CS Paper Radar

Daily listing

2026-11-08

400 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 130

Uncertainty-Guided Reward Labeling for Reinforcement Learning under Limited Feedback

cs.LG ·2026-11-08 ·Renhao Zhang, Shreyas Chaudhari, Bruno Castro da Silva
top venue
Abstract · Accepted at NeurIPS 2026

In reinforcement learning from limited feedback (RLLF), only a small fraction of an offline dataset can be labeled with rewards, and the central question is which samples should be labeled to learn a high-performing policy from the resulting partially labeled dataset. Prior work formalized this as a reward-selection problem by focusing on the selection stage while treating downstream policy learning as a black box, in a regime where queried rewards are not retained for reward-model training. We instead study the retained-label setting, where queried rewards can be stored and used to fit a reward model before policy learning. We bound the suboptimality of the learned policy in terms of two sources of error: one arising from offline RL on a fixed, offline dataset, and the other from uncertainty in the reward model. Since reward selection cannot change the offline dataset, the limited labeling budget must be used to strategically reduce reward uncertainty. Motivated by RLLF's observation that useful rewards tend to keep the agent on high-return trajectories, we introduce Successor-Guided Uncertainty REduction (SURE), a novel method that uses successor features to select rewards that are both uncertain and associated with states from which high-valued states are reachable. We formally derive SURE from a bound-induced design objective and characterize its exact one-step marginal gain. Empirically, SURE reaches near full-feedback performance with few reward labels across a variety of domains and offline-RL backbones, yielding an effective method for feedback-efficient reinforcement learning.

0.53

Symplectic Normal Coordinates Flow for Generative Modeling of Hamiltonian Systems

cs.LG physics.acc-ph ·2026-11-08 ·Christopher M. Pierce
top venuecode/data
Abstract · Accepted at NeurIPS 2026. 10 pages, 6 figures plus appendices. Code: https://github.com/electronsandstuff/Normal-Coordinates-Flow

In conditioned generative models of physical systems, the densities at different values of the conditioning parameters are often related through symplectic transformations. For stable linear Hamiltonian systems, there exists a symplectic transformation to "normal coordinates" such that the dynamics is reduced to rotations in each phase-space plane. This picture extends to the nonlinear case where, away from resonances, the Birkhoff normal form again consists of rotations in phase space, now with amplitude dependence. In this work, we introduce an invertible symplectic flow layer consisting of amplitude-dependent rotations in learned normal coordinates. We derive the condition under which the nonlinear rotation layer is symplectic and detail an architecture that guarantees it. Experiments are carried out on three conditional generative modeling problems: a synthetic chain of coupled nonlinear oscillators and two problems from accelerator physics. We show empirically that normalizing flows with the new layers achieve competitive or superior negative log-likelihoods compared to baseline flows, with gains that grow with the strength of the nonlinearity on the synthetic system.

0.51

Boundary Mass in Feature Space as a Label-Budget Diagnostic for Representations

cs.LG cs.AI ·2026-11-08 ·Siming Zhang, Zhehui Shen, Shijie Chen, Yansen Yu, Hang Yu
top venuetheorycode/data
Abstract · Accepted to NeurIPS 2026

Low-label transfer is often decided by a practical question: after a representation looks useful, how many labels are still needed to learn the downstream boundary? Transferability and probe-quality scores characterize global alignment, likelihood, margins, and calibration. Boundary-near target mass complements these diagnostics by measuring the ambiguous regions where scarce labels are most costly. We propose effective boundary complexity, a label-budget diagnostic for frozen representations and linear probes. It measures mass in a thin strip around the learned boundary and normalizes it by class balance, so dense ambiguous regions and predicted-class imbalance are both counted. We prove that, under a smooth representation map, density change and tangential surface change cancel at first order, leaving boundary-normal stretch or compression as the governing factor. This yields a cached-feature estimator that fits a probe on one split and counts independent held-out features near its boundary. Experiments support the diagnostic at three levels. Controlled transformations match the predicted law across 74 maps with 0.38% median relative error. On real frozen features, normal-direction changes move both measured boundary mass and label demand, while tangent controls preserve the first-order coefficient. A differentiable surrogate reduces measured complexity by 16-36% at matched top-1 utility. Finally, CIFAR-100 and Tiny-ImageNet experiments support boundary mass as a second-stage diagnostic for near-tied representations. Five-way experiments show strong OvR-interface backbone-ranking agreement (Spearman 0.872), while acquisition experiments show high correlations between initial boundary mass and realized label budgets across three sampling policies (0.779-0.830). Code is available at https://github.com/qqaqq-code/boundary_mass.

0.51

Daily Emergency Call Volume Forecasting using Univariate and Multivariate TimeGPT

cs.LG ·2026-11-08 ·Feriel Fass, Djemel Ziou
Abstract · Keywords: Emergency calls; Prehospital emergency services; Time-series forecasting; Foundation models; TimeGPT; Univariate forecast; Multiva…

This study investigates the potential of TimeGPT, a pretrained foundation model for timeseries, for forecasting daily emergency call volumes. Real-world emergency call data, comprising 2,101 daily observations, are used to evaluate both univariate and multivariate forecasting settings under one-step-ahead and seven-day multi-step-ahead scenarios. The multivariate forecasting incorporates temporal, meteorological, demographic, and socioeconomic variables. A chronological training, validation, and test split is adopted to prevent information leakage, while hyperparameters are selected using the validation set through grid search and early stopping when applicable. Forecasting performance is evaluated using multiple error metrics. Its forecasting performance is compared with several established forecasting approaches, including Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors (SARIMAX), Poisson regression (PR), Negative Binomial regression (NBR), Extreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM). The results show that SARIMAX achieves the best overall performance in the univariate setting. In the multivariate setting, NBR and PR provide best performance. These findings indicate that the large-scale pretraining of TimeGPT do not necessarily translate into superior forecasting accuracy for emergency call volumes. Frugal methods trained solely on the available data remain better for this application.

0.50

Identifiable alignment of unpaired representations

cs.AI stat.MLcs.LGstat.ME ·2026-11-08 ·Stas Syrota, Albert Lopez I Serrano, Johanne Franck, Søren Hauberg
published
Abstract · 40th Conference on Neural Information Processing Systems (NeurIPS 2026)

Aligning two models without paired data requires a distance that both share, yet existing methods use Euclidean or spherical geometry. This is either an arbitrary choice or imposed by their loss function. We show that, under a precise form of the manifold hypothesis, the pullback of the data metric gives such a distance for any two models of one data-generating process. This holds whether the models are encoders, decoders or both, and whatever their architecture, loss, prior or training data. The alignment map preserves this distance and the distribution of representations, and is identifiable if and only if the data distribution breaks every symmetry of the data manifold. We show that this symmetry breaking holds generically. Identifiability makes the alignment coupling the unique optimum of Gromov--Wasserstein and of its weighted combinations with a linear program. The linear program alone has the same optimum if almost every data point has a distinct distribution of distances. The coupling, and with it the alignment map, is recovered exactly from permuted samples, even without identifiability, and consistently from independent ones. Our method, Neural Gromov--Monge Alignment (NGMA), uses no pairs or labels and outperforms every unpaired and label-informed baseline. It relies on pullback distances to align image models with different priors and single-cell models trained on disjoint cells, where Euclidean distances fail.

