterra type. Here, we use this representation to construct approximation schemes based on finite-dimensional truncations of the transfer operators. Examples of using the package are provided.

0.54

Safe by Accident: Crash-Testing LLM-Written Code

cs.SE ·2026-11-07 ·Afridi Mahmud, Md. As-Aid Rahman, Rifat Shahriyar
Abstract · 9 pages, 4 figures, 4 tables

Evaluations of LLM-written code run it once, in one environment, and credit the code with whatever passes. We show that this credit can belong to the environment. We crashed LLM-written backend code just before each database, file and email call, re-ran it as an operator would, and judged the result under two transaction modes of Python's sqlite3: the default, in which the library quietly opens a transaction, and autocommit, in which it does not. Across 9,100 solutions from ten models on 24 backend tasks, more than half (54%) of the database code that survives every crash in the default mode is left half-updated in autocommit mode. This holds for every model, and it hides in ordinary code, including with conn: blocks that look like transactions but open none. Telling the model that its connection is in autocommit mode does not change what the local code models write, while the API models' unsafe share falls by 30 points. Asking for crash safety helps every model, and strong models repair nearly all unsafe code, but often write safe code only when asked. Thus, much of the crash safety that LLM-written database code appears to have is borrowed from the library rather than designed into it, and a second test under a less forgiving configuration reveals the difference.

0.37

Mismeasuring the Life of a Line of Code: Lessons Learned the Hard Way

cs.SE ·2026-11-07 ·Pavel Gurov
unusual
Abstract · 12 pages, 2 figures, 4 tables. Corrects results of arXiv:2606.04993

Studies of how long lines of code live, including lines written by AI agents, rest on choices that look technical: when a line counts as gone, what to do with lines still alive, whether to pool projects, and how to evaluate a model. We studied 32.5 million line births in 120 TypeScript repositories and got several of these choices wrong at first. For ten of them, we compare the result as we first computed it, or as commonly done, with the result after correction. All ten change conclusions, or would have. Six were our own mistakes. Counting moved and rewritten lines as dead shortens a line's median life from 3.0 years to 1.3. Counting any change as a death hides an outcome that can be predicted: within-project concordance falls from 0.71 to 0.56. Regressing observed lifetimes discards three quarters of the lines. Keeping rewritten lines at risk makes protective effects grow with age. Lines are born and removed together, so treating them as independent misleads twice: nine of fifteen effects look significant only under that assumption, and a split by file makes a barely predictable outcome look good (within-project concordance 0.71 instead of 0.61). After the corrections, removal depends on a line's age and project, but our models could barely rank it within a project. Local revision they could rank (0.71). Obvious cases aside, the top 1% of a project's new lines are revised at 2.3 times the base rate. Each pitfall leaves a symptom; we give a checklist.

0.34

Applicative theories can count

cs.LO ·2026-11-07 ·Melissa Antonelli, Marcel Ertel, Eduardo Skapinakis
hot direction
Abstract

Applicative theories offer a proof-theoretic framework for studying complexity. In this paper, we introduce a family of applicative theories to characterize classes ranging from constant-depth circuits to polynomial space. Our approach relies on a simple axiom scheme, the small variations of which allows to capture various forms of counting and alternation. This leads to highly uniform results. When applied on notation, the scheme captures bit counting, yielding a characterization of the functions computable by constant-depth, polynomial-size threshold circuits ($\textbf{FTC}^0$); when applied lexicographically, it allows for the counting of accepting computations, yielding a characterization of the functions in the counting hierarchy ($\textbf{FCH}$). Increasing the logical complexity of the induction formulas allowed in the scheme, we obtain characterizations of the functions computable by bounded Boolean circuits of logarithmic-depth and polynomial-size ($\textbf{FNC}^1$) and of the functions computable in polynomial space ($\textbf{FPSPACE}$). The latter characterization solves an open problem in the field. Finally, we investigate separation results. While it is unknown whether $\textbf{FNC}^1$ and $\textbf{FCH}$ are different, there is an oracle relative to which (relativized) $\textbf{FNC}^1$ is strictly contained in (relativized) $\textbf{FCH}$. We use this fact to separate the theories characterizing them, by exploring their ability to represent higher-order functionals.

0.29

Hybrid Source Code Change Impact Analysis

cs.SE cs.AI ·2026-11-07 ·Katharina Stengg, Christian Macho, Martin Pinzger
code/data
Abstract

Change impact analysis (CIA) helps developers identify code entities impacted by their initial changes. While several approaches for CIA were proposed, they still struggle with either low precision or low recall, thereby increasing manual effort for developers to verify the results. Furthermore, limitations in the used datasets and experimental setups can bias the results reported by many CIA approaches. In this paper, we first address the dataset limitations by manually labeling 1,931 changed code entities stemming from 263 commits of 22 open-source Java projects. Next, we introduce two variants of katIA, our code change impact analysis approach that combines static impact analysis with the reasoning capabilities of large language models (LLMs). The first variant uses a single prompt (katIA-SP) for refining an initial static impact set while the second variant uses three specialized agents (katIA-MA). The evaluation with our curated dataset shows that katIA-SP with Gemini 2.5 Flash achieved an F1-score of 22.4% when matching impacted entities by their fully qualified names (FQN) and 27.0% when matching them by their class and entity names (CE). The latter shows an improvement of 2.0 percentage points over the state-of-the-art. This score is further improved by katIA-MA which, using Gemini 2.5 Flash, yields an FQN-matching F1-score of 29.1% and a CE-matching F1-score of 33.3%. While these improvements come at the cost of increased median generation time from 0.570 s to 11.214 s and median token usage from 1,235 to 6,462 tokens, the improved performance justifies the additional cost. The comparison of three LLMs using katIA-SP finds that Gemini 2.5 Flash is the most consistent model. These results indicate that combining static impact analysis with LLM-based refinement, particularly with the multi-agent variant, can improve code change impact analysis.

0.26

Towards A Parametrized Complexity of Determinization

cs.FL ·2026-11-07 ·Ivan Baburin, Ryan Cotterell
Abstract

Given a nondeterministic finite automaton (NFA), we want to estimate the difficulty of finding an equivalent deterministic finite automaton (DFA). We demonstrate that computing the state complexity of an NFA within polynomial precision is \textsf{PSPACE}-hard, and furthermore the same holds for the complexity of the subset construction, Brzozowski's algorithm, or, indeed, \emph{any} determinization algorithm which runs at least as fast. This result implies that making any \textit{a priori} estimate of the running time of the major determinization algorithms is inherently hard. To circumvent this---in the spirit of parametrized complexity---we try to identify complexity measures which can guarantee efficient determinization. Our main idea rests on linking the complexity of the subset construction to transition monoids using the semigroup complexity. We show that the semigroup complexity by itself can be reasonably approximated for certain classes of NFAs, and then introduce \emph{subset complexity}---a refinement of the semigroup complexity which has a tighter connection with the subset construction. In certain scenarios, the subset complexity can be approximated in polynomial time, making it potentially viable for more practical applications.

0.25

Patch correctness in automated program repair: A five-dimensional evidence framework and cross-benchmark audit from classical repair to coding agents

cs.SE ·2026-11-07 ·Shuai Wu, Yanna Feng, Yufang Li, Xue Li, Zhijun Wang +2 more
benchmarkcode/data
Abstract · Technical report. 24 pages, 2 figures, 6 tables. Dataset and evidence matrix available at https://doi.org/10.5281/zenodo.23639376

Tests guide automated program repair and determine which candidate patches are reported as successful. Repository-level coding agents extend this feedback loop by generating tests and revising code after observing execution results. This technical report introduces a five-dimensional evidence profile and reporting protocol to audit how test observations support patch-correctness claims across classical repair systems and coding agents. We assemble an audit dataset of 39 benchmark and evaluation artifacts: 32 primary empirical studies, one institutional technical audit, and six prior taxonomic baselines. We code the 33 primary contributions along five dimensions: test provenance and exposure, behavioral scope, oracle basis and dependence, execution integrity, and label adjudication. The central distinction separates checks visible during repair search from evidence reserved for final assessment. Four worked comparative audits show how changed candidate populations, oracle rules, or execution budgets alter reported detection rates. Additional tests can expose missed faults while rejecting valid alternatives or reducing correct-repair yield under a fixed budget. Repository-level audits separate issue reproduction from candidate-patch discrimination and distinguish automated discrepancy detection from requirements-based adjudication. An evidence matrix links each coded decision to supporting passages and records shared benchmark ancestry. We specify a reporting protocol for repair benchmarks and maintenance pipelines that retains uncertain judgments and records validation cost alongside fault detection and valid-patch retention.

0.24

SIExBench: A Benchmark for Automated Repair of Sensitive Information Exposure

cs.SE cs.CR ·2026-11-07 ·Mingming Zhang, Xu Wang, Jian Zhang, Dihao Fan, Xiangxin Meng +1 more
benchmark
Abstract

Modern software systems routinely process credentials, personal records, and other sensitive information. A sensitive information exposure (SIEx) defect makes such information accessible to an unauthorized actor and can cause serious and often irreversible disclosure. However, prior research has focused largely on detecting such exposures rather than repairing the underlying defects, and existing repository-level issue-resolution benchmarks do not specifically target this defect class. We present SIExBench, to our knowledge the first benchmark dedicated to automated repair of SIEx defects. SIExBench contains 161 execution-verified instances from 64 widely used Python and Java repositories, each with an issue report, buggy repository snapshot, developer patch, executable exposure and regression tests, and defect characterization. We evaluate six repair systems with three LLMs, yielding 18 configurations, and systematically analyze their overall effectiveness, performance across defect characteristics, and failure mechanisms. The strongest configuration resolves only 39.1% of the instances, while 40.4% remain unresolved by all 18 configurations. More importantly, 62.1% of unsuccessful attempts produce a valid patch that still leaves sensitive information exposed, making incomplete exposure removal a central challenge. Systems often generate plausible patches and modify relevant code without fully eliminating the exposure, indicating that effective repair must ensure that all relevant paths to sensitive-information exposure are closed. Overall, SIExBench provides a foundation for systematically evaluating and advancing automated repair of sensitive information exposure defects.

0.22

Crustify: Repository-Level Generation of Safe Rust Bindings with LLM Agents and Coverage-Guided Testing

cs.PL cs.AIcs.CRcs.SE ·2026-11-07 ·Marius Momeu, Marius Arvinte, Cory Cornelius, Koushik Sen
Abstract

Wrapping C code in a safe Rust interface lets Rust applications use it without writing unsafe code, until the code itself is migrated to Rust. Such bindings are tedious and error-prone to write, however: their internals use unsafe code that the Rust compiler does not check, so a single mistake can introduce undefined behavior (UB), and they have so far been written mostly by hand. We present Crustify, an agentic co-translator that generates safe Rust bindings for repository-level C libraries. Crustify schedules a target's types and symbols into dependency-ordered waves of independent batches that agents bind concurrently. Its agents derive the ownership and lifecycle contract of every pointer and type they wrap, check each binding with coverage-guided equivalence and UB tests under sanitizers and Miri, and review every wave. They build on three tools that we contribute: (a) ffibox, Rust primitives for C/Rust interop that mirror Box, Arc, and Vec for C objects without ever forming a reference to memory that C may alias; (b) Wavefront, a neuro-symbolic static analyzer that orders a target's items by their dependencies, including those that casts and type-erased pointers hide; and (c) ffimap, a rustc-based analysis that maps the unsafe surface of bindings. On libgit2 and ffmpeg, with one developer in the loop, Crustify generated safe bindings for 41.9% and 42.8% of their public APIs in about 51.9 and 58.9 hours. On the interface consumers are expected to use, only 1.6% and 0.4% of the bindings' public functions are unsafe, and only 0.1% and 0.6% have a raw pointer in their signature. Their tests cover 98.92% and 99.07% of the Rust code and 45.95% and 67.20% of libgit2 and of ffmpeg's libavutil, and the reviewers fixed 8 and 13 defects that caused UB reachable from Rust. To our knowledge, this is the first work to automate the generation of safe Rust bindings for C libraries.

0.09

Theory & Algorithms 24

Close and Sort: Exact Maximum-Likelihood Aggregation of Cyclic Rank-Two Preferences

cs.DS cs.IRcs.LGstat.ML ·2026-11-07 ·Kehan Liu, Zhaoyun Chen, Xianquan Yan
publishedcode/data
Abstract · 10 pages, 3 figures, 5 tables. Accepted at the 20th ACM International Conference on Web Search and Data Mining (WSDM 2027). Code: https://gi…

Pairwise probabilities arise in rank aggregation, recommendation, and human or LLM evaluation, yet converting them into a maximum-likelihood list is a weighted Kemeny problem and is NP-hard in general. Known exact algorithms for rank-two tournaments use only comparison signs and do not optimize this likelihood. We show that the weighted problem is exactly solvable when the skew-symmetric log-odds matrix $W_{ij}=\log(p_{ij}/p_{ji})$ has rank two, including when the induced preferences contain cycles. Writing $W_{ij}=\det(h_i,h_j)$ and adding the closing vector $h_0=-\sum_i h_i$ turns the objective into twice the signed area of a closed polygon. This geometry yields CloseSort, which polar-sorts the $n+1$ edge vectors, cuts after $h_0$, and returns a global maximum-likelihood order in $O(n\log n)$ time from the factors. The same geometry identifies the runner-up and gives a stability certificate beyond exact rank two. When the smallest determinant $δ$ of two consecutive edges in the closed angular cycle is positive, it is exactly the best-to-runner-up log-likelihood gap. For $W=W_0+E$, the CloseSort order of $W_0$ has regret at most $\sum_{i<j}|E_{ij}|$ on $W$ and remains optimal whenever this sum is below $δ$, a worst-case tight threshold. In 5,120 determinant-logit simulations, increasing the observations per pair from 4 to 64 raises the fraction on which rank-two projection outperforms the empirical plug-in optimum from 68.6% to 85.7%. On 3,600 Arena-Hard LLM-judge profiles drawn from 63 systems and ten judges, the leading rank-two component captures 89.2% of the log-odds energy. On profiles containing a directed majority triangle, CloseSort reduces normalized regret by 44% relative to Borda and improves the mean normalized objective over each of eight baselines. With factors supplied, it is 8,407 times faster than exact dynamic programming on 18-item rankings.

0.52

Solving Large Rubik's Cubes within a Factor of $2.37+o(1)$ of Optimal

cs.DS cs.CC ·2026-11-07 ·Jeremy Mrzyglocki, Tomas Rokicki, Maxine Perroni-Scharf, Erik D. Demaine
theorycode/dataunusual
Abstract · 41 pages, 18 figures, 16 tables. Project page: https://largecubebench.org

We design optimized algorithms to solve the $n \times n \times n$ Rubik's Cube within $30.5 n^2 / \lg n + o(n^2 / \lg n)$ moves. This upper bound is asymptotically within a factor of $2.37$ of the known lower bound, which applies to almost all instances. The constant $30.5$ is also the first explicit constant factor for the known upper bound of $O(n^2 / \lg n)$, where the implicit constant of the previous algorithm was $1.01 \cdot 10^{40}$ (ten duodecillion). Thus we improve the upper-bound constant by a factor of more than $3.30 \cdot 10^{38}$ (330 undecillion). Our algorithm runs in time linear in the input size $N = Θ(n^2)$, and uses $O(\sqrt{N} \lg N)$ extra space (beyond the state of the cube). Our implementation is the first practical cube solver to generate solutions of length $O(n^2/\lg n)$. Our latest solver outperforms existing solvers for $n$ as small as $840$.

0.38

Tunnels: Compressing the Solution Subgraph of Regular Path Queries

cs.DS ·2026-11-07 ·Adrián Gómez-Brandón, Gonzalo Navarro
Abstract

Regular path queries (RPQs) are increasingly expected to return the paths that match them, not only their endpoints. Since the number of matching paths can be exponential or infinite, a standard way to deliver them (PMR) is the subgraph $\Hs$ of the product between the graph and the query automaton that contains exactly the nodes lying on some solution path. We show that $\Hs$ can be compressed much further when the graph is available to the decoder: most of $\Hs$ can be regenerated by re-running small parts of the traversal, and only the places where the traversal would waste significant effort on dead branches need to be stored. We formalize this as \emph{tunnels}: regions of $\Hs$ that are re-expanded from a stored entry point with a bounded amount of extra work, either an absolute budget or one proportional to the output produced. We handle the three classes of RPQs: constant-to-variable, constant-to-constant, and variable-to-variable. In the latter, we study the combination of tunnels with cutting the query at a selective articulation point of the automaton and encoding the answer as a union of Cartesian products of leftward and rightward traversals. Our experimental results show improvements of 2 to 3 orders of magnitude over the baseline space of PMR, for an acceptable increase in traversal time.