0.48

Decomposing Earth Embeddings with Sparse Autoencoders

cs.LG cs.CVphysics.geo-ph ·2026-11-08 ·Vitus Benson, Fanny Yang, Markus Reichstein
top venuecode/data
Abstract · Accepted at NeurIPS 2026. Code: https://github.com/EarthyScience/geosae

Earth observation foundation models produce global metre-resolution embeddings that underpin work across agriculture, ecology, hydrology, and beyond. Yet these embeddings are opaque: each dimension entangles unrelated physical concepts, which makes them difficult to interpret and their robustness hard to assess. Here, we propose sparse autoencoders (SAEs) as a principled way to decompose Earth embeddings into sparse, approximately monosemantic features over an overcomplete dictionary. Because LLM judges of interpretations do not transfer to satellite pixels, and spatial autocorrelation matters, we introduce a tailored evaluation protocol. It scores any SAE on reconstruction at matched sparsity, alignment with reference concepts, and spatial scale, each against baselines that include an SAE whose decoder is a frozen random dictionary. On this protocol, SAEs reconstruct AlphaEarth and Tessera embeddings better than every baseline at matched sparsity. Their features align with far fewer reference concepts than raw embedding dimensions (on AlphaEarth, 88% align with at most one of 20 concept groups, most with none, against a median of 12 per raw dimension). They also carry more of their variance at fine spatial scales (23 to 32% within a 640 m patch for BatchTopK and Matryoshka, against 9%). An SAE with a frozen random decoder shares both properties, which therefore follow from sparsity, whereas only a learned dictionary reconstructs the embedding accurately: on AlphaEarth, trained SAEs leave eight times less variance unexplained. We test the sparse code on two applications, discovering land-cover sub-concepts and detecting out-of-distribution pixels by area-of-applicability estimation. Both indicate how sparse features could be used, but neither is yet reliable enough for practical use. Together, these results are a step toward Earth observation models whose features can be inspected.

0.47

Discrete Diffusion Playground: A 2D Benchmark for Discrete Generative Models

cs.LG ·2026-11-08 ·Justin Shao, Sihyun Park, David Ryan Koes, Maria Chikina
top venuebenchmarkcode/data
Abstract · 18 pages, 7 main figures, 2 tables. Accepted at NeurIPS 2026 (Evaluations and Datasets Track)

Discrete diffusion is a promising framework for categorical generative modeling, but current evaluation protocols often fail to reveal whether models have actually learned the target distribution. We introduce the Discrete Diffusion Playground, a diagnostic benchmark that converts known 2D continuous distributions into binary token sequences through an exact discretized bijection. This creates a rare setting in which models are trained on discrete sequences, but generated samples can be decoded back into 2D and compared directly against a known ground-truth distribution. The benchmark spans smooth Gaussian mixtures and tilted checkerboards with sharp boundaries, disconnected support, and coordinate-dependent structure, allowing it to expose failures that are invisible in standard proxy metrics. It supports exact grid divergences, Wasserstein distances, density visualizations, and nonparametric two-sample tests, providing both quantitative evaluation and interpretable failure diagnosis. Using this testbed, we show that masked diffusion models can learn meaningful adaptive reveal orders but remain less stable than autoregressive baselines on smooth targets and largely fail on tilted checkerboards, where autoregressive models preserve the target geometry. Control experiments with reversed and interleaved token orderings show that autoregressive performance is not explained by favorable bit order alone. The Playground provides a simple but stringent benchmark for stress-testing discrete generative models and for developing evaluation metrics that detect real distributional mismatch rather than proxy performance. Our code is available at https://github.com/shaojus/DiscreteDiffusionPlayground .

0.46

InfraRSI: Harnessing Recursive Self-Improvement for Training Infrastructure

cs.AI ·2026-11-08 ·InfraRSI Team, Yuchuan Tian, Haocheng Mei, Ye Yuan, Chao Xu +6 more
code/datacommunity pick
Abstract · 22 pages, 5 figures, 8 tables. Project page and code: https://github.com/TokenRhythm/InfraRSI

While Recursive Self-Improvement (RSI) accelerates isolated algorithm and kernel synthesis, distributed LLM training depends on tightly coupled launch settings, operator kernels, and communication schedules, where single edits often hit out-of-memory (OOM) boundaries, compiler conflicts, measurement jitter, or silent numerical drift. We present InfraRSI, the agentic harness that turns open-ended human commands into verified training optimizations via a five-stage workflow: init locks editable scope and a fixed task contract over workload, loss, and timing invariants to prevent reward hacking, doctor profiles baseline phases and calibrates cluster noise, run co-optimizes launch knobs and code patches before re-verifying top candidates against the baseline and logging findings to a cross-hardware experience bank, and report and apply audit source paths, track failed-trial costs, and commit promoted changes. Rather than stepping greedily from a single incumbent, InfraRSI initializes from diverse experience-seeded configurations and fits global execution-time and censored peak-memory models across all trials, including OOMs. Using these models, it prices per-setting memory costs, revises miscalibrated LLM priors from live runs, and pairs memory-hungry optimizations with memory savers and verified code patches. Across four single- and multi-node dense and Mixture-of-Experts (MoE) LLM workloads, InfraRSI achieves a speedup of up to $2.75\times$ and lifts mean Model FLOPs Utilization (MFU) from $17.4\%$ to $30.0\%$, outperforming existing tuners and coding agents while transferring experience across GPUs. As it improves training infrastructure efficiency, InfraRSI forms a two-level Recursive Self-Improvement loop---revising priors online within runs and distilling experience across clusters---to push foundation model scaling.

0.44

A Hierarchical Tokenization Framework for Voxel-Level fMRI Representation Learning

cs.LG ·2026-11-08 ·Karanpartap Singh, Yue Zhao, Mohammad H. Abbasi, Ehsan Adeli
top venuecode/data
Abstract · Accepted to NeurIPS 2026. Code available at https://github.com/su-karanps/HBAR

We introduce HBAR (Hierarchical Brain Activity Representation), a hierarchical tokenization framework for learning compact, multi-scale representations of neuroimaging data. HBAR organizes voxel-level resting-state fMRI into a hierarchy spanning sub-parcel voxel groups, parcels, and large-scale functional networks, enabling up to 415x compression while preserving fine-grained spatial information beyond conventional parcellation. The framework supports fixed atlas-based, fully learned, and softly prior-guided hierarchies, allowing us to systematically study the role of structured priors in brain representation learning. On HCP-YA, at approximately 415x compression, HBAR achieves higher parcel and network correlations than rate-matched flat tokenizers, while at a lower 52x compression rate, shuffling the anatomical grouping reduces parcel correlation from 0.920 to 0.886. Demographic readouts with autoregressive training also reveal potential gains from the hierarchical structure. By combining sub-parcel inputs, compact discrete tokens, and multiscale brain structure, HBAR offers a practical framework for representation learning in high-dimensional neuroimaging data.

0.43

ALETHEIA: A Multi-Frequency Eddy Current Pulsed Thermography Dataset for Neural Operator Learning in Nondestructive Testing

cs.LG ·2026-11-08 ·Changbin Sun, Jie Xiao, Xiaotian Chen, Yuankai Wu
benchmarkdatasetcode/data
Abstract · 62 pages, 19 figures

Learning neural solvers for spatiotemporal partial differential equations (PDEs) under real-world constraints remains a key challenge in scientific machine learning, especially for inverse tasks with sparse and noisy boundary observations. We present the Aletheia dataset, the first 3D benchmark for learning data-driven solvers in the context of nondestructive testing (NDT). The dataset simulates eddy-current-induced heating in conductive solids and models the resulting transient heat propagation governed by the heat equation. Aletheia contains over 4,700 high-resolution samples across 10 excitation frequencies (1-100,kHz), each providing volumetric heat source and temperature fields over time. It supports both forward prediction of temperature evolution and inverse reconstruction of internal heat sources or defects from surface infrared measurements. Real infrared thermography data from cracked rail specimens are included for calibration and comparison. We define four canonical tasks on both regular and irregular grids and benchmark them using various neural operators. Aletheia establishes a unified platform for evaluating neural PDE solvers under realistic NDT conditions, enabling progress in reliable, data-driven inverse modeling. The dataset and code are publicly available at https://github.com/sherlock-0327/ALETHEIA.