0.36

On the Convergence of First-Order Methods in Signaling Games

cs.GT ·2026-11-07 ·Rilind Sahitaj, Artem Tsikiridis, Martin Bichler
top venue
Abstract · Accepted at NeurIPS 2026

We study first-order learning dynamics in Bayesian signaling games, a minimal extensive-form model of strategic interaction under asymmetric information. In these games, a Sender observes a private type and chooses a signal, after which a Receiver forms a posterior belief and selects an action. This Bayesian belief update makes the induced learning field fundamentally different from the affine game-gradient fields of normal-form games. Empirically, we find a sharp dichotomy: standard online learning algorithms reliably converge to strict Perfect Bayesian Equilibria (PE), but not to non-strict ones. We show that this behavior cannot be explained by the global geometric conditions commonly used to prove last-iterate convergence, such as monotonicity or the Minty variational inequality. Indeed, these conditions can fail even in binary signaling games. Instead, we identify the relevant local geometry. Our main result characterizes variational stability in finite signaling games: a PE is variationally stable if and only if it is strict. This yields local last-iterate convergence guarantees for online mirror descent and regularized dual averaging near every strict PE. We then study global average-iterate convergence and identify a class of signaling games in which coarse-trigger regret-minimizing dynamics converge to the unique PE. Finally, we complement the theory with experiments comparing projected gradient ascent, mirror-based methods, optimistic variants, and PPO across canonical signaling-game families. The results position signaling games as a compact testbed for understanding how Bayesian belief formation reshapes the convergence geometry of multi-agent learning.

0.35

Performance Analysis of Entanglement Swapping DV-QKD over MIMO FSO Channels

cs.IT eess.SP ·2026-11-07 ·Sushil Kumar, Soumya P. Dash, George C. Alexandropoulos
Abstract · 13 pages, 8 figures

This paper studies multiple-input multiple-output (MIMO) free-space optical (FSO) quantum communication systems employing polarization-encoded discrete-variable quantum key distribution (DV-QKD), based on entanglement swapping, with the goal of extending secure communications beyond the reach of a direct link. The FSO channels are subject to atmospheric absorption, diffraction, beam wander, pointing errors, and turbulence-induced fading, while background noise and detector imperfections degrade the performance. An eavesdropper controlling the channels and the relay is also considered. To analyze the performance of the considered system, both hops MIMO channels are decomposed into parallel spatial eigenchannels via singular value decomposition (SVD), photon loss is modeled using Kraus operators, and a threshold-detector model is developed for the relay. Based on these models, closed-form per-eigenchannel expressions for the relay gain are derived, which are then used to obtain expressions for the quantum bit error rate (QBER) and the secret key rate (SKR). Numerical results demonstrate that entanglement swapping maintains a low QBER over longer distances than direct transmission, extending the communication range beyond the noise-limited cutoff of the direct link, at the cost of a lower SKR at short distances. The crossover distance beyond which entanglement swapping prevails is determined mainly by the background noise and the detector technology, and it is demonstrated to increase with MIMO configurations. Moreover, the reach gain is shown to be maximized for a midpoint relay. Finally, since each swapping event requires a single-pair emission from both sources, a near-deterministic entangled-photon source is required, which spontaneous parametric down-conversion sources do not provide at practical key rates, whereas MIMO substantially relaxes this requirement.

0.33

BP-based qLDPC Decoding under Long-range Correlated Noise: Trapping Configurations and Prior Mismatch

cs.IT quant-ph ·2026-11-07 ·Abhinav Dangi, Avhishek Chatterjee
Abstract

We study the robustness of two state-of-the-art BP-based qLDPC decoders, Relay-BP and beam search, designed for independent noise when applied to spatially correlated noise with unchanged single-mechanism error probabilities. We generate correlated errors with a hidden Gaussian Markov random field, which lets us control the strength and range of the correlation while keeping every marginal error probability exact. This setting complements correlation-aware decoding: it asks how existing decoders perform when their assumed joint error model is mismatched. Across multiple tested circuits and code instances that we simulated, we observe that correlations can change logical error rates by orders of magnitude, reduce the finite-size distance advantage, and alter the relative ranking of Relay-BP, beam search, and BP-OSD. On one of the two tested circuits, Relay-BP and beam search become worse than BP-OSD, and increasing relay memory or beam width does not compensate for the degradation. Configuration analysis indicates that correlation primarily increases the probability of trapping configurations that are known to challenge these decoders; at matched error weight, correlated errors still cause decoding failures several times more often than i.i.d. errors. Finally, supplying the decoders with the marginals conditioned on the hidden field recovers most of the lost performance in the tested settings identifying prior mismatch as a major contributor to performance degradation under correlation.

0.32

A Grover-Amplified Oracle for Maximum Clique on Graph, Minimum Hitting set and Dualizations on Hypergraphs Built on Quantum Phase Estimation

cs.CC quant-ph ·2026-11-07 ·Francisco Piñero, Fernando L. \Pelayo, Hernan I. de la Cruz, Jose J. Paulet, Mauro Mezzini +1 more
Abstract

In this paper, we use a quantum circuit called the Graph Encoding Operator which, applied to the complement $\bar G$ of any graph $G$, yields an oracle for the maximum clique problem suitable for amplification via Grover's algorithm. Analogously, we devise a Hypergraph Encoding Operator which, when applied to a hypergraph $\mathcal H$, yields an oracle for the minimum hitting set suitable for amplification via Grover's algorithm. Interestingly, the latter oracle provides, when measured, an approximation of the number of hitting sets of the hypergraph, thus solving the dualization problem for positive Boolean functions represented in a positive irredundant disjunctive normal form. We give the full construction of both operators, prove their correctness, analyze their complexity, and provide an implementation on a quantum simulator. We discuss the resulting complexity, contrast them with existing quantum computer algorithms based on Grover search and we will show that our algorithms outperform, to the best of our knowledge, all of the previous algorithms on space computational complexity by requiring only $O(N)$ number of qubits with respect to the number of vertices $N$, for representing both graphs and hypergraphs in the encoding operator and, consequently, results in an oracle for both graphs and hypergraphs that uses only $O(N)$ qubits even when the hypergraph contains an exponential (on the number of vertices) number of hyperedges.

0.29

Initialization masks overcome decoding obstructions in compressed QAOA

cs.CC quant-ph ·2026-11-07 ·Emilio Semre, Steven Frankel
Abstract · 24 pages, 6 figures, including technical appendices

In compressed quantum optimization, a variational state is converted into a classical assignment by decoding noncommuting observables. We show that this step creates structural restrictions that changing the initialization can remove, even when increasing circuit depth cannot. For an explicit infinite family of connected cubic graphs on $n=3r$ vertices, with a fixed Pauli encoding and deterministic sign decoder, every all-plus QAOA circuit has cut at most $2r$, at every finite depth and every real angle sequence. A uniform signed product-state mask at fixed depth-one angles has expected cut $13r/4$, an additive advantage $5n/12$; an infinite subfamily admits an optimal mask. A second construction retains a linear depth-one separation under weak weighted couplings, a bounded cost-angle domain and a specified production-encoder ordering. Noiseless simulations then test classical mask selection before energy optimization. On 1,699 random cubic MaxCut instances, one selected mask achieves mean ratio $0.9100$, versus $0.7067$ for the all-plus $X$ baseline and $0.9125$ for the best of $r^2$ separately optimized random masks. On complete cohorts of weighted MaxCut, sparse Ising, Max-2-XOR and Max-2-SAT, the selected mask also improves on the baseline at every size, but it trails the pool at small sizes and leads it significantly only at the larger MaxCut and Ising sizes. Finite-pool order statistics separate the effect of selection from that of drawing a random mask. The results distinguish decoded expressivity from optimized energy and empirical performance; they do not establish a quantum computational advantage.

0.29

Positivity Proofs for Third-Order Holonomic Sequences via Differences Scaled by Rational Functions

cs.DM cs.SCmath.DS ·2026-11-07 ·Fugen Hagihara
Abstract · 23 pages, 1 figure

We study the positivity of sequences $f$ that satisfy a third-order holonomic recurrence $f(n+3)= P_2(n) f(n+2) + P_1(n) f(n+1) + P_0(n) f(n)$ whose coefficients $P_0$, $P_1$ and $P_2$ are rational functions in $\R(x)$. We provide a sufficient condition on $P_0$, $P_1$ and $P_2$ for the recurrence to admit a solution all but finitely many terms of which are positive, and whose positivity is robust under small perturbations of the initial values. This sufficient condition is also a necessary condition for a subclass of the recurrences. By developing the proof of this sufficient condition, for another subclass of the recurrences, we show that it is decidable whether all the terms of a given sequence are positive for any initial values outside an exceptional plane. It is common to construct a positivity certificate, which guarantees the positivity of a solution, to show this kind of decidability. In this paper, we construct a new type of positivity certificate via the $S$-scaled difference $f(n+1) - S(n) f(n)$ for a suitable $S \in \R(x)$. This result partially improves the work by Ibrahim and Salvy.

0.28

Quantum channel bounds for constant-composition codes

cs.IT math.CO ·2026-11-07 ·Artur Akhiiarov, Peter Boyvalenkov, Danila Cherkashin, Andrei Raigorodskii
Abstract · 45 pages, 1 figure, 5 tables

We extend the quantum-channel optimization of Alrabiah and Guruswami to $q$-ary constant-composition codes. We study explicit channel and flag constructions that recover selected classical asymptotic upper bounds, including the Ben-Haim--Litsyn, moving-subspace and Horn-channel bounds. We give two equivalent formulations, Gram and flag, and prove stability of the obtained upper bounds under permutation transport and generalized shortening. A hierarchy defined by bounding the output-state ranks within each flag improves strictly between consecutive levels at every composition and every positive distance below the Plotkin threshold. We provide finite matrix parametrizations of these bounds (i.e., the number of variables does not depend on the length of codes under consideration). Numerical optimization rapidly produces improvements over the known asymptotic upper bounds throughout all tested nonbinary parameter ranges. For rational input data, each finite level admits near-optimal feasible rational descriptions of polynomial length. Their feasibility can be checked exactly and their objective values computed to any prescribed accuracy in polynomial time; the associated threshold problem with a prescribed gap belongs to $\mathsf{PromiseNP}$.

0.27

A Cooperative Game Theoretic Model for Sustainable, Smallholder Carbon Farming

cs.GT ·2026-11-07 ·Geetha Charan, Thandava Sunkara, Rohit P Suresh, V. Priyanka, Manojkumar Patil +2 more
dataset
Abstract

Carbon farming offers a promising path to improve farmer livelihoods, mitigate climate change, and make farming sustainable. However, smallholder farmers, who dominate agriculture in developing economies, face structural barriers to participating in carbon farming due to high certification costs, limited landholdings, and limited budgets available to them. In this paper, we develop a cooperative game-theoretic framework to study how coalition formation can enable viable and profitable participation of smallholder farmers in carbon farming. Our approach explores three components: (i) Solo Portfolio Optimization (SPO), which determines an optimal portfolio of carbon farming practices that individual farmers can adopt under budget constraints; (ii) Joint Portfolio Optimization (JPO), which computes the maximum total economic surplus achievable through the formation of farmer coalitions; and (iii) surplus allocation among participating farmers in a principled way using cooperative game-theoretic solution concepts, such as Shapley value, core, least core, and nucleolus. We show that the underlying cooperative game is superadditive under an affordability condition and non-convex in general. As part of this study, we release CarbonFARM-Bench, a carbon farming dataset constructed by experts, and conduct controlled computational experiments on this dataset. We observe that solo participation in carbon farming is largely infeasible for smallholder farmers under baseline parameters, whereas coalition formation creates substantial value driven by economies of scale. In addition, we demonstrate that stability-aware allocation mechanisms are required for sustaining cooperation. Together, our results highlight that enabling smallholder carbon farming is fundamentally a challenge of institutional design, and we offer several policy recommendations to make it attractive to smallholder farmers.

0.27

The Power of Two Choices in Dynamic Pricing: As Good As Full Feedback, Instance-Wise

cs.GT cs.LG ·2026-11-07 ·Yan Dai, Tao Lin, Aleksandrs Slivkins
position
Abstract

We study the dynamic pricing problem, where a firm sells a product to $T$ myopic customers and their values are independent samples from an unknown value distribution $F$. With posted prices, the firm only sees whether each customer buys, and the minimax regret is $\widetilde Θ(T^{2/3})$ [KL03]. A menu instead offers several options, each selling the product with some probability at some expected payment. Infinitely many menu options allow the seller to fully reveal customer values, reducing the regret to $Θ(\sqrt T)$. How many options are needed? Perhaps surprisingly, a second menu option suffices, achieving the same regret as infinite menus (up to $\log$ factors). This holds broadly: without knowing any property of $F$, our algorithm with two menu options attains $\tilde O(\sqrt T)$ regret in general, $\tilde O(T^{1/3})$ regret for monotone-hazard-rate distributions, and $\tilde O(T^{d/(d+1)})$ instance-dependent regret when $d\in(0,1)$ is a suitable notion of "dimension". All are optimal even with infinite menus. Moreover, we argue, via a novel lower-bound construction, that our algorithm is near-optimal for each problem instance. Along the way, we study the query complexity of revenue maximization, where an algorithm issues a query to each customer to learn the property of interest (either the best price or the CDF) up to some precision. We show that interval queries suffice to achieve the same query complexity as value queries (up to $\log$ factors) for both properties and across several classes of value distributions. We provide a new interval-query algorithm for each property. The best-price-estimation algorithm is the key ingredient in our main dynamic pricing algorithm, and the CDF-estimation algorithm underpins a simpler algorithm attaining $\tilde O(\sqrt T)$ regret with three menu options.

0.26

Reliability Analysis for Ground-Air Cooperative Wireless-Powered Internet-of-Things

cs.IT ·2026-11-07 ·Xingyu Huang, Zhican Zhong, Yulei Wang, Yalin Liu, Yongqiang Cui +1 more
Abstract · 7 pages, 6 figures

Battery-constrained Internet of Things (IoT) devices face significant challenges in maintaining long-term uplink data collection, especially in network-limited scenarios. This paper investigates a Ground-Air cooperative Wireless-Powered IoT network (GA-WPIoT), where ground power beacons (GPBs) provide radio-frequency energy supply and unmanned aerial vehicles (UAVs) serve as flexible aerial access points for IoT data collection. Unlike conventional UAV-enabled wireless-powered solutions, GA-WPIoT decouples energy transfer and data collection, reducing the hardware and energy burden of UAVs. To capture the device-level performance variations among IoT devices, we develop a stochastic geometry-based analytical framework, where UAVs are modeled by a Matérn hard-core point process, IoT devices by a Poisson point process, and GPBs by a finite annular binomial point process. The conditional success probabilities of the GPB-device link (for energy transfer) and the device-UAV link (for data collection) are characterized, and device-level reliability distributions are obtained through the Meta-Distribution based analysis. Numerical results under suburban and urban environments reveal the impacts of UAV altitude and annular GPB deployment on reliability distributions, providing useful guidelines for GA-WPIoT.

0.25

Lower Bounds Above $\sqrt{3}$ for Online Makespan Minimization on Four Identical Machines

cs.DS math.OC ·2026-11-07 ·Maximilian Janke
theory
Abstract

For deterministic online makespan minimization, $m = 4$ identical machines is the smallest case whose optimal competitive ratio is open: it is known only to lie between $\sqrt{3} \approx 1.73205$ and $26/15 \approx 1.73333$. We prove a lower bound of $14383/8304 \approx 1.73206$ using a single branching layer, and refine the construction with a second, mirrored branching layer to obtain $14350/8283 \approx 1.73246$, closing about a third of the gap. Our constructions are the first branching adversaries for $m = 4$ machines, and the first explicit lower bounds for online makespan minimization in which the adversary presents jobs in non-monotone order of size.

0.25

On Relative Greedy Weights

cs.IT cs.CR ·2026-11-07 ·Olav Geil
Abstract

Relative greedy weights were introduced by Li and Liu in [19] as a measure of information leakage in coordinated two-party wire-tap channels of type II when the adversary follows a greedy strategy. Their definition can be viewed as a relative version of the CEZ greedy weights, which are only locally greedy. In this paper, we revisit the concept of greediness and relative greediness introducing new measures, and we discuss other perhaps more relevant applications of relative greedy weights than the coordinated two-party wire-tap channel of type II. We pay special attention to bottom-up greedy weights, which correspond to an adversary with an ultimate greedy approach, and we present families of nested codes for which such weights are larger than the relative CEZ greedy weights.

0.24

An Improved Algorithm for Quantified Boolean Formulas

cs.DS cs.AI ·2026-11-07 ·Zhiyang Chen, Hailong Yao
Abstract

Quantified Boolean formula (QBF), the quantified version of Boolean satisfiability (SAT), is a fundamental problem in constraint solving. Due to the $\mathsf{PSPACE}$-complete nature of QBF, designing theoretically efficient QBF algorithms is much harder and receives less attention in academia. In this work, we present a new algorithm based on memoized search, which solves QBF with $n$ variables and $m$ clauses in $1.685^m\cdot\text{poly}(n)$ time and space. This is the first improvement over the $1.709^m\cdot\text{poly}(n)$ bound by Williams in over 20 years. Our technique is based on the minimum vertex cover of the conflict graphs, which was proposed to study the parameterized complexity of SAT. We extend this idea to QBF and further show that QBF is slicewise polynomial-time solvable using this parameter.