0.42

One Table, Many Queries: Benchmarking Generalist Tabular Models Across Generation, Imputation, and Prediction

cs.LG ·2026-11-08 ·Yuanrui Wang, Kaifang Li, Xuanyue Li, Ziyun Li, Yan Lu +2 more
benchmarkcommunity pick
Abstract

Generation, imputation, and prediction query different aspects of the same tabular distribution, but are usually evaluated separately. This leaves unclear which capabilities they share, how their training objectives interact, and whether one model can perform all three well. We introduce the One Table Benchmark (\ours{}), spanning 56 datasets, 46 registered methods, and 20 capability axes under aligned data splits and explicit task contracts. We compare task ranks on aligned support, train five backbones under all seven non-empty objective combinations on five datasets with three seeds, and contrast fixed-model profiles with specialist portfolios. Rank comparisons connect imputation and prediction through numerical recovery and dependence preservation, and link generation to both through predictive utility; distributional fidelity follows different, sometimes opposing, rankings. Objective interventions show the largest pooled gains from including the recipient's own objective. Generation has the broadest coverage of positive pooled auxiliary effects, while prediction shows potential to improve imputation and synthetic-data utility, with responses that vary across backbones. Cross-addition effects reveal that objective benefits need not add and depend on the objectives already present. Fixed-model comparisons expose corresponding utility--fidelity trade-offs: some generalists surpass specialist portfolios on utility, while fidelity continues to favor specialization. Frozen foundation models show strong prediction and numerical point imputation, together with promising synthetic-data utility, highlighting their potential across all three tasks. These findings motivate unified tabular models that share useful conditional information while improving categorical reconstruction and faithful joint sampling.

0.42

Finite-Sample and Communication-Efficient Networked Information Aggregation

cs.LG cs.AIcs.GT ·2026-11-08 ·Mohammadhossein Bateni, Zahra Hadizadeh, MohammadTaghi Hajiaghayi, Mahdi JafariRaviz, Shayan Taherijam
top venue
Abstract · Accepted at the 40th Conference on Neural Information Processing Systems (NeurIPS 2026)

Building on SODA 2026 work Kearns et al. (2026), we study learning in directed acyclic graphs in which each agent observes only a subset of the input coordinates and may use the predictions of its parents as additional features. In the finite-sample protocol, the message sent along each edge is the agent's prediction vector on a shared sample, so the communication cost is linear in the sample size $m$. In this paper we study this model under communication constraints. First, we revisit the finite-sample analysis of the full-message protocol. We identify a gap in the finite-sample proof of Kearns et al. (2026) and develop a different argument that establishes a finite-sample population-risk guarantee along paths satisfying deterministic $M$-coverage. Second, we introduce a shared-sketch protocol in which each agent sends a $k$-dimensional linear sketch of its prediction vector instead of the full $m$-dimensional message, where $k\ll m$. We analyze this protocol through a reduction to least squares on the compressed sample together with an oblivious subspace embedding argument. The resulting bound separates statistics from communication: the statistical term is still governed by the original sample size $m$, while the additional error due to compression is controlled by the sketch dimension $k$. For Gaussian sketches, we obtain an explicit high-probability population-risk bound, showing that row subsampling is not the only possible communication-statistics tradeoff for networked information aggregation.

0.42

MCSA: Multi-Cutoff Spectral Aggregation in SO(3) Equivariant GNNs for Interatomic Potentials

cs.LG cs.AI ·2026-11-08 ·Junfu Tan, Mengfan Wu, Yu Zhu, Yu Shi, Yifang Qin +3 more
publishedcode/data
Abstract · NeurIPS 2026

Machine Learning Interatomic Potentials (MLIPs) based on Equivariant Graph Neural Networks (EGNNs) have achieved remarkable progress in atomistic modeling. However, accurately representing atomic interactions across different spatial scales remains challenging. In this work, we investigate the radial spectral properties of SO(3)-equivariant message passing and establish that the cutoff radius controls the characteristic spatial-frequency scale of radial interactions. Based on this analysis, we propose Multi-Cutoff Spectral Aggregation (MCSA), a plug-and-play module that aggregates radial kernels with different cutoff radii to enrich multi-scale geometric representations. To achieve efficient implementation, we exploit the compact support of radial kernels to fuse cutoff-dependent radial weights before angular tensor-product operations, avoiding redundant angular computations while preserving rotational equivariance. Extensive experiments across representative EGNN backbones demonstrate consistent improvements in energy, force, and stress prediction, as well as enhanced accuracy in higher-order properties, including phonon dispersion and lattice thermal conductivity. Further evaluations on layered materials with distinct interaction scales demonstrate the effectiveness of physically motivated cutoff selection. Code is available at: https://github.com/JF-Tan/MCSA.

0.42

Safeguarding Agentic AI in Radiology Workflows: A Multimodel Evaluation of Rule-based and Large Language Model-based Verification

cs.AI cs.CLcs.CY ·2026-11-08 ·Semil Eminovic, Robin Schmidt, Bogdan Levita, Maximilian Lindholz, Tobias Penzkofer +1 more
code/data
Abstract · 43 pages, 6 figures, 4 tables, supplementary material included. Tobias Penzkofer and Jawed Nawabi share last authorship. Code: https://githu…

Purpose: To estimate how often tool-using large language model (LLM) agents record clinically relevant false or unsupported information in simulated radiology documentation and how much safeguards applied before each action reduce these errors. Materials and Methods: In this simulation study (October 2026), 48 synthetic, radiologist-reviewed base scenarios covered finding communication, tumor board preparation and follow-up management. Forty-three also had a misleading-information variant (91 variants). Six LLM agents (Claude Sonnet 5.5, GPT-6.1-Sol, GLM-5.3, MiniMax M3, Gemma-4-31B, Qwen3.8-27B) completed each variant with permissions only, deterministic rules, and rules plus a locally run LLM checker (1638 episodes). The primary outcome was at least one executed high-severity violation per episode. Risk differences were estimated per agent at base-scenario level and, exploratorily, pooled across agents, with analyses of document access and blocks. Results: With permissions only, 7.3%-27.1% of episodes had a high-severity violation (pooled across agents, 19.4%). Pooled, it was 16.7% with rules (-2.8 percentage points; 95% CI: -6.1, 0.5) and 13.2% with rules plus the LLM checker (-6.2; 95% CI: -9.5, -3.0). Adding the LLM checker to rules changed it by -3.5 (95% CI: -6.3, -0.7). No difference between agents was detected (P = .43), and task success did not decrease detectably. In misleading variants, agents opened the altered document in 83% of episodes regardless of error, and 58% of erroneous statements cited it. Most blocked actions (64%-68%) would not have caused a violation and were returned for revision. Conclusion: Layered safeguards reduced but did not prevent high-severity errors. Errors after misleading information reflected misjudged rather than missed information.

0.41

Self-Supervised Collapse Prevention with Score-Based Entropy Maximization

cs.LG ·2026-11-08 ·Adarsh Kappiyath, Silpa Vadakkeeveetil Sreelatha, Xiatian Zhu, Yann LeCun, Lu Yin +1 more
theory
Abstract

Diffusion models have demonstrated the effectiveness of score-based objectives for generative modeling, yet their potential for self-supervised representation learning remains underexplored. We introduce Entropy-Based Regularization (EBReg), a framework that uses latent score estimation to regularize the latent distribution during self-supervised training. EBReg jointly learns an encoder and a latent score model, and utilizes the learned score to provide a diversity-preserving signal that prevents representational collapse. Motivated by a mutual-information lower bound, we formulate the objective as view alignment with smoothed marginal entropy maximization and optimize the marginal entropy without explicit density evaluation. The same score model can additionally be used at test time to refine corrupted representations toward the training latent distribution. Empirically, EBReg learns competitive representations on ImageNet-1K, achieving 75.57% linear-probe accuracy with ViT-L. Across diverse architectures, score-guided test-time adaptation further improves robustness on ImageNet-C, yielding an average relative accuracy gain of approximately 5%.