0.22

Quantum synchronizable codes, polynomial and quasi-polynomial shifts

cs.IT ·2026-11-07 ·Edgar Martínez-Moro, Miguel de los Ríos de Antonio
theory
Abstract

Quantum synchronizable codes protect quantum information against Pauli errors and against misalignments of the block boundaries. Almost all known constructions use cyclic codes, because the channel shifts the transmitted block cyclically. We study what happens when the algebraic shift of the code differs from the physical one. When one uses a twisted padding, constacyclic codes synchronize at no cost, and they are the only polycyclic codes with this property for a linear padding. For the other polycyclic codes, we measure the cost of each misaligned position, and we show that the phase damage it causes can be corrected as erasures. Using annihilator duality, we extend the construction to quasi-polycyclic codes, in which the cost is confined to a single block. We prove that the usual quasi-cyclic families have a distance bound that does not grow with the length, and show that quasi-polycyclic codes over repeated-root trinomials overcome it, giving synchronizable codes with larger dimension and distance than the known synchronizable codes of the same length.

0.21

A Simple Computer-Assisted Proof of the 5/6-Bound for Real-Period Pinwheel Scheduling

cs.DM cs.DSmath.CO ·2026-11-07 ·Hiroshi Fujiwara, Grzegorz Gutowski, Katsuhisa Ouchi, Alexander Wolff
theory
Abstract

In pinwheel scheduling with real periods, an instance consists of $n$ tasks with real periods $1\le a_1\le a_2\le\cdots\le a_n$. The goal is to find a schedule that assigns one task to each day indefinitely such that, for every task $i\in\{1,\dots,n\}$ and every positive integer $\ell$, task $i$ occurs at least $\ell$ times in every block of $\lceil\ell a_i\rceil$ consecutive days. Given an instance $(a_1,a_2,\dots,a_n)$, its dimension is $n$, its density is $\sum_{i=1}^n1/a_i$, and its potential is $\sum_{i=1}^{n-1}1/a_i$. We prove that every real-valued instance of density at most 5/6 admits a periodic schedule of period length less than $4\cdot2^n$. For integer periods, the 5/6 density bound was a long-standing conjecture settled by Kawamura using a computer-assisted proof [STOC 2024, PNAS 2026]. Our result confirms the real-period conjecture of Fujiwara, Miyagi, and Ouchi [DMTCS 2026]. Our computer-assisted proof builds on Kawamura's ideas while using fewer explicit schedules. It combines dimension reduction by folding with nine satisfiability modulo theories (SMT) checks over linear real arithmetic. A set of 14287 explicit periodic schedules contains a schedule for every instance of dimension at most 8 and density at most 5/6, and for every instance of dimension 9 and potential at most 5/6. Folding is used to reduce every instance of larger dimension and potential at most 5/6 to an instance of dimension 9 and the same potential bound. We also prove that every real-valued instance with minimum period at least 4 and density at most 17/20=0.85 admits a periodic schedule of period length less than $19\cdot2^n$. The density bound 17/20 is tight and improves the 0.84-bound of Mishra, Rho, and Kleinberg [arXiv 2025] for integer periods. It also resolves the $a_1=4$ case of Kawamura's 2025 conjecture that integer-valued instances of density at most $1-(a_1-1)/(a_1(a_1+1))$ are schedulable.

0.20

Learning-Assisted Beamforming Optimization for Hybrid XL-RIS-Aided Communication

cs.IT ·2026-11-07 ·Awuni Mohammed Abdul-Razak, Kyungchun Lee
Abstract · 13 pages, 8 figures

Hybrid active-passive extremely large-scale reconfigurable intelligent surface (XL-RIS) architectures mitigate the double-hop path loss of passive RIS while enabling near-field spatial energy focusing, but impose a computationally intensive mixed discrete-continuous beamforming design problem under strict radio-frequency (RF) power constraints. This paper investigates a hybrid XL-RIS-assisted downlink multi-user multiple-input single-output (MU-MISO) system in the near-field regime. The joint optimization of the base station (BS) precoder, discrete XL-RIS phase shifts, and active-element amplification gains is addressed via a discrete coordinate ascent (DCA)-based alternating optimization (AO) framework, in which discrete variables are updated through local index-domain searches and the BS precoder is derived in closed form via a noise-aware minimum mean-square error (MMSE) criterion. To reduce the per-sample computational cost for XL-RIS apertures, a deep unfolding (DU) framework is proposed as a low-complexity alternative, unrolling a fixed number of Riemannian gradient ascent steps on the sum-rate objective and recovering the discrete phase configuration via quantized phasor projection, with one learnable step size per layer trained in a self-supervised manner. Numerical results under near-field spherical-wave channel models confirm that the DU framework achieves sum-rate performance close to the AO scheme with substantially lower and nearly constant runtime, and that both schemes consistently outperform manifold optimization under discrete hardware constraints.

0.19

New Classes of CCZ-Inequivalent APN Functions with Boomerang Uniformity One in Odd Characteristic: Differential and Boomerang Spectra

cs.IT cs.CR ·2026-11-07 ·Namhun Koo, Soonhak Kwon
Abstract

Let $q=p^n>3$ be an odd prime power and let $g:\F_q\to\F_q$ be an even perfect nonlinear (PN) function with $g(0)=0$. We construct almost perfect nonlinear (APN) functions $G_{λ,M}(x)=g(x)+M(x)+λ(x^{q-1}-1)$, where $λ\in\F_q^*$, $-λ\notin g(\F_q)$, and $M$ is an $\F_p$-linearized permutation polynomial. These one-point modifications of $g+M$ have boomerang uniformity one, the smallest possible value for APN functions in odd characteristic. We give complete differential and boomerang spectrum formulas and obtain closed-form evaluations for selected constructions based on Coulter--Matthews monomials and Dembowski--Ostrom polynomials. For $q>7$, comparison of differential spectra establishes CCZ-inequivalence to all APN monomial families listed in the paper and every APN binomial in the original Ness--Helleseth family or its extension by Zeng et al. We characterize CCZ-equivalence within our construction, showing that any two linearly inequivalent even PN functions over the same field yield CCZ-inequivalent APN functions. We obtain three pairwise CCZ-inequivalent APN functions with boomerang uniformity one over $\F_{p^{15\ell}}$ for every $\ell\ge1$, six over $\F_{p^{120r}}$ for every $r\ge1$, and unbounded numbers of such classes in characteristic three. We further prove that CCZ-equivalence and EA-equivalence coincide for functions at Hamming distance at most $m$ from a PN function, provided that $m(m+1)<q-q/p$. We also show that all our APN functions have the maximal possible algebraic degree $n(p-1)$. Finally, we exhibit both CCZ-equivalent and CCZ-inequivalent APN pairs with boomerang uniformity one at Hamming distance one.

0.17

Weighted Matroid Intersection under Minimum Rank Oracle

cs.DM math.COmath.OC ·2026-11-07 ·Taihei Oki
theory
Abstract

We consider matroid intersection under the minimum rank oracle model, in which two matroids are accessible only through an oracle that returns the minimum of the ranks of a given set in the two matroids. By Edmonds' theorem, this oracle determines the matroid intersection polytope, although it does not reveal the exchangeability graphs on which classical algorithms rely. Recent work by Bárász, Bérczi, Király, Oki, Yamaguchi, and Yokoi (2025) showed that unweighted matroid intersection is polynomially solvable in this model. In this paper, we present a strongly polynomial-time algorithm for weighted matroid intersection in this model. The key to our algorithm is a polynomial-size extended formulation of tangent cones of common base polytopes that can be constructed using the minimum rank oracle. We also give min-max theorems in terms of the minimum rank function via 1/3-submodularity, and present a strongly polynomial-time algorithm to compute an optimal dual solution.

0.14

Optimal Sampling and Transmission Policy for Information Freshness in Gossip Networks

cs.IT cs.NIeess.SP ·2026-11-07 ·Irtiza Hasan, Ahmed Arafa
theory
Abstract · Potential IEEE journal publication

We study the optimal sampling and transmission policies for a transmitter communicating with n gossiping receivers with the objective of tracking a source's status over time using a semantic freshness metric: Version Age of Information (VAoI). The transmitter observes the version changes in the source, and can acquire a new information version by paying a sampling cost. The receivers gossip, allowing information exchange among each other without direct transmission. Gossiping can save on transmission cost, but it may propagate stale version information, necessitating transmitter intervention. With imperfect communication links, we formulate an infinite horizon average cost Markov Decision Process (MDP) in order to jointly optimize receivers' VAoIs, sampling and transmission costs in a time slotted fully connected (FC) gossip network. We first solve the MDP using Relative Value Iteration (RVI) algorithm. Then, we prove that the sampling decision has a threshold structure in the transmitter VAoI. Further, under reliable gossiping links, we show that transmission and idling has an VAoI-difference threshold structure with minimum VAoI activation. Numerical results obtained using RVI show that the optimal policy consistently outperforms natural baselines. These results characterize how source sampling, direct transmission, and receiver gossip should be jointly coordinated for cost-aware semantic freshness in gossip networks.

0.13

Efficient SC-Flip Decoding for Polar Codes through Decision-Boundary Normalization

cs.IT ·2026-11-07 ·Gulinisha Yiming, Zhongjun Yang, Zhicheng Liu, Zhiming Ma, Xumin Ni +1 more
Abstract

Accurate first erroneous position (FEP) identification is key to improving the error-correction performance of successive cancellation flip (SCF) decoding. However, FEP identification remains challenging with a limited number of re-decoding attempts. To address this issue, we propose decision-boundary normalization (DBN) to improve FEP identification and develop a novel DBN-SCF decoding algorithm. DBN normalizes the magnitude of each decision log-likelihood ratio (LLR) by the standard deviation predicted by the Gaussian approximation (GA) for the corresponding bit-channel. This allows a single normalized threshold to be shared across bit-channels within each code configuration, giving it a consistent statistical meaning and avoiding per-channel calibration. DBN-SCF decoding selects candidate flip positions from the information set in the natural successive cancellation (SC) decoding order using a threshold, without sorting. Moreover, DBN-SCF decoding achieves better error-correction performance with a limited number of re-decoding attempts. It also requires lower average decoding complexity than existing SCF decoding methods at a comparable level of error-correction performance.

0.09

Winning Reviews, Losing Discovery? The Co-Evolutionary Game of AI Research and Peer Review

cs.GT econ.GN ·2026-11-07 ·Hongze Liu, Mengru Wang, Zhiyuan Zhang, Yao Liu
Abstract

Autonomous research systems can produce convincing manuscripts whose scientific value is difficult to verify, while automated reviewers can reward the appearance of completeness. We examine how this gap can become a self-reinforcing threat to the research community. In a two-population evolutionary game, authors choose between acceptance-oriented and exploratory research, and reviewers choose between proxy evaluation and careful assessment. Cheaper acceptance-oriented production can increase review pressure, making proxy assessment more attractive; proxy assessment can in turn strengthen the incentive to produce reviewer-compatible work. We derive conditions under which this feedback sustains a stable regime that rewards publication success while reducing the scientific value of accepted work. An analytical basin boundary explains why the same institutional rules can support either this regime or exploratory research with careful review. Numerical illustrations show how publication counts can conceal scientific loss, how faster adjustment can reinforce existing practices, and why partial reform may fail once the cycle is established. Careful audits and reviewer rewards can reverse the feedback, while temporary support can produce lasting recovery if it moves the community across the basin boundary. The model identifies a warning mechanism rather than forecasting a particular conference: a review system can remain operational and privately rewarding while becoming less effective at supporting discovery. Its institutional implication is to reward independently validated contributions and consequential checking, rather than accelerate paper production and scoring alone.

0.06

Robotics, Control & Signals 54

LSRP*: Scalable and Anytime Planning for Multi-Agent Path Finding with Asynchronous Actions

cs.RO cs.AIcs.MA ·2026-11-07 ·Dachuan Liu, Shuai Zhou, Shizhe Zhao, Zhongqiang Ren
publishedhighly cited
Abstract · 15 pages, 14 figures

Multi-Agent Path Finding (MAPF) seeks collision-free paths for multiple agents from their respective starting locations to their respective goal locations while minimizing path costs. Although many MAPF algorithms were developed, most of them rely on a common assumption on synchronized actions, where the actions of all agents start at the same time and always take a time unit. This assumption may limit the use of MAPF planners in practice. To get rid of this assumption, recently, an algorithm called Loosely Synchronized Rule-based Planning (LSRP) was proposed, which can find sub-optimal solutions for many agents. However, LSRP often finds poor quality solutions due to its unbounded sub-optimality. This paper develops a new anytime planner called LSRP* that can keep improving solution quality after the initial solution is obtained until the runtime budget depletes. We analyze the properties of LSRP* and test it against several baselines with up to 1000 agents on various maps. LSRP* can handle up to 25% more agents than LSRP and can reduce up to 40% of the solution cost found by LSRP.

0.61

Spatial Representation Distillation and Knowledge Routing for Vision-Language-Action Models

cs.RO ·2026-11-07 ·Jungin Park, Chaoran Zhu, Changjae Oh, Kwanghoon Sohn
top venuecode/data
Abstract · NeurIPS 2026 Camera-ready

Understanding three-dimensional physical structure has become essential for vision-language-action models (VLAs). However, existing approaches either rely on explicit geometric inputs at inference time or adopt a single visual pathway, potentially compromising the visual-semantic knowledge inherited from pre-trained models. We present SPARK-VLA, Spatial Representation distillation and Knowledge routing for VLAs, a preserve-specialize-route framework for spatially adapting VLAs without requiring external 3D inputs or experts at inference. SPARK-VLA consists of two key components: (1) spatial representation distillation, which transfers 3D understanding from a frozen 3D expert to a LoRA-adapted visual branch by aligning attention responses while preserving the original visual pathway, and (2) a lightweight knowledge router, which selectively forwards action-relevant visual tokens to the language model using routing supervision derived from action-loss gradients. Across LIBERO, LIBERO-Plus, CALVIN, and real-world manipulation, SPARK-VLA achieves strong performance with only a 0.5B language backbone and approximately 53M trainable parameters, providing an effective and parameter-efficient approach to 3D-aware VLA learning. Code is available at https://github.com/park-jungin/spark-vla.

0.48

REINS: A Self-Evolving Agent Harness for Real-Time Trajectory Planning

cs.RO cs.AI ·2026-11-07 ·Zhihong Cui, Hengyu Liu, Haoran Tang, Shijun Liu, Amir Taherkordi +1 more
top venuecode/data
Abstract · Accepted at NeurIPS 2026 (Main Track). 36 pages, 7 figures, 20 tables. Code: https://github.com/ZhihongCui/REINS

Agent harnesses -- runtime scaffolds that wrap an LLM's reasoning loop with tool dispatch, verification, and persistent memory -- have become the standard substrate for deploying LLMs as autonomous agents, as demonstrated by Claude Code and Codex in software engineering. In trajectory planning, however, this paradigm has not taken hold: existing LLM-based trajectory planning methods either keep the LLM in the decision loop, exceeding the millisecond-level control budget, or offload it to offline rules or training-time supervision -- leaving the runtime planner non-evolving. Intermediate hybrids still lack the verification-and-consolidation loop that makes harnesses reliable. We ask: what runtime harness would let an LLM planner meet real-time constraints while self-evolving into a deployable one? We propose REINS, a harness built on three LLM-planner couplings: LLM outputs are (i) grounded to a learned variable-level dynamics model, (ii) verified via calibrated predicates and forward rollout, and (iii) consolidated into an indexed rule memory that serves future scenes in O(log n) without re-invoking the LLM. The harness closes a self-evolution loop: familiar scenes resolve by memory lookup; novel scenes invoke the LLM, whose verified outputs enrich the memory. Without modifying LLM weights, REINS attains 98%+ compliance, nearly 50% collision reduction, and 1.9-4.6 ms latency on simulated and real-world benchmarks at the tactical decision layer, with LLM invocation rate monotonically declining as memory matures. Code: https://github.com/ZhihongCui/REINS

0.41

Shifted Interaction Functions: A Unified Framework for Imperfect Actuation and Sensing in Consensus Networks

eess.SY ·2026-11-07 ·Yang Xu, Enric Morella Violeta, Hector Garcia de Marina
Abstract · 12 pages, submitted to TCNS

Practical networked robotic systems rarely implement ideal consensus interactions exactly, since local actuation constraints, sensing biases, and heterogeneous interaction maps may shift the input-output relations used by distributed controllers. Rather than focusing solely on whether consensus occurs, such imperfections can fundamentally alter the structure of the asymptotic collective behavior, typically yielding global or clustered velocity alignment, while exact static agreement arises only in special cases. This paper develops a shifted-interaction framework for analyzing collective alignment under imperfect actuation and sensing. Shifted functions provide a compact representation of local actuator and sensor maps, allowing actuation constraints and sensing biases to be represented by horizontally and vertically shifted monotone interactions. The framework distinguishes nodal shifted actuation from edge-wise shifted sensing. For shifted actuation, a Lyapunov analysis establishes global velocity alignment under compatible actuation ranges, including bounded, non-invertible, and deadzone-type maps. For shifted sensing, a maximum-principle argument and a paired-sensitivity condition establish convergence to velocity clusters over edges with infinite accumulated sensitivity; uniform sensing sensitivity yields exponential global velocity alignment. In the linear case, closed-form expressions characterize asymptotic behavior and identify actuation as a special case of sensing. Simulations illustrate static, globally aligned, and clustered velocity-aligned behaviors.