0.40

Weight-Space Programming of Variational Quantum Circuits: Large Language Models Write Quantum Circuit Parameters Without an Optimizer

cs.AI quant-ph ·2026-11-08 ·Debarshi Kundu, Swaroop Ghosh, Vasant G Honavar
Abstract

Variational quantum algorithms (VQAs) optimize circuit angles through repeated quantum-device evaluations, often facing vanishing gradients and high optimization costs. We ask whether a large language model (LLM) can write these angles directly. Extending *weight-space programming* from classical networks to quantum circuits, we provide a frontier LLM (\gpt) with the problem, circuit, and gate conventions. Its angles are executed exactly as written, without returning losses, gradients, or accuracy feedback. Across 139 requests, we test handwritten-digit classification, VQE ground-state estimation for spin chains and H$_2$, and MaxCut QAOA, comparing against random weights, classical optimization, and optimization initialized with LLM weights on the same circuit. When parameters have interpretable structure, the LLM matches or approaches the optimizer: it achieves 95--100\% of the energy spectrum on every VQE problem, derives the exact H$_2$ ground state in closed form, finds optimal depth-one QAOA angles for every graph, and reaches 93--96\% accuracy on digits 3 versus 8 with 12--18 hand-written weights, outperforming an optimizer trained on the same ten examples. Its weights are also strong warm starts: optimization reaches its best value after a median of one evaluation rather than hundreds. Allowed to design circuits, the LLM chooses data-dependent rotation chains encoding pixel importance in all 30 cases (one qubit for binary tasks, one chain per qubit for ten classes), reaching 65\% accuracy across ten digits. Yet the approach fails in opaque weight spaces: accuracy drops to chance when fixed circuits exceed 30 weights, while optimization improves. It is also sensitive to gate conventions: 6 of 42 QAOA replies reverse mixing-angle signs and fail. We release all prompts, replies, and scores.

0.40

MG-Cache: Multi-Granularity Caching for Accelerating Diffusion Transformer Inference

cs.LG cs.PF ·2026-11-08 ·Zesen Chen, Youxuan Xu, Shigang Li
publishedcode/data
Abstract · Published in Pacific Graphics 2026 (PG'26), https://doi.org/10.2312/pg.20261030

Diffusion Transformer (DiT) models, while having become a powerful backbone for generative synthesis, suffer from high inference cost on generating high-resolution images. Diffusion caching has emerged as a promising acceleration technique to reduce inference cost. However, existing caching strategies fail to accurately identify the varying importance of different computational modules and rarely coordinate cache reusing across the timestep and layer granularities, making it difficult to control error accumulation. To address these limitations, we propose multi-granularity caching (MG-Cache) for DiTs, a novel and fully training-free approach to exploit computational redundancy both on timesteps and Transformer layers. Our core innovation contains a two-stage caching strategy: an offline timestep profiling analysis that identifies redundant timesteps for cache reuse, followed by an adaptive layer-wise caching strategy that combines input stability with importance of each Transformer layer to selectively cache stable Attention and MLP submodules. Extensive experiments on mainstream open-source DiT models validate the effectiveness of our approach, achieving significant acceleration while maintaining high fidelity to the uncached output, with up to a 4.18x acceleration on HunyuanVideo.

0.40

Deterministic Narrative Gravity: Bridging Theory of Mind and Historical Simulation through ABM-LLM Architectures

cs.MA cs.AIcs.CL ·2026-11-08 ·Alicia Colmenero-Fernández, Lidia Ortega Alvarado
publishedbenchmarkcode/data
Abstract · 28 pages, 2 figures. To appear in the Journal of Computer Applications in Archaeology. Code: https://github.com/alixiacf/cognitive-abm

Historical social simulation faces a fundamental reproducibility challenge: the stochastic nature of large language models (LLMs) introduces a digital butterfly effect that prevents exact replication across multi-turn runs. Furthermore, the inability to control low-level server parameters in commercial LLM hosting environments severely limits reproducibility. To overcome this opacity, we present a deterministic multi-agent framework coupling LLMs to an Agent-Based Modelling (ABM) pipeline grounded in Theory of Mind (ToM) and Cognitive Archaeology, and validate exact artifact identity within local, pinned execution environments using SHA-256 hashing. Run locally at temperature zero with clean per-call model reloads, the system injects differentiated cognitive states per turn through dissonance-driven behavioural modes and gated internal voices. Evaluating seven model configurations spanning 4B to 31B parameters against an ablated baseline reveals clear operational boundaries. Small models ($\leq$8B) suffer from context capacity overload, producing repetitive, specular dialogue loops. However, at $\geq$12B parameters, the ABM scaffolding breaks the monotonous response patterns seen in ablation. Crucially, these scaffolded models exhibit spontaneous diegetic expansion: they coherently expand scenario lore through tactically motivated, anachronism-free details rather than escalating into aggression. By resisting the structural pull toward premature scene closure, scaffolded sessions sustain an open terminal Narrative Vector ($V_N = \mathrm{Open}$ in 12/21 overall; 9/9 in official $\geq$12B configurations) whereas unscaffolded ablation never does (0/21). This framework provides a transparent, auditable benchmark architecture for virtual archaeology, with the full pipeline released as open-source software to enable independent replication.

0.39

Spatial Harness: Grounding Spatial Reasoning in Executable World States with Physics Engines

cs.LG ·2026-11-08 ·Chentao Cao, Zhanke Zhou, Sangni Duan, Bo Han, Hang Li
publishedcode/data
Abstract · NeurIPS 2026

Spatial reasoning requires understanding how objects and observers are arranged in a scene and how their relations change with viewpoint or movement. Vision-language models (VLMs) typically perform both steps through model-internal reasoning, forming an implicit scene interpretation from partial views and imagining the consequences of spatial transformations. A misread scene can lead to a wrong answer, while even a correct interpretation does not ensure that imagined transformations satisfy spatial constraints. We introduce Spatial Harness, which represents scenes as explicit and executable world states to ground spatial reasoning in physics-engine feedback. A Perceiver integrates scene constraints extracted from the question with specialized visual estimates to place objects and cameras in a shared coordinate frame. The resulting state is instantiated in a physics engine, where a Reasoner executes viewpoint changes and uses computed measurements and rendered views to guide reasoning. We evaluate on MindCube, SAT, MMSI-Bench, and SpatialRGPT-Bench with open- and closed-weight VLMs. On MindCube, Spatial Harness improves Qwen3.5-27B from 52.70% to 71.14%, surpassing pySpatial by 7.52 percentage points. Given identical annotated object and camera poses, it reaches 98.25% versus 71.81% for model-internal reasoning. Code is publicly available at: https://github.com/tmlr-group/Spatial-Harness

0.39

A Trust Region Approach for Learning Schrödinger Bridges

cs.LG stat.ML ·2026-11-08 ·Takeshi Koshizuka, Denis Blessing, Max Zimmer, Sebastian Pokutta, Lorenz Richter
top venue
Abstract · Accepted to NeurIPS 2026; Takeshi Koshizuka and Lorenz Richter contributed equally

We study the numerical approximation of Schrödinger Bridges in the setting where samples from the target distribution are unavailable and only an unnormalized target density is given. In this regime, the widely used iterative proportional fitting (IPF) procedure (and its variants) often exhibits severe numerical instabilities, stemming from ill-conditioned underlying control problems and from the fact that the optimal solution may lie far from the initialization. To address these issues, we propose a trust region variant of IPF that constrains successive updates to remain at a fixed Kullback-Leibler divergence from the current iterate. This restriction significantly improves numerical stability. Moreover, we show that the resulting method admits an interpretation as a mirror descent scheme with an adaptively chosen step size. We demonstrate the effectiveness of our approach on several challenging sampling tasks.

0.38

Do Feature Interactions Break Explainability? Provable Explanations for Any-Order Neural Additive Models

cs.LG cs.AIcs.CC ·2026-11-08 ·Idan Refaeli, Yizhak Yisrael Elboher, Guy Katz, Shahaf Bassan
top venue
Abstract · Accepted at NeurIPS 2026. 36 pages, 2 figures

Post-hoc explanation methods for neural networks are often heuristic and lack provable guarantees. A common alternative to such methods that was studied in recent years is to compute a cardinally-minimal subset of input features that is provably sufficient to determine the model's prediction. For general neural networks, however, finding such explanations is computationally intractable. In contrast, recent work showed that this problem becomes more tractable for a particular family of neural networks, namely neural additive models (NAMs), though standard NAMs are highly restrictive because they exclude feature interactions altogether. In this work, we study the computation of provably minimal and sufficient explanations for neural networks with restricted feature interactions, bridging the gap between fully general neural networks and purely additive models. We first prove that even with constant-order interactions, the problem is $Σ_2^P$-hard, implying that worst-case exponential complexity is unavoidable. We then identify two provably tractable settings: one based on sparsity in the interaction structure and another based on a stronger interaction-aware notion of sufficiency. Under either of these settings, we develop algorithms that compute provable explanations for neural networks with any-order interactions while retaining much of the computational efficiency of additive models. Empirically, we show that these explanations are significantly smaller and faster to obtain than those produced by standard algorithms, while being derived from models that achieve much higher accuracy than standard NAMs. Overall, our results help expand the class of neural networks that admit efficient provable explanations, while clarifying the fundamental role of feature interactions in shaping their complexity.

0.38

When Self-Consistency Hurts: The Hidden Cost of Majority Voting in Small Language Models

cs.AI ·2026-11-08 ·Wanrong Yang, Lingfang Li, Jun Zhang, Dominik Wojtczak
publishedtheorycode/data
Abstract · 15 pages, 3 figures, 2 tables. Accepted by the 33rd International Conference on Neural Information Processing (ICONIP 2026)

We study prompted chain-of-thought (CoT) with self-consistency (SC), i.e. sampling N traces and plurality-voting, on small language models, a regime where per-query compute budgets are tight yet SC is commonly applied with a fixed sample budget inherited from large-model practice, regardless of instance difficulty. Across two model families (Qwen-2.5-0.5B, Gemma-3-1B) at FP16 and MLX-Q4 (4-bit quantisation in Apple's MLX framework), on GSM8K and MATH-500, we run 8 per-task and 4 task-pooled experiments at n=200-400 with 1,000x paired-bootstrap confidence intervals (CIs). Our first result is an upper bound: an oracle gate deciding per prompt whether to invoke SC reaches 109-174% of always-SC's accuracy at 23-30% of its compute in all 12/12 cases, so at N=5 always-SC is dominated everywhere on our grid. That margin is exactly the rate at which SC flips correct greedy answers to wrong, here 3-10.5% of prompts, which on Qwen-Q4 x MATH-500 exceeds the wrong-to-right rate and leaves always-SC 3 pt worse than always-greedy. Our second result is a deployable approximation: a gate over greedy-pass signals skips 45-99% of SC calls, with bootstrap CI half-widths under 5 pt, removing 36-79% of always-SC compute. Accuracy retention at those operating points has point estimates of 98.1-98.8%, which we report as point estimates because their CIs are wide (lower bounds 76-86%) over an SC accuracy of only 15-36%. The evidence covers N=5 on two <=1B families and two math benchmarks, which is the scope we claim. Code is available at https://github.com/heywanrong/SiftSC.

0.38

Inside the Trading Loop: Markets, Models, and Moves with LLM Agents

cs.AI ·2026-11-08 ·Renee Jia, Di Mu, Yao Yu, Tianrui Liu
Abstract · 37 pages, including appendices. Code and supporting materials available upon request

We built a hybrid multi-agent trading system that combines large language model (LLM) analysts with momentum-based portfolio construction and rule-based execution. Company-news and macroeconomic agents return structured reports to a coordinator that sets stock weights and cash reserves. We evaluate the portfolio engine using its broker-recorded paper account, which operated with deterministic fallbacks, and compare LLM-enabled configurations through retrospective replays. From March 11 to August 28, 2026, the account returned 34.5%, compared with 14.4% for the S&P 500 fund SPY over the same dates. This raw return advantage came with a larger maximum drawdown. Reconstructed policies also finish above matched SPY and QQQ references when assessments are used in the same session or deferred to the next. Design comparisons examine cash authority, selective invocation, and response contracts. A constant-cash control gives a return estimate close to stance-based sizing, and shorter report contracts improve observed parsing but can change recommendations.

0.37

A Stability Analysis of AdamW: Unstable Equilibria and Non-Convergence

cs.LG math.OCstat.ML ·2026-11-08 ·Baiyu Su, Lizhang Chen, Qiang Liu
top venue
Abstract · Accepted by NeurIPS 2026

AdamW is ubiquitous in deep learning, yet its behavior remains poorly understood. We analyze its dynamics through the lens of dynamical systems and show that AdamW admits an implicit fixed-point objective: its fixed points coincide with the stationary points of a constrained and regularized optimization problem. However, not all of these fixed points are stable under AdamW's dynamics, and stability depends sensitively on curvature, weight decay, and momentum parameters. Even in one-dimensional settings, equilibria may be unstable, and trajectories can exhibit persistent oscillations rather than convergence to them. We further extend the analysis to higher dimensions, deriving sufficient conditions for local stability of the continuous-time dynamics, and propose a convergent modification that preserves AdamW's fixed points. These results clarify what optimization problem AdamW is associated with, when its convergence can be expected, and how its dynamics could inspire more reliable optimizers.

0.37

Full Fine-Tuning is not the Problem: Why Adam Fails and SGD Succeeds in Few-Shot CLIP Adaptation

cs.LG cs.CV ·2026-11-08 ·Nairouz Mrabah, Ismail Ben Ayed, Eric Granger
top venue
Abstract · Accepted at NeurIPS 2026

Few-shot CLIP adaptation is often treated as a parameterization problem. Full fine-tuning is assumed to overfit, so parameter-efficient fine-tuning (PEFT) methods are preferred. In this paper, we show that this conclusion conflates parameterization with optimization. Across 11 datasets, 3 CLIP backbones, and 5 shot levels, dense full fine-tuning with plain SGD outperforms representative PEFT methods under the reported evaluation protocol, whereas Adam/AdamW collapses. We trace this collapse to unreliable support-set preconditioning. The few-shot second moment systematically underestimates the corresponding full-data statistic, already at initialization. Adam's inverse-square-root normalization converts this error into an oversized adaptive update. Matching Adam's first-step update budget to SGD resolves most of the collapse, whereas numerator-side alignment does not explain the failure. Shampoo and SOAP also fail, so richer preconditioning alone does not resolve the problem. Theory formalizes how few-shot estimation error is amplified into unsafe updates. SGD succeeds because it avoids unreliable preconditioners and stays close to pretrained CLIP. Interpolation probes show low-loss connections along tested directions, and curvature probes show favorable directional curvature. These results identify optimizer-induced preconditioning, not dense full fine-tuning itself, as the main source of failure in few-shot CLIP adaptation.