0.40

An Interline Multiport DC Circuit Breaker with Power Flow Controller

eess.SY ·2026-11-07 ·Maliha Ferdous, Zhi Jin Zhang, Qichen Yang, Maryam Saeedifard
published
Abstract

Currently, the high-voltage multi-terminal DC (MTDC) grids face two major challenges. First, fast rise rate and lack of zero-crossing point in DC-side fault currents render AC circuit breakers inoperable. Second, DC line currents are passively determined by terminal DC voltages and line resistances, which inherently limits the capability of changing individual line currents for essential grid services such as power flow management and overloading prevention. To address these two limitations, DC circuit breakers (DCCBs) and DC power flow controllers (DCPFCs) have been developed. The conventional DCCBs are two-port devices installed in series with DC lines at their terminals in an MTDC converter substation, so that any fault current flowing through the line can be rapidly cleared. Additionally, a DCPFC can be installed at line terminals to adjust terminal voltages and power exchanges among adjacent lines. As a result, for a substation that connects to multiple lines, it needs to host multiple DCCBs and a DCPFC, which significantly increases the cost and footprint of both the substation and the MTDC grid. To this end, this work proposes an integrated equipment combining multiple DCCBs and a DCPFC, called the interline multiport DCCB with DCPFC, which can reduce the number of components and enhance the economic feasibility of MTDC networks. This paper presents the topology, operation principles, and component selection procedures of the proposed equipment, and compares it with the state-of-the-art technologies. Electromagnetic transient (EMT) simulation results validate the fault interruption and power flow control performance of the interline multiport DCCB with PFC.

0.38

Deep Reinforcement Learning with Massively Parallel Simulation for Low-Thrust Multi-Revolution Rendezvous

eess.SY ·2026-11-07 ·Chenglong Wu, Yi Zhou, Bin Cheng, Nan Zhang, Hexi Baoyin
Abstract · 23 pages, 8 figures, 13 tables

The growing demand for on-orbit servicing and constellation management in low Earth orbit requires autonomous and reliable low-thrust rendezvous. Traditional numerical methods struggle with the long-horizon nature of multi-revolution maneuvers under perturbations, while standard deep reinforcement learning (DRL) methods are limited by low simulation throughput and poor sample efficiency. To address these challenges, this paper proposes a massively parallel DRL guidance framework. A fully tensorized, GPU-resident orbital propagator is developed to integrate $J_2$-perturbed orbital dynamics and policy optimization within a zero-copy pipeline, achieving up to a 67-fold increase in simulation throughput over conventional CPU baselines. In addition, the rendezvous problem is formulated in modified equinoctial elements and trained with a progressive curriculum learning strategy to improve convergence and generalization across increasingly challenging rendezvous scenarios. Simulations show that the proposed framework can learn robust rendezvous trajectories for complex multi-revolution rendezvous over a broad range of orbital configurations. Additional evaluations under a high-fidelity model and eclipse-induced thrust interruptions further demonstrate the robustness of the learned policy under practical mission constraints. Moreover, the trajectories generated by the proposed framework can be used as effective initial guesses for optimization-based trajectory solvers, improving the convergence of high-precision trajectory optimization. The method offers a scalable approach for real-time low-thrust guidance and trajectory optimization.

0.36

Soft-then-Hard: Joint Link Association and Precoding Optimization for LEO Satellite Networks

eess.SP ·2026-11-07 ·Hyeongtak Yun, Wonjae Shin, Jeonghun Park
hot direction
Abstract · 13 pages, 8 figures

Spatial multiplexing in dense low Earth orbit (LEO) satellite downlinks is hindered by two bottlenecks. First, line-of-sight (LoS)-dominant propagation from a high orbital altitude places the users in a narrow angular region, which limits the spatial resolution achievable by a single satellite. Second, the visible satellites cause severe inter-satellite interference. Both effects call for joint link association and precoding, because the association decides which users are served together and which streams interfere. However, conventional block coordinate descent with hard association updates suffers from premature turning-off, where an inactive link loses precoder power and cannot be reactivated in subsequent iterations. To overcome this limitation, we propose a soft-then-hard update scheme that relaxes the association into continuous weights via iterative scaling, keeping candidate links active. Annealing progressively concentrates the association weights as the precoders are updated, after which a single hard matching determines the serving links. We further extend this framework to the LEO space-time channel, where distinct Doppler shifts provide user separability beyond angular geometry. Numerical results show that, over the space-time channel, the proposed design outperforms the baselines while effectively exploiting this separability.

0.36

Simulating Nonlinear Resistive Networks using a Behavioral Splitting

eess.SY math.OC ·2026-11-07 ·Ryan Alizadeh, Thomas Chaffey, Dominic Liao-McPherson, Alberto Padoan
Abstract · Submitted to European Control Conference 2027

An operator splitting method is developed for computing operating points of nonlinear resistive networks. Our behavioral formulation separates the device laws from Kirchhoff's laws and casts simulation as a zero-finding problem. Branch port resistances then provide a diagonal preconditioner for the relaxed Peaceman-Rachford iteration in scattering variables. Each iteration requires solving independent scalar equations and a linear system with a fixed weighted Laplacian matrix. Convergence is established for maximal monotone device laws when an operating point exists. Contractive branch reflections also yield a linear convergence bound independent of network topology. Numerical experiments on crossbar arrays illustrate the computational advantages.

0.35

Wireless Interference Recognition with Limited Data via Utilizing Diffusion Model

eess.SP ·2026-11-07 ·Rui Wang, Tong Li, Chanchan Zhu, Shaolong Gu, Zhu Xiao
Abstract

Accurate recognition of interference signals is essential for maintaining the reliability and security of modern wireless communication systems. However, existing approaches face several significant challenges: the high cost and the scarcity of accurately labeled datasets adaptable to diverse environments, coupled with the limited ability to evaluate uncertainty factors arising from propagation in complex electromagnetic environments. To address these challenges, this paper proposes a diffusion-driven model for wireless interference recognition. Specifically, we develop a composite data augmentation strategy that integrates geometric transformation and modal augmentation to enrich the limited data distribution. Geometric transformation expands scarce signal data, and modal augmentation generates semantically consistent multimodal data by modeling the correlations among cross-domain features. Following that, we propose a novel diffusion-based generative model with an interference classifier to accurately capture the distribution of interference signals within given signal data. It hence significantly improves the understanding of environmental uncertainty factors. Extensive experiments demonstrate that the proposed model outperforms state-of-the-art methods by over 5.84\% in wireless interference recognition.

0.35

A Bussgang-based Latent Lifting Method: Discrete RIS Channel Estimation and Beamforming

eess.SP ·2026-11-07 ·Mintaek Oh, Namyoon Lee, Jinseok Choi
Abstract · 13 pages, 8 figures

This paper addresses the challenge of discrete phase-shift optimization in reconfigurable intelligent surface (RIS)-aided multiuser communication systems. Practical RIS hardware often utilizes low-resolution phase shifters, leading to challenging combinatorial optimization problems. To address this challenge, we propose a Bussgang-based latent lifting (BuLL) method that represents discrete phase shifts as quantized outputs of continuous latent variables. Our BuLL method calibrates the Bussgang gain and distortion covariance under a reference distribution and keeps these moments fixed during deterministic latent optimization. These fixed moments are then used to construct a distortion-aware continuous surrogate that incorporates reference-averaged discretization effects without penalty terms. For uplink channel estimation, we develop a matrix quadratic transform-based algorithm to optimize a mean-square-error surrogate. For downlink beamforming, we develop a generalized power iteration-based RIS phase-shift optimization algorithm for a spectral efficiency (SE) surrogate. Numerical results demonstrate improvements in channel estimation accuracy and sum SE over conventional benchmarks in the evaluated low-resolution RIS settings.

0.34

Task-Critical Part Alignment Improves Imitation of Object Interactions

cs.RO cs.CV ·2026-11-07 ·Denis Hadjivelichkov, Sebastian Ægidius, Valerio Modugno, Dimitrios Kanoulas
published
Abstract · 6 pages, 6 figures. Accepted at ICAR 2025. This is the preprint version; the published version may differ slightly

This paper presents enhancements to the DINOBot framework for imitation learning of object-object interactions. We introduce two key contributions aimed at improving the accuracy and efficiency of alignment in imitation-based robotic tasks. First, we integrate RANdom SAmple Consensus (RANSAC) for outlier rejection during the alignment phase, effectively mitigating the impact of noisy depth data and spurious feature matches. This improves robustness when transferring demonstrations to real-world applications. Second, we refine the semantic correspondence process by focusing on task-relevant interaction regions, ensuring that the learned representation prioritizes critical object features that influence successful manipulation. We evaluated our approach in a range of real-world manipulation tasks, including picking, insertion, shaking, and pouring, using a robotic system. The experimental results demonstrate an overall improvement of 13.8% in task success rates compared to the baseline, along with faster convergence in the alignment process. These results highlight the effectiveness of our enhancements in improving precision and efficiency for object-object interaction imitation in robotics.

0.34

MoE-DAgger: Transition-Aware Router-Entropy Gating for Selective Human Correction in Phase-Structured Manipulation

cs.RO ·2026-11-07 ·Sungjun Heo, Sungjun Yang, Heedon Jeong, Junhyoung Ha, Jeong hwan Jeon
code/data
Abstract · 9 pages, 5 figures, 5 tables. Project page: https://SJ824.github.io/MoE-DAgger/

Selective data acquisition in Dataset Aggregation (DAgger) on robots requires deciding which learner rollouts to correct and when to intervene. In manipulation with ordered phases, router entropy from a phase-supervised soft mixture-of-experts (MoE) is insufficient: the same value can reflect either a normal adjacent-phase handoff or support for nonadjacent or earlier phases. We propose MoE-DAgger, which uses causal router history to distinguish these cases without providing the policy with a phase label, progress variable, or hard phase command during execution. The acquisition rule allows an adjacent transition to settle, but requests correction when high entropy persists outside such a transition, the transition times out, or routing persistently favors an earlier phase. Under matched active human-control budgets, MoE-DAgger achieves 84.0% task success on Green and 78.0% on Red, exceeding phase-unaware filtered entropy by 22 percentage points on Green and 26 points on Red, with no observed collisions in 50 rollouts per task. On 50 shared intervention-free traces per task, 83.3% and 80.0% of its first proposals fall on failed rollouts within predefined failure windows when 15 and 25 traces are selected, compared with 70.0% and 72.0% for temporal action-chunk disagreement and 63.3% and 60.0% for ensemble disagreement. Shadow replay and an active no-veto ablation indicate that suppressing queries at normal adjacent handoffs is a key contributor to the improved allocation.

0.34

Two-Timescale Trajectory Control, Task Reassignment and Resource Scheduling for Multi-UAV Cooperative Mobile Edge Computing

eess.SY ·2026-11-07 ·Bin Li, Shiyu Sun, Lei Liu, Zesong Fei
top venue
Abstract · 16 pages, 10 figures

Uncrewed aerial vehicle (UAV)-assisted mobile edge computing (MEC) is a promising technology, leveraging the advantages of UAVs including flexible deployment, wide coverage, and reliable wireless communication. In this paper, we study the task reassignment and offloading problem in a multi-UAV cooperative operating environment, and consider the real occurrence that the computing tasks arrive more frequently than UAVs make flight decisions. In addition, we also take into account the position offset of the UAV during flight and the unavailability of the UAV due to external interference. We propose a two-timescale optimization framework, where the trajectory of the UAV is carried out on a large timescale, while the task reassignment and computing resource allocation are jointly optimized on a small timescale. The core objective is to maximize the long-term average system gain, which takes both energy consumption and task delay into account. To address the UAV trajectory optimization problem under large timescale constraints, we propose a constrained multi-agent deep reinforcement learning algorithm. For decision-making within small timescale, to jointly optimize task offloading decisions and computing resource allocation, we integrate convex optimization with a heuristic algorithm. Finally, we conduct extensive simulations to evaluate the proposed algorithms against the baseline, achieving a 24% reduction in latency and a 28% reduction in total system cost.

0.33

Sparse Graph Neural Networks With Spectral Virtual Nodes for MIMO Detection

eess.SP ·2026-11-07 ·Fatemeh Jalili, Ghaya Rekaya-Ben Othman
Abstract · 6 pages, 4 figures, 1 table. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice,…

Graph neural network (GNN) detectors achieve high accuracy in large multiple-input multiple-output (MIMO) systems, but exchange messages between all pairs of transmitted symbols, so their complexity grows quadratically with the number of streams. We first propose a sparse GNN (SGNN) detector that restricts each symbol's messages to its most strongly coupled neighbors, so the message count grows linearly with the system size. However, pruning discards small couplings that jointly shape the weak channel directions, which limit detection accuracy. Even on the dense graph, learned messages capture these directions only partially. We therefore propose a sparse GNN with spectral virtual nodes (SVGNN), which filters the node states through a scaled regularized inverse of the channel Gram matrix to emphasize the weak directions. With two neighbors per symbol, SVGNN reduces the number of multiplications by a factor of 18 relative to the approximate message passing GNN (AMP-GNN) in a 64 x 64 MIMO system. Simulations with QPSK, 16-QAM, and 64-QAM over i.i.d. and correlated channels show that SVGNN matches or outperforms both AMP-GNN and expectation propagation, with gains of up to 3.2 dB, and lowers the error floors of AMP-GNN in several configurations. The virtual nodes improve AMP-GNN even without pruning, indicating that message passing alone, whether sparse or dense, cannot fully capture the weak directions.

0.32

MPSi: Neural Extraction of Site-Specific Multipath Propagation Properties for Massive MIMO

eess.SP ·2026-11-07 ·Maryam Rezvani, Raviraj Adve, Amr El-Keyi, Akram bin Sediq
Abstract · Submitted to IEEE for possible publication

Accurate extraction of site-specific multipath propagation properties enables a wide range of wireless applications, including initial access, wideband channel reconstruction, and the generation of channel datasets for learning-based and continual-learning systems. Classical parametric channel-sounding algorithms such as RIMAX provide physically interpretable estimates, but their data-dependent model-order selection and iterative nonlinear refinement introduce substantial delay and computational cost. This paper proposes MPSi, a physics-guided neural framework that replaces these costly stages with a compact path-count network and a fixed-depth, deep-unfolded Levenberg-Marquardt refinement. MPSi estimates the complex gain, angle of arrival, and propagation delay of each multipath component while preserving the underlying physical channel model. We evaluate MPSi in two site-specific ray-traced environments, a small cell and a dense urban area, and show that it achieves initial-access and full-channel-estimation performance comparable to an enhanced RIMAX baseline while requiring significantly fewer optimization iterations. The reconstructed channels also yield bit-error-rate performance close to the perfect-channel knowledge reference. These results position MPSi as a promising low-overhead framework for extracting multipath propagation properties in massive-MIMO systems and for generating cell-specific channel pseudo-labels for training when direct ground-truth measurements are unavailable.

0.31

A Nested Heterogeneous Multiscale Framework for Power System Dynamic Simulation

eess.SY ·2026-11-07 ·Kaiyang Huang, Min Xiong, Yang Liu, Kai Sun
Abstract · This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may…

Power system dynamics involve multiple timescales, from electromagnetic transients of the network and inverter-based resources to electromechanical oscillations of synchronous generators and quasi-steady-state evolution of the operating point. Simulating these dynamics with a detailed electromagnetic transient model is computationally prohibitive, whereas existing methods reduce the model to focus on a single timescale and lose accuracy when dynamics of different timescales interact. This paper proposes a nested heterogeneous multiscale framework for highly efficient multi-timescale simulation of a power system represented by one unified detailed model. Rather than varying only the integration step, the framework adaptively shifts the simulation resolution among three nested levels so that the dynamics of interest at each stage are always captured, with the slower dynamics estimated on the fly from short detailed simulations at controlled accuracy. Theoretical analysis proves the stability of long steps during quasi-steady-state evolution and quantifies the attainable speedup. On a modified IEEE 39-bus system with frequency regulation and load restoration, the framework accurately reproduces the dynamics of all timescales and runs 15 to 18 times faster than conventional simulation of the detailed model. The accuracy and speedup of the proposed method are further verified on a 390-bus system, demonstrating its scalability to large power systems.