0.37

DISCOVER: Online Variance-Guided Data Discovery for Budgeted Multimodal GRPO

cs.LG cs.AI ·2026-11-08 ·Peyman Gholami, Shivam Chandhok, Saber Malekmohammadi, Shayan Shams, Robert Xiao +1 more
top venuecode/data
Abstract · Accepted at the 40th Conference on Neural Information Processing Systems (NeurIPS 2026). Project page: https://peymoon.github.io/discover/

Recent advances in reinforcement learning have made verifiable rewards a central post-training paradigm for improving multimodal reasoning. However, methods such as GRPO remain data- and compute-inefficient, spending rollout and annotation budget on prompts that provide little learning signal. We propose DISCOVER, an online data-discovery framework for budgeted multimodal GRPO that enables vision-language models to dynamically identify prompts that are most useful for the current policy. DISCOVER first organizes the multimodal sample pool into semantic clusters and queries a small set of density-diverse representative samples to obtain initial reward observations. During training, it estimates the utility of unannotated samples by transferring sparse reward feedback from queried samples. Since sample utility changes as the policy evolves, DISCOVER dynamically updates these estimates with recency-weighted rollout observations. It then selects only high-utility prompts for annotation using a policy-conditioned score that accounts for reward variance and difficulty. Once the discovery budget is exhausted, DISCOVER continues with utility-based replay. Experiments on ViRL39K and VLAA-Thinking with Qwen3-VL-2B show that DISCOVER substantially improves multimodal mathematical reasoning under a strict 10% unique-sample budget, outperforming budget-matched random selection, hard-example sampling, full-pool GRPO under the same training-step budget, and online selection baselines. Ablations further validate the importance of semantic coverage, online reward transfer, and stable discovery cadence, establishing DISCOVER as a practical framework for data-efficient multimodal RL post-training.

0.37

When Calibration Data Matter in Sub-4-Bit Language Model Quantization

cs.LG cs.CL ·2026-11-08 ·Yehong Jiang, Yen-Kuang Chen, Xinmin Tian
top venue
Abstract · Accepted to NeurIPS 2026. 10 pages main text, 47 pages total including references, checklist, and appendices

Low-bit post-training quantization reduces language-model memory but can sharply degrade downstream task accuracy. Calibration usually relies on generic text, but when task-informed data recover accuracy and which quantizers benefit remain poorly understood. We introduce FragCal-DC, which supplements generic text with training documents from task families with low retention of full-precision accuracy in a two-bit scan. We use a two- to four-bit sweep to locate functional regimes and four-seed calibration comparisons to evaluate six data-dependent methods across six dense models (1.7B-8B) and ten tasks. At their lowest tested functional bit widths, iterative SignRound V2 and OmniQuant improve at two bits, whereas single-pass GPTQ and AWQ show small mean responses at three bits. At two bits, SignRound V2 gains 1.83 percentage points in mean ten-task accuracy over generic-text calibration across six models. At 70 billion parameters, GPTQ and SignRound V2 both improve at two bits, with functional generic-calibration baselines. For two-bit SignRound V2, task-level controls show that recovery concentrates on families included in calibration, while incorrect answer designations retain about 90% of the mean accuracy gain. Changes in reconstruction error do not consistently track accuracy gains. These findings support task-informed calibration as a design principle for preserving downstream accuracy in low-bit quantization.

0.36

scBIG++: Learning Across Genes and Cells to Predict Single-Cell Perturbation Responses

cs.LG cs.AIq-bio.GN ·2026-11-08 ·Jiafa Ruan, Zican Qiao, Zongxin Yang, Ruijie Quan, Yi Yang
code/datahot direction
Abstract

Predicting single-cell transcriptional responses to genetic and chemical interventions supports gene-function discovery and regulatory analysis. Existing studies often focus only on highly variable genes, restricting the expression changes available for downstream analysis. Cells exposed to the same intervention can also show heterogeneous expression responses, with coordinated gene programs active to different degrees. We propose scBIG++, a full-gene perturbation prediction framework that models this heterogeneity by connecting expression-adaptive gene modules with population-conditioned generation. To preserve shared gene structure while accommodating cellular variation, Gene-Relation Clustering organizes genes into common windows, within which Dynamic SoftSub learns expression-dependent submodule memberships. This shared organization enables Set-GCAE to exchange information about corresponding gene groups across control cells, placing individual representations in their source-population context. The cell representations initialize a latent flow, while their pooled summary and the perturbation representation condition response generation. Generated states are decoded into expression profiles covering all assayed protein-coding genes retained after preprocessing, supporting comprehensive analysis of perturbation-induced expression changes. Experiments across six genetic and chemical settings spanning 6,539--19,451 genes demonstrate competitive perturbation-effect recovery and full-gene expression reconstruction under unseen targets and combinations, cross-cell-line transfer, and held-out compounds. Our implementation is available at https://github.com/ttruan2426-dot/scBIG.

0.36

How Much Training-Inference Mismatch Can On-Policy Distillation Tolerate?

cs.LG cs.AI ·2026-11-08 ·Cong Yu, Ziting Wang, Songlin Jiang, Sitong Zhang, Chengliang Chai +3 more
code/data
Abstract · 28 pages, 10 figures, 12 tables

On-policy distillation (OPD) trains a student language model on its own responses, guided by a fixed teacher. Generating these responses is expensive, so systems use quantized inference and asynchronous rollouts to reduce the cost. Both, however, introduce training-inference mismatch between the rollout engine and the model being trained. Such mismatch can destabilize reinforcement learning (RL), but its effects on OPD remain unclear. This paper investigates how much mismatch OPD tolerates, where its effects arise, and how training systems can use this tolerance. Measuring all mismatch sources on one scale, we examine quantization across three student-teacher pairs, with additional tests of weight staleness and injected noise on one pair. OPD tolerates substantial mismatch without meaningful loss of training quality but degrades at higher levels, revealing an empirical mismatch budget. Both the quality cost of mismatch and the mismatch introduced by quantization depend on the model, with some layers particularly sensitive to reduced precision. In controlled comparisons, OPD remains stable under rollout configurations that destabilize RL methods. Under weight noise, mismatch also shifts the prefixes reached during generation, introducing errors that per-token corrections cannot address. To put these findings into practice, we introduce a lightweight admission check that screens rollout configurations against a calibrated budget without requiring a full training run for each candidate. Model-specific FP8 recipes preserve training quality while providing capacity for at least $1.93\times$ as many KV-cache tokens as BF16. Combining FP8 with asynchronous rollouts reaches the same training quality $1.62\times$ sooner than synchronous BF16 training. Code is available at https://github.com/benyucong/train-inference-mismatch-opd

0.36

ZeroPoint: Lightweight Energy-Based Decision Model

cs.AI cs.CLcs.LG ·2026-11-08 ·Javier Marín
Abstract

Energy-based models (EBMs) assign a scalar energy to every joint value of the observed and predicted variables. In decision models, the prediction is chosen from a finite list of options provided with each input. Inference then reduces to comparing the energies of the options, and training lowers the energy of the correct option relative to the others. A complete decision must also account for the possibility that no option is acceptable. However, the energies learned by a standard EBM define only relative compatibility: adding the same input-dependent quantity to all energies leaves the predictions, the Gibbs distribution, and the usual training losses unchanged. The model therefore has no learned reference level against which to decide whether any option fits. We propose ZeroPoint, an EBM that makes the empty answer a state of the model with energy zero for every input. This fixed state provides an explicit reference for the energies of all listed options, whose non-negative depths measure their compatibility with the input relative to the empty answer. The options and the empty answer share one distribution, in which an option at zero depth receives no probability. The decision to select an option or to abstain is thus part of ordinary inference, trained with the negative log-likelihood loss, rather than a separate rejection rule. The energy function is a small multilingual encoder that reads the input together with all the options and is invariant to their order. Ablations show that most of the benefit comes from training with examples in which no option fits. With the same encoder and data, the fixed zero separates inputs with and without an acceptable option slightly better than a learned rejection class and clearly better than a thresholded softmax, while leaving the choice among options unchanged.

0.35

Learning calibrated probabilistic forecasts using Koopman mode decomposition and the continuous ranked probability score

cs.LG math.DS ·2026-11-08 ·Benjamin M. Kent, Richard E. Turner
hot direction
Abstract · 31 pages plus appendices. 18 figures and 5 tables

Well-calibrated forecasting for non-linear dynamical systems is essential in science and engineering applications such as fluid dynamics and weather prediction. We present a forecasting method combining probabilistic machine learning with the structured linear dynamics of Koopman mode decomposition. Within this framework, the observation, linear dynamics, and state reconstruction operators are constructed using standard machine learning architectures, producing predictive distributions via the pushforward of injected noise samples. The model is trained using a proper scoring rule, the continuous ranked probability score, to directly target calibrated, accurate forecasts. The approach is validated on two non-linear ordinary differential equation examples and two partial differential equation examples, vortex shedding and cavity flow, using both large and small, noisy datasets. Results demonstrate that known Koopman eigenfunctions and spectra can be identified and used for accurate forecasting. The model can also produce multi-modal predictive distributions, demonstrated for a trajectory initialised on the boundary between two basins of attraction. Uncertainty is captured by the sample ensemble spread, with spread--skill ratios close to one over the training horizon in several experiments, although for extended rollouts and noisy datasets, the ensemble may become under- or over-dispersed. In low-noise experiments, temporal rollout is generally accurate for at least twice the training horizon, and up to twenty times in some cases. Approximation quality is comparable to a deep ensemble approximation over the training horizon, but at a reduced computational cost and with better robustness in identifying Koopman eigenvalues.

0.34

Agent Self-Improvement with Frozen Weights: The Added Value and Limits of Failure Traces

cs.LG cs.AIcs.CL ·2026-11-08 ·Arkid Mitra, Snehil Saluja
code/data
Abstract · 6 pages, 1 figure, 1 table. Code and data: https://github.com/hopit-ai/harness-mistakes-artifacts

We evaluate whether an AI agent's harness, the notes around its model, can learn from the agent's failure traces beyond a matched search that never sees them. It can, but less than it looks. We tested an open 9-billion-parameter model with fixed weights; a larger teacher model wrote its notes. On generated customer-service decisions, where both note writers were told the target rules, notes based on the agent's mistakes improved accuracy by 4.0 percentage points over notes from the same process with the mistakes withheld. That process alone, trying and keeping better-scoring notes, reached about 60% of the total improvement. In full customer-service conversations the same notes showed no demonstrated improvement. Separately, a strict screen of single lessons admitted two customer-service lessons, still awaiting tests on new data, and one tool-use lesson, written without seeing the agent's mistakes, which showed no demonstrated gain on new tool-use tasks. A neutral note with no procedure scored higher than the learned lessons (exploratory). A blind audit judged 146 of 240 tool-use test items defective.

0.34

Amortized Bayesian Experimental Design with In-Context Knowledge Conditioning

cs.LG stat.ML ·2026-11-08 ·Zhanghu Zhao, Daolang Huang, Xinyi Wen, Samuel Kaski
top venue
Abstract · 38 pages, 25 figures. Accepted at the 40th Conference on Neural Information Processing Systems (NeurIPS 2026)

Amortized Bayesian experimental design (BED) enables real-time design strategies by shifting acquisition cost offline, but existing methods remain tied to the prior and task distribution they were trained on and cannot exploit additional information available at deployment. We introduce in-context amortized BED with unified knowledge encoding (IMBUE), a BED framework that admits external knowledge at deployment through a unified in-context interface. IMBUE accepts two forms of such knowledge without retraining, namely user-specified prior knowledge and observations from previous instances of the experiment. Both are encoded as additional input tokens and processed in-context with the accumulated observations by a shared Transformer, and a learned reliability filter scores each external token against the observations and excludes inconsistent ones. Across four benchmarks, IMBUE accelerates early-stage information acquisition when the external knowledge is reliable, and remains close to the no-external baseline when it is not.

0.34

One-Shot Hierarchical Federated Clustering for Distributed Non-IID Biomedical Data Analysis

cs.LG ·2026-11-08 ·Shenghong Cai, Yiqun Zhang
published
Abstract · Accepted at IEEE BIBM 2026

In multi-institutional biomedical data analysis, comprehensive disease profiles are often fragmented across different data silos. Furthermore, local data distributions typically exhibit varying levels of granularity or even complex nested structures. Although Hierarchical Clustering (HC) is a powerful tool for exploring such data topologies, its cumbersome recursive process incurs prohibitive computational overhead and renders it highly susceptible to privacy leakage. This paper, therefore, introduces an efficient one-shot hierarchical FC framework that performs client-end (institutions) distribution exploration and server-end distribution aggregation through one-way prototype-level communication from clients to the server. Although the designed one-shot prototype transmission guarantees minimal computational overhead and mitigates privacy risks, it inevitably incurs information loss. To counteract this, a multi-granular competitive learning mechanism is proposed on the server to fuse the clusterlets, effectively maximizing the retention and mining of global distribution topologies under limited, one-way communication constraints. It turns out that the complex cluster distributions across clients can be efficiently explored, and extensive experiments comparing state-of-the-art methods on public datasets demonstrate the superiority of the proposed framework in balancing robust clustering, privacy preservation, and computational efficiency.

0.34

Optimizing Multi-Tier Intermediation in Programmatic Advertising: A DAG-Based Mechanism for Supply Path Optimization

cs.AI ·2026-11-08 ·Amelia Bădică, Costin Bădică
benchmark
Abstract

Digital programmatic advertising operates through complex, multi-hop supply chains. The opacity of these intermediary networks leads to significant efficiency losses, often referred to as the ``ad-tech tax.'' While current industry practices utilize heuristic Supply Path Optimization (SPO) to trim inefficient routes, they fail to optimize the equilibrium of the entire network. This paper extends previous work on Directed Acyclic Graph (DAG) intermediation models by applying them to traceable programmatic supply chains mapped via the OpenRTB SupplyChain object. We introduce multi-attribute utility functions and Mixed-Integer Linear Programming (MILP) to account for substitutable routing (OR nodes) and percentage-based fee structures. Furthermore, we outline a computational methodology and experimental framework using public \texttt{schain} and \texttt{sellers.json} data to benchmark optimal Nash social welfare against current uncoordinated market outcomes.

0.33

Means vs. Foundation Models for Emergency Call Forecasting

cs.LG ·2026-11-08 ·Feriel Fass, Djemel Ziou, Marie-Flavie Auclair-Fortier
Abstract · Keywords: Prehospital emergency calls; Foundation models; TimeGPT; Hölder means; Lehmer means

This study investigates whether generalized means can achieve forecasting performance comparable to that of TimeGPT for daily emergency call volume prediction. Specifically, two families of generalized means, the Hölder means (HM) and the Lehmer means (LM), are evaluated against fine-tuned TimeGPT across forecasting horizons ranging from one to seven days. Forecasting performance is assessed on a real-world emergency call dataset using multiple error metrics, and Diebold-Mariano test. Experimental results show that HM and LM achieve nearly identical predictive performance across all forecasting horizons. For short-term forecasting, both generalized means outperform TimeGPT, with reductions of mean absolute error (MAE) approximately 1.24\% and 2.48\% at horizons of one and three days, respectively. Conversely, TimeGPT achieves a lower error at the seven-day horizon, with an descresing of MAE by 1.98\%. However, Diebold-Mariano tests reveal no statistically significant differences between the methods at the evaluated horizons. These findings suggest that generalized means can provide competitive forecasting performance compared with a energyvore foundation model.