0.31

Trajectory Inversion for a Rudderless Dual Vectored-Thrust Flying Wing

cs.RO ·2026-11-07 ·Weicheng Di, Qiqi Liu, Jinwu Xiang, Daochun Li, Zhan Tu
Abstract

A rudderless dual vectored-thrust flying wing combines vertical takeoff and landing with efficient for- ward flight in a compact tailless configuration. Without aerodynamic control surfaces, vectored thrust couples translation and pitch control. Therefore, generating dy- namically compatible trajectories remains an open chal- lenge. To address this challenge, this study proposes a trajectory generation method that jointly optimizes path geometry and timing. Trajectory inversion yields state and control histories that satisfy the coupled equations of motion. The optimization balances curvature-induced acceleration against aerodynamic support, lowering thrust- vectoring demand while retaining the defining maneuver features. Representative looping and rolling sequences are tracked in closed-loop simulations. These results provide a basis for planning autonomous aerobatic maneuvers with rudderless vectored-thrust aircraft.

0.31

Bisect-Index-Choose: A Fast and Exact Algorithm for EV Charging under Minimum Charging Rates

eess.SY ·2026-11-07 ·Can Berk Saner
Abstract · 10 pages, 6 figures. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after w…

Electric vehicle (EV) charging stations are increasingly operated with an aggregate power limit below the sum of their port ratings, and require scheduling algorithms to divide the available power among the connected EVs. Charging standards permit a current that is either zero or at least a minimum rate, typically 6 A, which makes this allocation problem non-convex. Existing methods either solve a mixed-integer program with a commercial solver, which can be computationally expensive, adds software dependencies, and incurs licensing costs, or apply heuristics without optimality guarantees. This paper proposes Bisect-Index-Choose (BIC), a fast and solver-free scheduling algorithm that computes a globally optimal allocation among EVs under mild assumptions on their utilities. For $n$ EVs, BIC runs in $O(n \log n)$ time when all EVs share one maximum rate and in $O(n^2)$ time for two, times a logarithmic factor in the numerical precision. In tests with up to 1000 EVs, BIC attains global optimality and is up to three orders of magnitude faster than the solver. On real-world charging data, it needs less capacity and delivers more energy than existing heuristics. We implement BIC on a low-cost microcontroller and show that it solves instances with 1000 EVs in less than 90 ms.

0.30

An Overview on Multimodal Wireless and Visual Sensing for Environment Reconstruction: Opportunities and Challenges

eess.SP ·2026-11-07 ·Guanlin Wu, Chao Hu, Yilong Chen, Yudong Guo, Rongfei Fan +3 more
survey
Abstract

Physical-scene reconstruction through visual sensing and channel knowledge map (CKM) reconstruction through wireless sensing have attracted growing research interest but have largely evolved independently within the computer vision and wireless communications communities, respectively. When applied to the same physical environment, the two sensing modalities provide complementary observations. Visual sensing captures detailed scene geometry and appearance but is constrained by occlusion and poor lighting. Wireless sensing can provide information beyond visual line of sight regardless of lighting conditions, but its measurements are often sparse and difficult to associate with scene structures. These complementary strengths motivate a unified multimodal approach to physical-scene and CKM reconstruction. This paper provides an overview of such unified environment reconstruction through wireless and visual sensing. We first compare the capabilities and limitations of the two modalities and discuss potential applications that benefit from jointly reconstructing both targets. We then review two complementary research directions, namely wireless-assisted physical-scene reconstruction and vision-assisted CKM reconstruction, highlighting how each modality supports the reconstruction task associated with the other. Building on these perspectives, we discuss design principles for a joint representation that supports both targets and identify mutual refinement, in which each modality helps improve the reconstruction associated with the other. Finally, we discuss open challenges and future research directions, offering insights into developing site-specific environment models for sixth-generation (6G) and beyond networks.

0.30

ReSkill: Selective Self-Improvement for Robot Manipulation

cs.RO ·2026-11-07 ·Zheng Sun, Zhihao Li, Zhuo Li, Wenrui Liu, Xin Wang +1 more
Abstract · Submitted to intelligent computing for embodied AI and robotics

Self-improvement on compositional manipulation tasks requires deciding what to reuse, where to improve, and which changes to retain. We present ReSkill, a selective self-improvement framework that concentrates revision on capability gaps. A coding agent constructs an object-centric task graph from an instruction and a skill library, then uses node-level execution records to locate failures. At the task level, it revises stage organization, hand-off conditions, and recovery routes; at the skill level, it strengthens perception, motion, manipulation, and verification capabilities that repeatedly fail. Candidate revisions undergo full-task evaluation and paired comparisons before adoption, allowing local changes to accumulate into validated skills and an independently executable policy. The agent participates only during development; the deployed policy calls the retained skills without invoking the agent at run time. On the same development set, average success across ten RoboDojo tasks rises from 30.0% for the initial versions to 86.2% for the final adopted programs; retained programs achieve 83.6% in recorded robustness and model-selection evaluations. Across eight real-world tasks, ReSkill succeeds in 72 of 80 rebuilt-simulation trials and 62 of 80 real-robot trials (77.5%). The median revision changes only three functions, approximately 0.5% of a program, while 39% of task-specific functions in later-added tasks directly reuse existing implementations. These results show how locating local deficiencies, validating candidate changes, and accumulating reusable skills can improve execution of compositional manipulation tasks.

0.29

Multi-Objective Adaptation After Deployment: A Position on World Models for Remote Robot Teams

cs.RO ·2026-11-07 ·Jamie Santos, Kagan Tumer
position
Abstract

Autonomous robot teams deployed in remote, unstructured environments may face changes in the environment, hardware, and team composition that degrade previously effective policies. Shifting mission objectives can be handled by selecting a different solution from an existing Pareto set, but shifting deployment conditions can reorder the set itself, making previously dominated policies competitive or new policies necessary. However, evaluating these policies in new contexts while minimizing risk to the hardware remains an open problem. We argue that, although largely untested in this domain, world models merit investigation for post-deployment multi-objective adaptation. By predicting policy outcomes under updated estimates of current conditions, world models could enable teams to re-evaluate existing policies without physically executing each candidate, provided predicted performance changes reflect those realized on deployed systems. We identify research priorities for model-based policy re-evaluation, validation of predicted against realized performance, and incorporating shared information into world models to support coordinated team adaptation.

0.27

Online Geometric Consistency Assessment of LiDAR Measurements via Recursive Bayesian Comparison

cs.RO eess.SY ·2026-11-07 ·Mingqiao Han, Qigeng Duan, Xuchen Liu, Jinqiang Cui, Xi Chen +1 more
Abstract

LiDAR-based localization and mapping systems produce registered point clouds that serve as geometric measurements for downstream estimation and navigation. Registration errors introduce duplicated surfaces and structural deformation into these measurements. In the absence of an independent reference, online assessment relies on geometry accumulated from earlier measurements. Such assessment requires tolerance to sampling variation and an accumulation process in which observations enter the history only after evaluation. We therefore formulate a recursive Bayesian comparison that couples geometric assessment with history maintenance. A Normal-Inverse-Wishart model represents historical voxel geometry and its uncertainty, while a weak isotropic reference adapts to current sampling. Comparing the marginal predictive support of the same current points under both models yields Normalized Drift Evidence (NDE). Assessment precedes assimilation: positive NDE triggers an alarm and preserves the preceding historical state; otherwise, the current observations update the history. Across 14 real-world sequences totaling approximately 73 min from four platform types, alarms identified all nine corrupted sequences and were absent from all five consistent sequences. In four representative scenarios with event references, first alarms occurred within 0.9 s of the reference times, with no preceding alarms. Cross-backend evaluation demonstrates consistent responses to reconstructed geometry. Ablation and controlled perturbation results establish the roles of uncertainty modeling, recursive history, displacement, and geometric support. Mean processing time is 4.2--6.1 ms per update at 1 Hz. In ten UAV trials, alarms initiated odometry isolation, landing, relocalization, and mission continuation.

0.27

RoboMoCap: A Large-Scale Whole-Body Motion Capture Dataset with Egocentric Vision for Robot Learning

cs.RO ·2026-11-07 ·Mayur Deshmukh, Aaron Tung, Junyi Li, Nikash Bhardwaj, Ademi Adeniji +3 more
benchmarkdatasetcode/data
Abstract · Project page: https://bitrobot-foundation.github.io/robomocap/

Humanoid robots must coordinate locomotion and manipulation to navigate complex environments and perform everyday tasks. Learning such behaviours from human demonstrations requires accurate whole-body motion paired with egocentric vision, particularly for lower-body movements and foot placements that are frequently invisible from a first-person perspective. However, existing egocentric datasets often lack high-fidelity optical motion capture or systematic coverage of diverse loco-manipulation behaviours. To address this gap, we introduce RoboMoCap, a large-scale multimodal dataset combining egocentric vision with marker-based optical motion capture, specifically designed for locomotion-centric humanoid learning. RoboMoCap comprises over 200 hours of recordings from 14 participants, featuring six head-mounted cameras, wrist-mounted visual and inertial sensing, and optical full-body motion capture at 120 Hz. The dataset systematically covers diverse locomotion primitives, dynamic whole-body movements, and loco-manipulation tasks, including variable-speed walking, turning, constrained navigation, terrain traversal, object transportation, and manipulation during locomotion. It provides temporally aligned multimodal observations and metric whole-body motion representations, enabling the study of human movement beyond directly observable body regions. We systematically evaluate existing methods on three benchmark tasks: egocentric full-body motion reconstruction, foot contact and placement estimation, and human-to-humanoid motion retargeting. By pairing egocentric vision with high-fidelity human motion across diverse locomotion and interaction scenarios, RoboMoCap provides a foundation for egocentric motion understanding and humanoid learning from human demonstrations.

0.27

On Real Eventual Exponential Positive property of complex-valued Laplacians and their pseudoinverses

eess.SY ·2026-11-07 ·Aditi Saxena, Twinkle Tripathy, Rajasekhar Anguluri
Abstract

The property of real eventual exponential positivity guarantees the state trajectories of a linear time-invariant system to evolve asymptotically to right half of the complex-plane [1]. It has applications in the characterization of herdability and reachability of states in networked systems. Networks with complex-valued edge-weights arise in AC power systems and opinion dynamics. Recent work has introduced real eventual exponential positive (rEEP) Laplacians, for which the trajectories of the associated network flow eventually converge to consensus. We establish further properties of rEEP Laplacians and their pseudoinverses. Firstly, we show that the rEEP property for complex-valued matrices exhibits invariance under specific rotational transformations. Secondly, we establish that the property of rEEP is preserved under the MoorePenrose pseudoinverse operation, highlighting its robustness under algebraic manipulations. These results apply to effective impedance computation in AC networks and to consensus in multi-agent systems. Finally, we provide numerical examples to validate our theoretical results.

0.27

Modified CNN-Learned Surrogate Covariance With Krylov/Lanczos Subspace for DoA Estimation

eess.SP eess.SY ·2026-11-07 ·R. Brandt, F. Rangel de Sousa, V. J. Brusamarello
Abstract

Recently, learning-based approaches have been employed to improve the robustness of subspace methods in DoA estimation. A Convolutional Neural Network (CNN) was used as a data-driven framework employing an autoencoder to generate a well-conditioned surrogate covariance optimized end-to-end through a differentiable classical estimator, such as the Root-MUSIC. The complete system is designed to estimate the directions of arrival based on a spectral decomposition using the EVD of this covariance matrix. In this work, a new approach is proposed by formulating the spectral stage of the Root-MUSIC algorithm in terms of the orthogonal complement of the signal subspace, thereby eliminating the need for the noise subspace and noise eigenvalues. Furthermore, the full EVD is replaced by a Lanczos tridiagonalization with reorthogonalization, reducing the overall computational cost.

0.26

Sampling-based Model Predictive Control as Sequential Monte Carlo Inference

cs.RO cs.LGeess.SY ·2026-11-07 ·Austin Garrett, Muyang Yan, Zachary Kingston
code/data
Abstract · 8 pages, 5 figures. Submitted to IEEE Robotics and Automation Letters (RA-L). Code: https://github.com/CoMMALab/PFMP

Sampling-based model predictive control (MPC) is treated as a family of distinct algorithms: predictive sampling, MPPI, the cross-entropy method, and their biased, colored-noise, and annealed variants all sample control sequences and reweight them by cost. We show a single Sequential Monte Carlo (SMC) inference algorithm implements them all, specified by three independent axes: the proposal, the update rule that reduces cost-weighted particles to a nominal plan, and the effective-sample-size threshold for resampling. We evaluate SMC-based planning on MuJoCo non-linear control, open-loop navigation through obstacle fields, 7-DoF motion planning, and non-prehensile push-T, in simulation and on a Franka FR3. Resampling combats particle degeneracy at long horizons. At a 28-knot horizon in our navigation domain it cut the collision rate by 30% over non-resampling baselines. Programmable proposals improve push-T success over perturbative baselines, improving from 1% to 100% in simulation and 10% to 90% on hardware.

0.26

From Discrete Snapshots to Geometric Entity: Motion-Native Channel Tracking for THz Near-Field XL-MIMO

eess.SP eess.ASeess.IVeess.SY ·2026-11-07 ·Yuheng Fan, Chuang Yang, Mugen Peng
Abstract

Channel tracking in mobile terahertz (THz) near-field extremely large-scale multiple-input multiple-output (XL-MIMO) systems is challenging because spherical-wave propagation couples angle and range, leading to rapid and nonlinear channel evolution under sparse pilot observations. Existing approaches typically estimate pilot-local channel or position snapshots before applying temporal tracking or smoothing, making them vulnerable to information loss during snapshot formation. To break this bottleneck, we propose a paradigm shift from tracking discrete snapshots to estimating a holistic geometric entity, namely, one shared continuous UE trajectory over the tracking window. Specifically, we parameterize the trajectory using a low-dimensional cubic B-spline representation and directly couple the per-pilot profile likelihoods through this shared trajectory before any pilot-local position estimate is finalized, thereby inducing continuous near-field channel evolution. We further characterize the efficient geometric information supplied by each pilot observation and its aggregation in the shared spline-coefficient space. Based on this formulation, we develop a regularized maximum a posteriori estimator and a quadratic-penalty lifted solver that separates parallel pilot-local geometric updates from a structured global spline update. The resulting global system is banded and, for a fixed spline degree, admits linear-complexity factorization, enabling an efficient structured implementation. Simulations demonstrate consistently improved tracking accuracy over representative temporally refined baselines in the tested medium-to-high-SNR regime, together with a favorable accuracy--runtime trade-off. These results highlight the benefit of coupling sparse pilot observations through a shared continuous trajectory rather than temporally processing independently estimated snapshots.

0.26

DreamVLA: From Generated Human Motions to Humanoid Vision-Language-Action Policies

cs.RO ·2026-11-07 ·Kevin Park, Izat Rasyad, Dvij Kalaria, Shankar Sastry
published
Abstract · Accepted to CoRL 2026 workshops G2A: From Generative Models to Robot Actions and Open Problems in Contact-Rich Loco-Manipulation

Training vision-language-action (VLA) policies typically depends on substantial robot demonstration data, making human teleoperation a costly and cumbersome bottleneck, particularly for whole-body humanoid interaction. Reinforcement learning in simulation offers an alternative for scaling this data, but designing task-specific rewards remains challenging. To address this, we propose DreamVLA, which uses generated motion trajectories to guide reinforcement learning (RL) in simulation, converting them into reliable task execution trajectories. These trajectories train VLA models in simulation without new task-specific teleoperation data. Our central contribution is a reference-conditioned DAgger distillation procedure: after a failed learner rollout, we restart the RL expert from a sampled learner state with the same reference suffix and aggregate successful corrective trajectories, addressing the distribution shift left by offline imitation. We evaluate picking from a platform, opening a drawer, pressing a button, and walking and picking from a platform, with ablations of DAgger iterations, residual RL, and generated dataset scale. Project website: ember-center-berkeley.github.io/DreamVLA.

0.25

Missed Interferers in Ray-Traced Radio Maps: A Digital-Twin and Stochastic Geometry Study of Aerial Coverage

eess.SP cs.PF ·2026-11-07 ·Saad Masrur, Ismail Guvenc
Abstract · Under Review

Ray-traced digital twins are increasingly used to plan where cellular-connected uncrewed aerial vehicles (UAVs) can fly. In this paper, we build a Sionna RT digital twin around Raleigh, NC, with 18,591 buildings and two overlapping sub-1 GHz operator networks, and study the downlink signal-to-interference-plus-noise ratio (SINR) of UAVs at 40-300 m. We show that the Monte Carlo radio-map solver of Sionna RT misses distant line-of-sight (LOS) interferers on aerial planes. With $10^7$ rays per sector and 5 m cells, it therefore underestimates the coverage at low SINR thresholds and overestimates it at high thresholds, by up to 26 percentage points at $\pm$2 dB and 50-100 m. Using a LOS-aware deterministic engine validated with Sionna's path solver, we then develop a stochastic geometry model with strongest-sector association and own-site leakage that matches the engine without fitting. The analysis shows that the interference grows logarithmically with the modeled network radius, so a twin that omits distant LOS sites overestimates the coverage, and that altitude mainly reduces the coverage at high SINR thresholds. Finally, selecting the better of two operators raises the minimum SINR along the best corridor by 1.0-3.9 dB up to 100 m.

0.25

Intelligent Reflecting Surface for ISAC: From Beamforming to Joint Design With Surface Orientation

eess.SP ·2026-11-07 ·Qiaoyan Peng, Qingqing Wu, Guangji Chen, Yong Zeng, Sai Huang +3 more
Abstract

Rotatable semi-passive intelligent reflecting surfaces (SIRSs) provide an additional spatial degree of freedom (DoF) by proactively adjusting the orientation, while dedicated sensors directly receive target echoes. To exploit this DoF for integrated sensing and communication (ISAC), we investigate the joint optimization of SIRS rotation, base station (BS) transmit beamforming, and SIRS phase-shifts to minimize the Cramér-Rao bound (CRB) for target's spatial-frequency estimation subject to per-user communication signal-to-interference-plus-noise ratio (SINR) constraints. We first establish an orientation-dependent system model that characterizes the effective aperture gains, based on which the CRB is derived. To gain useful insights, closed-form expressions of the beamforming and rotation designs are derived in the sensing-only and communication-only cases. For the co-located user and target case, we further characterize the condition under which the sensing-oriented rotation design also satisfies the communication requirement. In the single-user single-target case with ideal unit aperture gains, we analyze the channel correlation and its impact on ISAC performance. For the general multi-user case, an effective method based on the genetic algorithm is developed. Numerical results demonstrate the superiorities of the rotatable SIRS over the fixed-orientation SIRS and show that jointly optimizing the SIRS orientation and beamforming can improve the effective aperture gains and the correlation between communication and sensing channels.

0.24

Tactile Sensors Are an Effective Substitute for Wrist Cameras in Contact-Rich or Occluded Manipulation

cs.RO ·2026-11-07 ·Aijia Yang, Arsh Tangri, Sergey Zakharov, Krishna Murthy Jatavallabhula
Abstract · Accepted at the 3rd Workshop on Dexterous Manipulation: Full-Stack Robotics on Learning and Design, CoRL 2026

Learned manipulation policies often rely on wrist-mounted cameras for close-range visual feedback that complements external views of the workspace. However, contact-rich manipulation requires establishing and maintaining contacts that can remain hidden even from wrist cameras. Tactile sensors mounted on fingers can reliably and directly measure contact, without needing visual access. In this work, we show that this feedback enables policies to recover the performance advantage of wrist cameras using scene and egocentric vision together with touch. Our approach incorporates two low-cost tactile arrays into a visuotactile ACT policy, without wrist images during training or deployment. We evaluate six real-world task families spanning placement, insertion, occluded retrieval, and sustained-contact tool use. On three contact-rich tasks, we independently vary wrist-camera and tactile availability while keeping the demonstrations, policy architecture, and optimization settings fixed. Adding wrist cameras improves success by 18.3-28.3 percentage points over vision without wrist images. Adding tactile sensing instead improves success by 31.7-60.0 percentage points and yields higher observed success than vision-only policies with wrist cameras on all three tasks, with significant advantages on two under Holm-corrected tests. These results support finger-mounted tactile sensing as an effective alternative to wrist cameras in the evaluated contact-rich settings.

0.24

CPNav: Calibrated Vision-Language Navigation for Object-Goal Search in Unseen Environments

cs.RO cs.AI ·2026-11-07 ·Qizhao Chen, Yuanhong Zeng, Shoh Nishino, Anushri Dixit
code/data
Abstract

Directing a robot to locate a target object in unseen environments remains a central challenge in embodied AI. Vision-Language Model (VLM)-based methods provide rich semantic reasoning but often produce uncalibrated and overconfident action decisions. We introduce CPNav, a navigation framework that combines collision-aware planning with statistically calibrated VLM action scoring for map-free object-goal navigation. From RGB-D input, CPNav constructs a navigability map, generates collision-free motion primitives, and queries two VLMs for goal detection and action scoring. The resulting scores are calibrated offline via conformal prediction (CP) using oracle actions obtained through a tree-structured exploration procedure with backtracking, providing finite-sample guarantees at a user-specified risk level. The CP layer is model-agnostic and can wrap any VLM policy. Online, calibrated VLM scores guide action selection, while a tree-structured exploration history records alternative actions and allows backtracking. We evaluate CPNav in both simulation and real-world hardware settings. In AI Habitat simulation, compared with the uncalibrated baseline method, CPNav achieves relative SR and SPL improvements of 6.2% and 11.4% on HM3D v0.2, and 12.4% and 8.8% on MP3D, respectively. In real-world deployment on a modified Hiwonder MentorPi robot, the corresponding improvements in SR and SPL are 24.9% and 15.2%, respectively. Code is available at https://github.com/practice-lab-ucla/CPnav.

0.24

Passivity-Guaranteed Variable Impedance and Compliance Control via Online Parameter Optimization

cs.RO eess.SY ·2026-11-07 ·Huseyin Tugcan Dinc, Ersin Daş, Nikolas Thelenberg, Christian Ott
Abstract · submitted to Control Engineering Practice

Variable impedance control enhances the interaction capability of robot manipulators by adapting impedance parameters online. However, time-varying impedance parameters may inject energy into the system and violate passivity, particularly when tasks demand rapid stiffness changes. Existing methods address this by increasing damping or limiting the stiffness profile, which often degrades the dynamic response. In this work, we propose a quadratic programming-based parameter optimization framework that guarantees passivity by modulating the desired inertia to compensate for energy injected by stiffness variations, while preserving a critically damped system. For the case of diagonal impedance matrices, we optimize the rates of change of inertia and stiffness subject to a passivity constraint formulated as a convex quadratic program. We extend the framework to compliance control, where the desired inertia is coupled to the robot's configuration-dependent inertia matrix. We use an eigendecomposition based on Cholesky factorization to optimize the spectral weights while maintaining passivity. Experiments on the Kinova Gen3 7-degree-of-freedom manipulator validate both methods with human interaction perturbations. Comparisons with an energy tank approach and a nonpassivity-based method demonstrate that the proposed framework achieves less conservative stiffness tracking without requiring an energy reservoir or damping overhead.

0.23

HSGNav: Hypothesis Scene Graphs for Multi-Scale Lifelong Navigation under Object Relocation

cs.RO cs.CV ·2026-11-07 ·Kaiyao Liao, Qingfeng Li, Chen Chen, Qing Sun, Ningbo Gu +4 more
Abstract · 11 pages, 4 figures, 4 tables

In real homes, assistive robots must repeatedly locate targets from large furniture to tiny tabletop items, as people move objects between household surfaces. Uniformly instantiating detections at different scales as 3D objects can be brittle for small on-surface items; without explicit object-state validation, stale entries may persist after scene changes. Our key insight is to separate persistent 3D furniture entities from revisable surface-object hypotheses, retaining their visual evidence without requiring independent object geometry. We formalize this design as the Hypothesis Scene Graph (HSG), which preserves small-object evidence and supports local revision without rebuilding the stable entity topology. Building on HSG, HSGNav uses this separation for multi-scale lifelong navigation: retrieval selects entity or hypothesis evidence according to target scale, exploration acquires additional surface observations, and verification confirms target presence or invalidates contradicted hypothesis records. Because existing lifelong benchmarks such as GOAT-Bench rarely require mid-episode memory invalidation or multi-scale search, we introduce FAROS-Bench (Find And Re-find Objects across Scales) for mid-episode relocation of surface objects and multi-scale find goals. Experiments show that HSGNav performs competitively on GOAT-Bench while substantially outperforming prior methods on FAROS-Bench.

0.23

Balancing Confidentiality and Attack Detectability in Networked Control Systems over Lossy Wireless Channels

eess.SY ·2026-11-07 ·Mahyar Mohammadi, Subhrakanti Dey, Arunava Naha
Abstract · To be Submitted to ECC27

We study the joint protection of a networked control system (NCS) against eavesdropping and data integrity attacks over a lossy wireless channel. The eavesdropper intercepts the transmitted sensor measurements and control inputs to reconstruct the system state. In contrast, the integrity attacker in this study replaces the sensor measurements with fabricated data to drive the system to undesired operating conditions. Although these two threats are typically studied separately, a wireless NCS should be protected against both simultaneously, and a defence against one affects the other. We employ an observation-encoding scheme that renders the eavesdropper's state-estimation error covariance unbounded under packet drops, and show that transmission energy governs how long this error remains bounded. In particular, lower transmission energy improves confidentiality. For integrity protection, we combine stochastic physical watermarking with a CUSUM-based sequential change detector. The proposed attack detection scheme accounts for intermittent communication and encoded measurements. The detection delay is governed by the Kullback--Leibler divergence (KLD) between the pre- and post-attack distributions of the test signal, and we show that a higher transmission energy increases the KLD and thereby reduces the detection delay. Furthermore, we derive analytical expressions for the eavesdropper's estimation error and the worst-case expected KLD, providing guidelines for selecting transmission energy to balance confidentiality and integrity protection and characterising the privacy--security trade-off. Finally, we run numerical simulations to validate our theoretical findings and demonstrate the privacy--security trade-off in practical scenarios.

0.23

Selective Cooperative Perception via Blind Spot Reasoning and Agentic Dialogue with Vision-Language Models

cs.RO cs.AIcs.CVcs.MA ·2026-11-07 ·Chengyuan Ma, Shixiao Liang, Hangyu Li, Zheng Li, Hang Zhou +4 more
Abstract

Real-time cooperative perception (CP) in multi-agent traffic environments is highly resource-intensive, which posing scalability challenges for large-scale deployment. Efficiency can be significantly improved by selectively performing CP only in safety-critical scenarios and for key occluding objects. Vision-Language Models (VLMs) are well-suited for complex reasoning in dynamic and uncertain environments, making them a promising tool for this selective CP task. This study proposes a VLM-driven agentic framework for blind spot reasoning and dialogue-based CP. A mixed-model architecture is designed that integrates conventional object detection for stable perception, vision-language reasoning for identifying impactful occlusions, and a physics-based geometric module to align reasoning with real-world visibility constraints. The framework also incorporates a Chain-of-Thought (CoT) prompting mechanism to enhance interpretability and intent-aware decision-making. We collect real-world data from connected vehicles operating in urban traffic and compare the driving-critical blind spot identification and dialogue performance of four state-of-the-art VLMs. Experimental results demonstrate that our method achieves over 80% accuracy in driving-critical blind spot detection and over 90% success in dialogue-driven perception sharing. This work provides a scalable, interpretable, and bandwidth-efficient alternative to conventional CP pipelines, and offers new insights into agentic communication design for future connected traffic systems.

0.22

Explicitly capturing demonstrator multi-modality for distillation into diffusion policies

cs.RO ·2026-11-07 ·Shao Yuan Chew Chia, Rebecca H. Jiang, Abhinav Agarwal, Andreas Hofmann, Leslie Pack Kaelbling +1 more
Abstract · Presented at Pretrain to Adapt Workshop at the 10th Conference on Robot Learning (CoRL 2026). Project website: https://shaoyuan.cc/research/…

The diversity of a distilled diffusion policy's outputs matters beyond how it impacts the policy's success rate. In particular, downstream applications like test-time search and policy improvement require exploration. Thus, we want to distill multi-modal demonstrators into diffusion policies which reproduce their conditional action distributions. Standard data collection for behavior cloning applies a sliding window to demonstration episodes, which pairs unique observation histories with actions. These datasets can be fit by a policy that is either multi-modal, or unimodal with high Lipschitz constant to capture a diversity of actions seen at nearby observation samples. Prior work finds that diffusion policies trained on sparse data learn the latter, memorizing one action per observation. We adapt vine sampling to behavior cloning data collection to resolve this ambiguity: at some states the demonstrator visits, query it $k$ times from exactly the same observation. On a synthetic task, we study how the number of actions per observation, $k$, affects the learned policy, and find that larger $k$ yields a policy that better matches the demonstrator's conditional distribution, including its multi-modality and diversity. On a 2D truck unloading task, vine sampling likewise leads to a policy closer to the demonstrator and substantially improves both best-of-$N$ and expert iteration performance.

0.22

Real-Time Threat Classification in Multi-Agent Systems via Topology-Coupled Hidden Markov Models

eess.SY ·2026-11-07 ·Ailin Barzegar, Davide Spinello, Mohammad Pirani
Abstract

Networked multi-agent systems coordinating in adversarial environments face compound cyber threats that alternate between false-data injection (FDI), denial-of-service (DoS), and stealthy decoy perturbations designed to mimic benign operational anomalies. Because the switching policy is latent and the plant dynamics evolve continuously, deterministic threshold detectors struggle to isolate deception without causing excessive false alarms. This paper presents an online, data-driven framework that couples distributed continuous time unknown input observers (UIOs) with a topology-aware Hidden Markov Model (HMM) for real-time attack detection and classification under unknown adversarial strategies. By decoupling FDI signals from augmented node states, distributed observers deployed across the network generate local augmented residuals that are discretized into emission sequences. The HMM embeds the underlying directed communication graph into the off-diagonal blocks of its transition probability matrix, explicitly capturing inter-agent attack propagation. Both transition and observation probabilities are updated recursively online via an Expectation-Maximization (EM) scheme, adapting the classifier to evolving threat behaviours without prior offline calibration. Numerical simulations on a directed network demonstrate that incorporating graph topology into the latent transition dynamics resolves the severe classification ambiguity between nominal and decoy modes, outperforming decoupled agent-independent decoders while restricting observer deployment to a sparse subset of nodes.

0.22

Which Taxels, Not How Many: Tactile Sensor Placement for Dexterous Policies

cs.RO ·2026-11-07 ·Taeyun Kim, Jinsoo Shin, Taehwa Kim, Taeyoon Kim, Jinyoung Choi
Abstract · Accepted at the Workshop on Everything Beneath the Policy, CoRL 2026. 8 pages, 5 figures, 12 tables

Policies for dexterous manipulation depend on design choices made before training: which sensors enter the observation, where they are placed, and how their readings are arranged for the encoder. These choices are rarely examined. For tactile sensing, simulation studies place taxels uniformly over the hand, and this placement has become a default without being tested. We test this default under a single controlled protocol. The protocol holds the teacher, the training procedure, and the candidate taxels fixed and varies only the sensor configuration and the representation. We evaluate it on in-palm rotation and grasping with three robot hands. A small set of taxels selected from contact statistics matches a far larger uniform set. Taxels at rarely contacted sites are equivalent to no tactile input. With all candidate taxels, their layout in the input tensor does not affect performance, even for a convolutional encoder. Uniform placement therefore overestimates the number of taxels a task requires. We distill these results into lessons for designing tactile policies and into a minimal set of design choices that a tactile policy study should report.

0.18

Bridging Semantics and Signals: An LLM-Driven MoE Framework for Radio Map Prediction

eess.SP ·2026-11-07 ·Qi Yang, Xinyue Jiang, Tong Li, Chanchan Zhu, Shaolong Gu +1 more
Abstract

High-fidelity radio map prediction is essential for wireless network optimization, yet existing single-network models frequently fail to generalize across diverse multi-band and multi-terrain scenarios due to rigid parameter constraints. We introduce LLM-MoE, a framework that employs a divide-and-conquer Mixture-of-Experts strategy to dynamically allocate modeling tasks to matched expert sub-networks. Our approach features a heterogeneous expert pool for characterizing varied physical propagation behaviors, combined with an LLM-driven semantic extractor that provides global physical awareness. A cross-modal fusion module integrates these macroscopic semantics with local spatial features for adaptive routing. Experimental results on the SpectrumNet dataset show that LLM-MoE outperforms top-tier baselines, achieving a 10.0\% reduction in Mean Squared Error (MSE) and 9.7\% in Normalized Mean Squared Error (NMSE), while ensuring robust generalization across heterogeneous environments.

0.18

SimDuoReal: Object-Centric Bimanual Policy for Emergent Dexterity

cs.RO ·2026-11-07 ·Qiyang Yan, Yulin Liu, Akash Sharma, Haozhi Qi, Guanya Shi +1 more
Abstract · Website: https://simduoreal.github.io

Bimanual dexterous manipulation requires coordinated arm and hand control, yet specifying such coordination manually is challenging and often task-specific. In this work, we explore whether complex bimanual dexterity can emerge from a simple object-centric learning objective. We train reinforcement learning policies under a shared object-centric reward, varying their training environments rather than tuning separate rewards for each behavior. We find that the policies discover pre-grasp manipulation, grasping and regrasping, in-hand adjustment and handover to reach object-pose goals without hand-designed motion sequences. We then refine the experts on transition states and use DAgger and reinforcement learning to distill them into a single policy that composes these behaviors for diverse bimanual tasks. On real hardware, the same policy succeeds in 61 of 80 trials (76.3\%) across four tasks with unseen objects and no task-specific fine-tuning. In simulated barcode scanning, our policy achieves 85.0\% success, compared with 60.4\% for two single-arm policies without handover. These results suggest that this objective provides a simple and scalable interface for learning general bimanual dexterity, shifting the burden from manually designing robot coordination to specifying what should happen to the object.

0.17

Coexistence of LEO Satellite Communication Systems under Regulatory Protection Constraints

eess.SP ·2026-11-07 ·Sunwoo Min, Daesik Hong, Jemin Lee, Ian P. Roberts
Abstract

This paper investigates the coexistence of two low Earth orbit (LEO) satellite communication systems sharing the same frequency band: a primary system with regulatory precedence over the spectrum and a secondary system that must not cause excessive interference to the primary system's ground users. This spectrum sharing paradigm was recently formalized by the United States Federal Communications Commission (FCC) through short-term and long-term protection criteria that limit primary-link unavailability and throughput degradation due to interference inflicted by the secondary system. To facilitate coexistence, we propose a satellite selection mechanism that associates the secondary system's satellites with clusters of ground cells at each handover to maximize secondary-user throughput while abiding by these regulatory constraints. Our mechanism enforces both constraints through a cumulative assessment that combines past primary-link performance with predicted performance over the upcoming handover interval. Simulation results show that our proposed method achieves up to 31% higher secondary-user throughput than prior work that imposes interference-to-noise ratio (INR) constraints, even when its constraints are tuned to satisfy the FCC's throughput-based protection criteria.

0.17

Deploying Physically and Epistemically Compliant Agents by Distilling Belief-Space Task and Motion Planners

cs.RO cs.AI ·2026-11-07 ·Phillip J. K. Christoffersen, Rachel Ma, Patrick Barragán, Dylan Hadfield-Menell
Abstract · 9 pages, 5 figures

In this work, we train agents to act under uncertainty according to specified, belief-dependent rules. Deep reinforcement learning (RL) can in theory train agents to follow rules if encoded into well-designed rewards, but in practice, RL fails to perfectly optimize these rewards, leading to ineffective or noncompliant agent behavior. Belief-space planners by construction produce behavior in line with developer specification, but require hand-programming online planners and belief-state estimators, adding significant engineering complexity at scale. We combine the two: we plan offline to generate a dataset of sample trajectories and then distill the plans into a policy that acts on raw observations. We propose two changes to planner distillation that address problems specific to history-based policies: Dual Averaging, a loss that combines per-timestep predictive loss and average loss on task changes, and $ρ$-Optimal Reinforcement, a feedback mechanism that reinforces the policy's own successful rollouts when these are near-optimal. We provide a sufficient condition under which static search problems with noiseless observation models is complete, and show that maximum-likelihood-observation planning implies this condition in static search with continuous beliefs. We create a challenging new domain, Scavenger, where an agent needs to gain information in order to succeed. We evaluate our distilled policy in two versions of Scavenger, one where information is provided visually and one where it is provided as a command in natural language. In either case, the learned policy attains a success rate above 90% in Scavenger, compared with at most 5% for RL. We also find that distilled planners transfer with greater than 60% success to following unseen and intentionally vague commands and greater than 75% success to following floorplans larger than those seen in training.

0.15

Grip, Trace, Write: Force-Aware Imitation and Reinforcement Learning for High-DoF Robotic Handwriting

cs.RO ·2026-11-07 ·Rongqian Chen, Yiding Chen, Jijie Zhou, Tom Z. Jiahao
Abstract · This paper is accepted by CoRL 2026 Workshop PhysWM

Most robotic handwriting rigidly mounts the pen and reduces writing to trajectory tracking. A freely held pen instead couples grasp stability, tip motion, and contact force. We present Grip, Trace, Write, a force-aware simulation environment and learning framework for high-DoF handwriting. A two-branch actor separates pen retention from arm-wrist writing while preserving their closed-loop coupling. Training learns a grasp controller, initializes writing from inverse-kinematics demonstrations by behavior cloning, and improves writing through reinforcement learning with rewards on ink, force, time, and retention. Across diverse tasks, the policy improves completion, precision, force regulation, and efficiency over behavior cloning and inverse kinematics, and generalizes without fine-tuning to unseen English and Chinese content. These results establish force-aware handwriting as a testbed for dexterous tool use.

0.15

EmbodiedIntentBench: Benchmarking Intent-Driven Runtime Adaptation for Mobile Robotic Missions

cs.RO cs.ARcs.CVcs.DC+1 ·2026-11-07 ·Rajat Bhattacharjya, Changwoon Lee, Hyeonjong Ju, Sing-Yao Wu, Hamidreza Alikhani +4 more
benchmark
Abstract · Paper is currently under review. Authors' version posted for personal use and not for redistribution

Mobile robotic sensing systems must adapt to operator requirements, limited resources, changing connectivity, and observation opportunities shaped by physical motion. Evaluating runtime strategies therefore requires assessing whether their decisions fulfill mission objectives, beyond isolated perception accuracy or execution efficiency. We present EmbodiedIntentBench, a benchmark for intent-driven runtime adaptation in mobile robotic missions. It represents mission intent through structured contracts for coverage, temporal evidence, reporting, urgency, and resources, and couples simulated unmanned aerial vehicle (UAV) workloads with causal, resource-accounted execution. Computation decisions affect mission time, resource use, and which observations can contribute evidence. We evaluate fixed and adaptive perception strategies in multi-phase missions and under changing urgency, connectivity, and battery conditions. In the multi-phase Depot study, a configuration with 87.5% full-bank target recall across visible target observations completes only 1 of 16 missions, whereas a lower-recall configuration completes 14. In an ablation, ignoring execution-induced observation loss increases aggregate mission success and reverses the ordering of two runtime strategies. Replays using Jetson AGX Orin CPU and GPU profiles change the relative success of local and remote strategies. In selected Jetson GPU-profile replays, intent-driven adaptation reduces modeled task energy by up to 28% with unchanged mission success. EmbodiedIntentBench provides an extensible framework for evaluating runtime decisions against mission fulfillment under physical and resource constraints.

0.15

Zero-Shot VLA Adaptation with Retrieved Guidance

cs.RO cs.AIcs.LG ·2026-11-07 ·Subin Kim, Anh-Quan Pham, Dinesh Jayaraman
Abstract

Pre-training large parametric "Vision-Language-Action" (VLA) models on large demonstration datasets has recently yielded promising results for general-purpose manipulation. However, these models still struggle under even small domain shifts relative to the training data, such as changes in camera viewpoint, background, object identity, or robot morphology. Motivated by the success of Retrieval-Augmented Generation (RAG) in language models, we pursue retrieval to boost VLA performance. Specifically, we seek to leverage retrieval from domain-mismatched, "in-the-wild" demonstration datasets like DROID to guide a VLA's behavior in a new target domain. To achieve this, we introduce RAGu, which overcomes robotics-specific challenges for RAG: (1) It incorporates simple geometric calibration to transform source demonstrations into the target setup and reduce cross-domain differences. (2) It uses a coarse-to-fine retrieval strategy that selects, from a large pool of pre-warped candidates, the trajectory most compatible with the base VLA. (3) It guides the VLA's flow-matching inference procedure rather than seeking to retrain the base model or expand its limited context window. We evaluate RAGu across real-world, simulation, sim-to-real transfer, and cross-embodiment humanoid manipulation. We find it improves real-world VLA success rates from 20% to 80% and outperforms the closest-related baselines in sim. Project website: raguvla.github.io

0.15

Burden, Bounty, or Bliss? Muscle-Aware Retargeting and Null-Space Exploration for Whole-Body Musculoskeletal Control

cs.RO ·2026-11-07 ·Ruiyang Kong, Takeshi Oishi
benchmark
Abstract · 9 pages, 4 figures, 4 tables. Submitted to IEEE ICRA 2027

Learning whole-body musculoskeletal control from human motion is important for developing human-like robots, assistive devices, and rehabilitation systems. This task presents two main challenges, namely reference trajectories that are kinematically plausible but difficult to reproduce through muscle-driven dynamics, and a highly redundant 700-dimensional action space in which unstructured exploration can easily disrupt the target motion. In this paper, we propose a motion-to-policy pipeline that exploits muscle redundancy in two complementary stages. First, muscle-aware retargeting (MAR) performs bounded adaptation of kinematic trajectories subject to hard semantic, anatomical, and contact constraints. We motivate the constraint bounds by analyzing how reference-trajectory artifacts affect muscle-driven tracking. Second, null-space exploration (NSE) estimates a one-control-frame mapping from muscle excitations to generalized forces and injects bounded, anatomically structured perturbations along its null space into the actions of a value-guided flow policy. On a 24-clip motion-capture benchmark, MAR and NSE together solve every reference, reach the full horizon with half the samples of the baseline pipeline, and lower joint-center error by 44% without late collapse. To the best of our knowledge, this is the first end-to-end pipeline that transforms monocular video into a trained whole-body imitation policy controlling 700 muscles.

0.13

Computationally Efficient Data Generation for Learning Nonlinear MPC Policies

eess.SY ·2026-11-07 ·Pietro Gori, Anton Pozharskiy, Franco Angelini, Manolo Garabini, Moritz Diehl +1 more
Abstract

Learning MPC requires large datasets generated by repeatedly solving MPC problems. For Nonlinear MPC (NMPC), dataset generation is computationally expensive because feasible initial states are difficult to sample efficiently. The feasible set is often non-convex or even disconnected, and many randomly sampled states are infeasible, leading to wasted optimization solves and unnecessary computational effort. In this work, we propose a data-generation framework for learning NMPC policies that combines active-set exploration with projection-based sampling. First, a bilevel formulation samples activity weights to promote solutions associated with different active sets while preserving optimality for the original NMPC problem; second, a simpler slack-based formulation uses weighted slack variables to target different active sets at the single level. These reference solutions are then used to construct local linearized feasible sets, from which candidate initial states are sampled and projected through the nonlinear NMPC problem to recover feasible and locally optimal training samples. Our approach is up to $37\times$ faster than baseline uniform sampling with $99\%$ acceptance rate, while maintaining a representative coverage of the feasible set.

0.13

GUARD: Anomaly-Informed Q-Steering for Generative Robot Policies in Dexterous Manipulation

cs.RO ·2026-11-07 ·Aryaman Gupta, Xiaoyue Wu, Rana Soltani Zarrin, Mark Van der Merwe
Abstract · 8 pages, 4 figures

Generative robot policies based on diffusion and flow matching can model complex, multimodal behaviors, but remain susceptible to covariate shift during closed-loop execution. Existing approaches mitigate such failures through additional demonstrations, policy fine-tuning, or test-time guidance, but often require human intervention, engineered rewards, or other task-specific signals. We introduce GUARD, an anomaly-informed Q-steering framework that leverages the policy's training distribution to construct a dense, failure-anticipatory steering signal. GUARD pairs sparse task rewards with dense anomaly scores learned from policy training data and propagates both through a learned Q-function. Its gradients then steer the denoising process toward actions that improve task return while preventing trajectories from entering poorly supported regions, without directly modifying the pretrained policy. Across four Adroit dexterous manipulation tasks, GUARD achieves average success rates of $96.0\%$ and $80.0\%$ under nominal and perturbed dynamics, outperforming the strongest baseline by $21.0$ and $20.0$ percentage points while maintaining the lowest anomaly score. GUARD further demonstrates direct sim-to-real transfer on in-hand screwdriver rotation and reorientation tasks, achieving $40$ and $25$ percentage-point gains in success rate. These results demonstrate that anomaly signals can serve not only for failure monitoring but also to complement sparse task rewards as effective supervision for robust policy steering.

0.12

BRACE: Belief-guided Residual Adaptive Control for Safe Navigation under Unknown Disturbances

cs.RO eess.SY ·2026-11-07 ·Jaeyoun Choi, Niklas Schmid, Oswin So, Augustinos Saravanos, Chuchu Fan
Abstract

Disturbances encountered during autonomous navigation generally vary across space, often discontinuously, as obstacles redirect wind and changing terrain alters traction. Tracking controllers typically assume sufficient control authority to mitigate disturbances and maintain safety; the regime where this assumption fails remains underexplored. We present Belief-guided Residual Adaptive Control (BRACE), a hierarchical reinforcement learning (RL) framework for reach-avoid navigation under unknown, bounded, spatially varying disturbances. BRACE infers a disturbance belief from model prediction errors and uses that to condition a high-level policy that adapts the reference trajectory. The policy combines a nominal and a belief-guided residual term, trained jointly with a likelihood-ratio advantage reweighting heuristic. We evaluate BRACE in three simulated quadcopter environments with abruptly switching wind fields, and validate it on Crazyflie hardware with fan-generated disturbances. Across these settings, BRACE improves reach-avoid performance over model predictive control and RL baselines.

0.12

Beating the Pool, Not the Game: A White-Box Equilibrium Agent for Language-Based Economic Games

cs.RO ·2026-11-07 ·Sudhanshu Srivastava, Faisal Aqlan
Abstract

We describe sudsho, a rule-based agent for the GLEE Competition that plays bargaining, negotiation, and persuasion with zero inference-time LLM calls. Games are scored as payoff percentiles against an empirical pool of prior play on the same configuration, so pool percentiles set our demand targets while clamped equilibrium shares bound demands and acceptance. Every move is a deterministic function of the visible state (persuasion mixing aside), so alignment is verified by replaying the post-freeze move log through the policy. Across 117,280 games the agent reached an overall rating of 2056 at $0 inference cost, and the account, ranked by its best agent (a same-policy replica), placed 26th of 71 ranked accounts. A replay against three frozen snapshots of the opponent pool credits the pool-aware clamp with 274 to 330 points of mean per-game bargaining rating, an effect that holds where logged opponent decisions fix the replay.

0.10

Model-Based Imitation Learning For Monocular Vision-Based Navigation

cs.RO ·2026-11-07 ·Basant Sharma, Prajyot Jadhav, Saqi Hussain Kalan, Arun Kumar Singh
Abstract · Under Review

Reliable navigation with a single monocular limited field-of-view camera is challenging because the lack of depth measurements and limited spatial awareness complicate collision checking. Existing approaches to tackle these challenges fall into two categories. Model-based planners can incorporate explicit collision-avoidance guidance which leads to better generalization but are often computationally expensive and require manual tuning of the algorithmic parameters. In contrast, end-to-end approaches trained by imitation learning can be flexible but typically encode safety only implicitly and thus can suffer from high collision rates in novel scenarios. In this paper, we present a middle ground through model-based imitation learning. The model-based part is a learned collision model that can map a sequence of past images and a candidate control sequence to a distribution of worst-case obstacle clearance. This predictive distribution in turn is used to define a collision-risk cost in a trajectory optimization formulation. The imitation component uses human demonstrations to learn scene-dependent balancing of collision risk with other task costs such as goal progress. We also use it to learn the algorithmic parameters (e.g., a warm start) of our optimizer to accelerate the solution process. We show that our approach can demonstrate highly reliable navigation in cluttered environments on robots with different embodiments. We also demonstrate a substantial reduction in collision rate over recent end-to-end and model-based baselines.

0.10

Detecting and Explaining Robot Policy Failures from Sparse Internal Features

cs.RO ·2026-11-07 ·Riad Ahmed, Sujosh Nag, Moniruzzaman Akash, Momotaz Begum
Abstract

Visuomotor policies can handle many robot manipulation tasks, but understanding their failures remains difficult. Unfamiliar conditions and small action errors can lead to failure, yet watching the robot alone tells us little about what went wrong inside the policy. We introduce MIND (Mechanistic Interpretability for Neural Diagnosis), a method that looks inside a trained policy to detect failures during execution and identify the internal concepts involved. MIND uses a sparse autoencoder to break down the policy's internal activations into features that can be studied individually. We give meaning to these features by linking their activity to the robot's position, movement, and surroundings, and by steering their activations to examine how they affect the robot's actions. From successful and failed executions, we learn how the features behave on the way to a failure. During execution, MIND flags deviations from these patterns and uses the meaningful features contributing most to the change to describe the problem. Experiments in simulation and on a physical robot show that changes in these internal features help detect and explain failures. By using the same features for detection and diagnosis, MIND connects failure signals to understandable changes inside the policy.

0.09

BAT-WAM: Bridging Action and Touch in World-Action Models for Dexterous Hands

cs.RO ·2026-11-07 ·Dong Fang, Feixuan Huang, Yuanxin Zhong, Yuxiang Gao, Qi Song +3 more
Abstract

Dexterous manipulation with world action models (WAMs) relies on hand configuration and contact information, yet joint-state and whole-hand tactile observations may be partially or entirely missing. RGB images provide scene context and hand-object geometry but are insufficient to reliably recover these missing modalities. We further observe a correlation between hand state and tactile information. Motivated by this observation, we introduce BAT-WAM, a framework that bridges visual, joint-state, and tactile representations for WAMs. BAT-WAM learns separate codebooks and decoders for the tactile and action streams. An RGB-conditioned continuous latent bridge connects the two streams for bidirectional learning, while visual information serves as continuous context. Under the HTBench evaluation protocol, BAT-WAM reduces the root mean squared error (RMSE) of tactile reconstruction by 34.73% under fully masked conditions and 38.84% under partially masked conditions, relative to an image-conditioned HandTouch checkpoint. For future tactile prediction conditioned on target-time RGB and historical actions, it reduces mean RMSE by 85.56% relative to OpenWAM with HandTouch. Real-robot experiments further show that BAT-WAM consistently outperforms related WAM baselines in dexterous manipulation. Generalization experiments on unseen objects and scenes also demonstrate substantial improvements over baseline models. These findings support joint modeling of visual context, hand state, and touch as a representation approach for multimodal WAMs.

0.09

People & Society 13

Do links to retracted papers predict future retractions? An out-of-sample test on the open biomedical citation graph

cs.SI ·2026-11-07 ·Hossein Mohasel Arjomandi
code/data
Abstract · 16 pages, 3 figures, 2 tables. Data: doi:10.5281/zenodo.23874477. Code: https://github.com/HosseinMA96/retracted-neighbours-open-data

Tools that screen manuscripts for research-integrity problems flag papers that cite retracted work, and retrospective studies report that such papers are retracted more often. This study tests the signal prospectively on open data. Predictors were computed on the January 2024 OpenCitations graph using the retractions known at that date, frozen, and compared with retractions in the Retraction Watch database from February 2024 to October 2026. The design was pre-specified. Among 10.5 million biomedical papers published 2015-2023, 2.1% cited a paper already retracted, and 4,323 were retracted later (0.041%). Citing a retracted paper raised the later retraction rate from 0.035% to 0.27% with one such reference; excluding reviews (post hoc), it rose to 2.3% with five or more. Within journal and publication year the odds ratio was 2.99 (95% CI 2.73-3.27), and 3.03 when reference-list length was also held fixed (post hoc), so the signal is not explained by retraction waves in journals or by long reference lists. As a screen it is weak: 0.9% of the 1,000 highest-ranked papers were retracted later (2.2% with reviews set aside, post hoc), and 84% of later retractions cite no retracted paper. When the time point was moved into the late-2023 retraction wave, the journal's own retraction rate became the far better screen. NIH-funded papers cited already-retracted work as often as other papers in the same journal and year. Links to retracted papers are a useful flag but not a screen on their own.

0.42

Can open citation data detect citation stacking? An audit against Clarivate's journal suppressions

cs.SI ·2026-11-07 ·Hossein Mohasel Arjomandi
code/data
Abstract · 18 pages, 7 figures, 2 tables. Submitted to Scientometrics. Data and code: https://github.com/HosseinMA96/citation-stacking-open-data

Clarivate suppresses journals from the Journal Citation Reports when self-citations or concentrated citations from another journal ("citation stacking") distort their Journal Impact Factor. Its data are closed. Yearly journal citation networks were built from the OpenCitations snapshot of January 2024 (1.96 billion citations; 85,437 journals) and compared with Clarivate's suppression lists for data years 2015-2023, collected from primary sources. Where reference lists are present, open data reproduce Clarivate's published self-citation shares within ten points for 36 of 50 journals. The failures follow publishers that do not deposit references with Crossref, which also hides several stacking donors. Biodiversity data downloads registered with DataCite inflate naive impact-factor numerators of taxonomy journals. The leading open method, CIDRE, detects none of 15 named Clarivate pairs: it drops large donor journals, and lowering its donor threshold merges the pairs into large groups. A recipient-centric reading of CIDRE's own test flags stacking-listed journals 6 to 14 times more often than other journals, depending on consensus level, and 5 to 7 times more often than size-matched controls, sometimes years before the listing. At equal flag rate it finds more listed journals than a plain 43% donor-share rule (10 vs 3 of 28), and about as many as a rule with a citation floor. A prospective check on the 2024-2025 lists agrees in direction but is not significant (one of six). Flagged pairs show more citations per citing paper and weaker topical similarity than matched pairs. Open audits are feasible if coverage gaps are reported.

0.39

The one-step surface-air lag in the EnergyPlus zone heat balance: order of accuracy, correction and consequences

cs.CE ·2026-11-07 ·Wooyoung Jung
unusual
Abstract · 24 pages, 7 figures, 10 tables. HUBS Lab Paper No. 2026-010. Code and data: doi:10.5281/zenodo.24041718

The order of accuracy sets how fast numerical error falls as the timestep shortens. This study measured that order for the zone heat balance of EnergyPlus 26.2 against exact or converged references and tested a correction on 45 U.S. Department of Energy prototype cases. All three zone air formulas, the third-order default included, give first order or lower, because the surfaces use the previous timestep's air temperature. Solving surfaces, air and HVAC to agreement within each timestep raises the default to about second order, and in a room with an exact solution two extra passes cut the error eightfold at 1.65 times the runtime. In the prototype cases, the correction changed annual site energy by at most 0.45% but an outpatient clinic's heating by 5.1%. Window-upgrade savings changed by 10% or more in 13 of the 34 applicable cases, about as often as under a change of timestep.

0.31

Worst-Case Contagion in Financial Networks: MILP Clearing and Robust Loss Analysis

cs.CE q-fin.RMeess.SY ·2026-11-07 ·G. C. Calafiore, M. Della Rossa, A. Proskurnikov
theory
Abstract

We study worst-case contagion losses in financial networks with interbank liabilities and common exposures to marketable assets. Existing linear-programming approaches provide exact stress-test values only within an insolvency-resilience margin, where all institutions remain solvent with respect to the external sector. We show that this margin is computed exactly by a finite family of linear programs, one per vertex of the uncertainty ball, whereas a single linear program based on the componentwise worst deterioration of external resources yields only a certified lower bound. We extend this framework beyond that margin in three directions. First, we derive an exact mixed-integer linear programming (MILP) formulation for the maximal clearing vector under arbitrary signed external net positions, providing a clearing oracle for robust stress testing when insolvency can occur. Second, we develop certified bounds on the worst-case loss beyond the margin, including insolvency-pattern enumeration, scenario lower bounds, fixed-pattern dual upper bounds, localized certificates, and a regime-generation procedure with monotone gap certificates. Third, we analyze the worst-case loss curve as a function of the perturbation radius. For $\ell_1$ and $\ell_\infty$ price-perturbation sets, the curve is convex, nondecreasing and piecewise linear up to the margin, with breakpoints generated by switches in worst-case shock directions, LP bases, and bank-level activity patterns. Beyond the margin, it remains nondecreasing and piecewise affine, but convexity may fail, the curve may jump upward, and the worst case need not be attained. The framework provides a certifiable MILP/LP methodology for parametric stress testing and interpretable diagnostics identifying the banks responsible for changes in systemic loss.

0.26

Who Benefits, and When? Trajectories of Academic Achievement and Interest in AI-Supported Learning

cs.HC ·2026-11-07 ·Hyejeong Lee
Abstract

AI-supported learning environments are increasingly used, yet relatively little is known about how students' achievement and interest develop over time or whether these trajectories vary across learners. This study examined longitudinal trajectories of mathematics achievement and interest in a semester-long AI-supported mathematics learning. Using a mixed-methods longitudinal design, repeated measures of mathematics achievement were collected at four time points and mathematics interest at three time points. Linear mixed models (LMMs) were used to examine overall changes over time, while linear quantile mixed models (LQMMs) were used to examine differences in trajectories across the lower, median, and upper portions of the outcome distributions. Mathematics achievement increased substantially during the intervention, with the largest gains emerging between the second and third measurement occasions. However, the trajectory differed across the achievement distribution. Mathematics interest increased significantly after the first month, but this increase was not sustained through the end of the semester. Qualitative findings further indicated that students experienced the AI platform differently. The study highlights the importance of examining both cognitive and motivational trajectories.

0.24

Early Signals of Problematic AI Use: The Longitudinal Roles of Motivation and Emotional Engagement

cs.HC cs.AIcs.CY ·2026-11-07 ·Jina Suh, Constance Hadley, Koustuv Saha
survey
Abstract

Longitudinal evidence about the psychological impact of conversational AI use remains scarce. We report a survey study of 114 adults who reported their level of problematic AI use one year after reporting their motivations (for escape, social, entertainment, and instrumental reasons) for engaging with AI. We found high levels of AI motivation of all types predicted later problematic AI use, especially when participants turned to AI for managing their feelings and overcoming difficulties. Our findings imply that AI motivation could be used as an early warning signal of potential problematic use, especially if individuals engage with AI for emotional purposes over time.

0.23

Safer for Whom? A Systematic Exploration of AI Safety in Eating Disorder Interactions

cs.HC cs.AI ·2026-11-07 ·Jinyuan Wang, Minh Duc Chu, Frances Ross, Ellen Fitzsimmons-Craft, Kristina Lerman
Abstract · 23 pages, 7 figures, 2 tables

AI assistants advise users about diet, exercise, and weight loss---areas in which unsafe guidance can worsen body dissatisfaction, dietary restriction, compulsive exercise, and other behaviors associated with eating disorders (EDs). Existing evaluations emphasize diagnostic accuracy, overtly harmful prompts, or synthetic scenarios, providing limited insight into whether models recognize implicit vulnerability in natural interactions or protect users consistently. We evaluate responses from three open-weight large language models to 100 questions sourced from an online pro-anorexia community. We grade responses using clinically informed criteria that extend a seven-dimensional taxonomy of ED-related harms. We also construct matched conditions in which questions are accompanied by information about the user varying race, gender, body mass index (BMI), and explicit ED vulnerability. Safety varies substantially across models and harm dimensions. High-risk responses are approximately twice as frequent without bios, suggesting that models do not consistently recognize vulnerability from language alone. Explicit signals of ED vulnerability improve safety by 10--14 points across models. Responses also differ significantly by race, gender, and BMI, with body size producing the largest disparities. These patterns are consistent with reliance on the stereotype that EDs primarily affect thin, White women, potentially leaving users outside this profile with less protective responses.

0.23

EgoVeil: Personalized Privacy Reasoning for Egocentric Video

cs.HC cs.CVcs.CY ·2026-11-07 ·Xue Wang, Yang Zhang
Abstract · 19 pages, 8 figures, 4 tables

Recent advances in vision-language models (VLMs) enable AI systems to interpret egocentric video. Yet continuous visual capture raises privacy concerns because sensitive content is pervasive and varies across users and contexts. We present EgoVeil, a personalized privacy reasoning system that determines which visual content to obscure based on individual preferences while retaining context useful for downstream assistance. It converts localized visual regions into structured semantic observations and combines a privacy reasoner with an explicit user-specific memory for personalized privacy reasoning. This design supports personalization and continual adaptation as preference evidence accumulates. We evaluate on in-the-wild data from 10 smart-glasses users under a cross-user chronological protocol. EgoVeil achieves 91.93% balanced accuracy and 84.42% Mask recall, outperforming direct VLM baselines and other controlled privacy reasoning baselines. Our results show that personalized, adaptive privacy reasoning better aligns privacy preservation with individual preferences while retaining the visual context needed for effective smart-glasses assistance.

0.21

Place and Political Behavior Online during the Los Angeles Blaze

cs.SI ·2026-11-07 ·Sofi Koponen, Ali Salloum, Luca Maria Aiello, Mikko Kivelä, Arttu Malkamäki
Abstract

Understanding how online communication during extreme weather crises enables or constrains climate mobilization is critical. During a crisis, the online world connects distant audiences and complicates the relationship between place and political behavior. We analyze all sociopolitical content from Bluesky about the January 2025 Los Angeles wildfire event, reporting how participation in collective action and allocation of attention vary by geographic distance. Participation varies nonlinearly with proximity, with local and international users participating more than regional and national users. Attention concentrates around regional and national users engaging primarily in viral, non-participatory political discussion. Analysis reveals a geographic mismatch between participation and attention, illuminating how online communication concentrates attention away from collective action amid an exacerbating climate crisis.

0.18

CompAIssionate Other: Identity Assumptions in LLM Responses to Distress

cs.HC cs.AIcs.CLcs.CY+1 ·2026-11-07 ·Drishti Goel, Ethan M. Nguyen, Violeta J. Rodriguez, Koustuv Saha
Abstract

Individuals experiencing suicidal ideation (SI) or psychological distress may disclose their social identities when seeking support from LLMs. Attending to identity-related stressors can be important for providing contextually appropriate and empathetic support. However, disclosing an identity does not necessarily mean that it is a cause of distress. In this work, we audit how LLMs interpret distress through disclosed identities. Using paired counterfactual interventions, we evaluate six LLMs across two SI and distress datasets, 16 identity and control conditions, and two prompting protocols. Our findings suggest that models can transform identity disclosures into explanatory narratives, attributing distress to presumed identity-related adversity. These assumptions frequently appear within otherwise empathetic responses, particularly when marginalized identities are disclosed. We contextualize our findings through the lens of compassionate othering to characterize these patterns as a form of harm in which empathetic responses impose identity-linked assumptions on individual suffering. These responses can misrepresent individuals' experiences by treating their identities as explanations for distress without supporting evidence. Our findings highlight the need to mitigate these harms by grounding interpretations of distress in individuals' disclosures and avoiding unsupported identity-linked assumptions.

0.16

Collaboration Signatures: Tracking Long-Horizon Dynamics of Cross-Meeting Collaboration Networks

cs.SI cs.CLcs.CYcs.HC ·2026-11-07 ·Shengqi Zhu, Sheshera Mysore, Bahareh Sarrafzadeh
Abstract

Enterprise platforms have received increasing attention as a core component of social networks and the Web. However, existing work mostly relies on fragmented logs of small units of work, like emails and chats, and focuses on who connects with whom without sufficient account of collaboration content and dynamics. This work systematically explores tracking collaborations across sequences of meetings as a new approach to interpreting their long-horizon dynamics. Linking 45,426 meetings in a large multinational technology company based on full transcripts and 1.6 million extracted agenda items, we establish a cross-meeting collaboration network of 6,065 collaborations. The network recovers the natural shapes of workplace collaborations: meetings, as snapshots of ongoing collaboration, link into largely tree-shaped structures, while the work content forms nested directed acyclic graphs (DAGs). Further, we explore how these networks evolve over time, examining three established perspectives: task division, team composition, and temporal continuity. Each explains part of the dynamics but does not capture the distinctness of long-horizon workplace collaboration: e.g., work flows in a consistent direction between a few pairs of subjects, yet a collaboration's subject explains little of how its meetings are conducted. Instead, accumulating the agenda, interaction, and language of meetings along each tree yields a signature of the collaboration, a distinctive profile that picks out the collaboration of a held-out meeting at 4.5 times chance. The signature models collaboration better than team or task views, individually or combined, in next-meeting prediction and tree-level fitting. We discuss the implications of these cross-meeting dynamics for designing office applications and support tools, and call for next-generation Web systems that represent collaborations over their long horizon.

0.15

The Market Is the Reward: Design Markets, Not Just Traders for Human-AI Collective Intelligence

cs.CE ·2026-11-07 ·Yankai Chen, Xueqing Peng, Alejandro Lopez-Lira, Yuehua Tang, Michał Dzieliński +4 more
position
Abstract

Learning agents are becoming market participants, yet machine learning still treats the market as a fixed environment, although its payoff rule is the reward agents learn from. This position paper argues that financial AI should build HyperMarkets for human-AI collective intelligence, and not only the agents that trade in them: markets whose liquidity provision, information production, outcome determination, and trading rule are performed or adapted by learning agents and designed jointly toward explicit market-level objectives. In a pre-registered simulation with Claude agents trading for principals, joint design raises a market-level objective by 79% over separately tuned functions, through a trading rule that helps only when slow traders are present, and how evidence is paid for decides whether a stronger agent earns more for its principal. We state when fixed rules or non-market aggregation should be preferred and what findings would refute the case.

0.11

Regularity in Cryptocurrency Market Cycles: A Halving-to-Halving Analysis Using Data from Google Trends and Wikipedia

cs.CE stat.APcs.SI ·2026-11-07 ·Marek Zatwarnicki, Piotr Stolarski, Włodzimierz Lewoniewski
Abstract

Bitcoin halvings provide a natural temporal structure for studying long-term dynamics of cryptocurrency markets. This paper presents a methodological framework for analyzing the regularity of Bitcoin and aggregate cryptocurrency market cycles by normalizing historical periods between consecutive halving events and projecting them onto a common relative-time scale expressed as the percentage of cycle completion. The proposed framework integrates market variables, namely Bitcoin price and aggregate cryptocurrency market capitalization, with measures of public attention derived from Google Trends search volume and Wikipedia pageviews. To obtain long-horizon daily attention indices, Google Trends data are reconstructed from overlapping 90-day windows linked by median scale ratios, following established approaches to combining high-frequency Google Trends samples. Rather than attempting to forecast future prices, the study aims to compare historical cycles, identify recurring temporal patterns, and evaluate the relationship between market dynamics and retail investor interest across different cycle-relative stages. Exploratory results point to a possible recurring structure. The maxima of the aggregate market capitalization in the two most recent completed cycles occurred at 39.0% and 37.9% of cycle completion, and the highest value of the ongoing cycle observed to date at 38.1% (534-547 days after the respective halvings), although the corresponding multiple of the halving-day value fell from 66 to 1.8. Public attention shows a similar compression of peak multiples, but not a similar timing. The paper also discusses current data limitations and outlines future research directions, including formal lead-lag analyses and expanded multi-asset investigations.

0.09