0.33

Resolvent Ellipsoid for Minty Set Inclusions

cs.LG math.OC ·2026-11-08 ·Ashkan Soleymani, Gabriele Farina, Patrick Jaillet
publishedtheory
Abstract · The Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS) 2026

We study Minty set inclusion on a compact convex set $\mathcal{X} \subseteq \mathbb{R}^d$ with a set-valued operator $A : \mathcal{X} \rightrightarrows \mathbb{R}^d$. Under the Minty promise, the goal is to compute $x \in \mathcal{X}$ and $u \in A(x)$ such that $\sup_{y \in \mathcal{X}} \langle u, x - y \rangle \le \varepsilon$. Our approach is geometric and we work in the fixed-scale constrained-resolvent oracle model, where a query at anchor $a$ is assumed to return a resolvent point $x^+$ together with the Yosida vector $g_η(a) = (a - x^+)/η$. We prove that each oracle response yields a sharp dichotomy: either the returned point provides an $\varepsilon$-SVI certificate, or the Yosida vector defines a strict separating normal for the Minty set. Thus, a single black-box resolvent query supplies either a certified candidate solution or a valid geometric cut. Combining this separation principle with the ellipsoid method, we obtain RESOLVENT-ELLIPSOID, an algorithm that computes an $\varepsilon$-SVI in $\mathcal{O}\!\left(\operatorname{poly}(d, \log(1/\varepsilon))\right)$ black-box oracle calls to the fixed-scale constrained resolvent. For monotone inclusions, our algorithm gives, to our knowledge, the first $\mathcal{O}\!\left(\mathrm{poly}(d, \log(1/\varepsilon))\right)$ black-box guarantee in the fixed-scale resolvent model.

0.33

UTOPI: Efficient Egocentric Long-Video Understanding in AR via User-Guided Token Pre-Compression

cs.LG ·2026-11-08 ·Ziqi Wang, Su Chen, Qiance Tang, Jieyu Lin, Ziyun Li +2 more
code/data
Abstract

As long-video understanding gains increasing attention, augmented reality (AR) devices are emerging as a compelling deployment platform. However, egocentric videos exhibit significant temporal redundancy, imposing substantial memory and computational burdens on resource-constrained AR hardware. To address this challenge, we introduce UTOPI, a plug-and-play pre-compression module designed specifically for egocentric long videos on AR platforms. First, we leverage user motion signals to estimate inter-frame overlap and propose a viewpoint-shift-invariant pruning algorithm. Second, we exploit eye-tracking sensors to identify user fixations, enabling the separation of attended foreground content from background and its preferential retention via a fine-tuned neural network-based adjustment mechanism. Across multiple benchmarks and model backbones, UTOPI achieves up to 95% compression while preserving performance. Code and project page are available at https://sai-lab-nyu.github.io/utopi/.

0.33

Hybrid Product Search Without the Copy: Lexical, Semantic and Filtered Retrieval Served Directly from a Data Lake

cs.LG cs.DBcs.IR ·2026-11-08 ·Raghvendra Yadav, Xiang Fu, Chinmay Soman
benchmark
Abstract · 6 pages. Submitted to ISIR-eCom 2026 (workshop at IEEE BigData 2026)

E-commerce catalogs increasingly live in lakehouse tables: Apache Iceberg metadata over Parquet files in object storage. Product search, however, still runs on copies of that catalog, in a lexical engine and a vector database, each with its own ingestion pipeline, schema, staleness window and bill. We show that search applications can be hosted directly on the data lake, with the lakehouse table as the single system of record. We extend Apache Pinot, a real-time OLAP engine that already serves Iceberg tables in place, with BM25 relevance scoring over its Lucene text index, corpus-wide term statistics so that scores are comparable across files, and reciprocal rank fusion in plain SQL; a real-time table fed by a stream can sit behind the same table, merged by the broker. On the Amazon ESCI benchmark (1.2M US products, 22,458 test queries), hybrid search served from the lake matches exact retrieval (nDCG@10 .361). The lake table is queryable 149 s after its files are registered, and a lake commit is searchable in 3.8 s, close to indexing directly into a search engine. On a brand-sorted lake of 277M products (0.8 TB), one 16-vCPU node answers brand-scoped vector, BM25 and client-fused hybrid searches in 6.5-9 ms (p50), sustains 1,100 vector and 635 BM25 searches per second at under 180 ms p99, and makes a commit searchable in 4.3 s. Search adds only its indexes: no copy of the catalog, no sync pipeline and no second schema, which cuts storage, rebuild time and the operational work of keeping copies consistent.

0.32

LLM-based Multi-Criteria Deep Reinforcement Learning for Dynamic Efficient 5G UPF Selection

cs.LG cs.AIcs.CLcs.NI ·2026-11-08 ·Abdelrahman Soliman, Ahmed Refaey Hussein, Lei Lei
published
Abstract · Accepted at IEEE Globecom 2026 CSM

User plane function (UPF) selection in 5G networks is an active area of research, with various limited approaches proposed to enhance network efficiency and manage resources effectively. In this paper, we address this gap in the related literature by proposing a Deep Reinforcement Learning (DRL) UPF selection method that utilizes a dynamic, multi-objective reward function and validating it on a real-world 5G emulation platform. The proposed method is based on the large language models' (LLMs) translation of network operator intents and the network data analytics function (NWDAF) abnormality alerts. The experimental results validated in the real-world experimental setup demonstrate that our dynamic DRL solution achieves an effective balance between latency, power consumption, and traffic distribution, outperforming the heuristic baselines and fixed DRL by 16.89%. These findings highlight the real-world applicability of our approach for dynamic, intent-driven UPF selection.

0.32

Rescene-ESM: an Earth system model built around a frozen AI weather model

cs.LG physics.ao-ph ·2026-11-08 ·Minjong Cheon, Kaoru Kakinuma
hot direction
Abstract · 47 pages, 6 figures, 11 Extended Data figures; Supplementary Information (20 figures, 9 tables) included

Earth system models underpin climate projections, but their cost limits how many scenarios, ensemble members and impact pathways can be explored. Machine-learning weather models forecast as skillfully as physics-based models at a fraction of the cost, yet coupled to an ocean they have produced weak El Nino variability or have learned it from climate-model output, and none carries the climate through rivers to the people exposed to it. Here we present Rescene-ESM, an Earth system model whose coupler, YAHO, yokes a frozen 20.8-million-parameter AI weather model to physically formulated ocean, sea-ice, land, river, carbon and energy components constrained by observations. Adding the couplings one at a time shows what each contributes: over a climatological ocean the AI atmosphere varies from year to year by a third of the observed amount over the oceans, and an interactive mixed layer, an equatorial wave ocean, the residual of the observed sea surface temperature (SST) equation as stochastic forcing and the land's moisture flux raise this step by step and bring El Nino and its teleconnections, leaving the mean climate almost unchanged. The full model simulates about 1,000 years per day on one graphics processing unit, holds the observed climate for a thousand years without drift, reproduces the observed 1979-2022 global warming, carries forced change with a physically based structure around the frozen network (transient climate response 1.60 K), routes runoff to 212 river gauges with a median seasonal-cycle correlation of 0.79 and, driven by emissions, separates the heatwave exposure of an SSP1 and an SSP3 world by a factor of 4.7 in 2081-2100, three-quarters of it through the warmer climate.

0.31

FRESCO: A Novel Consistency Control for Asynchronous Pipeline Parallel Training

cs.LG cs.AIcs.DC ·2026-11-08 ·Dongyeop Lee, Namhoon Lee
published
Abstract · NeurIPS 2026

Asynchronous pipeline parallelism can accelerate distributed model training, but suffers from instability caused by inconsistent gradient computations. We identify a fundamental source of this inconsistency: error signals are coupled to forward-time features, not just parameter states. This means that parameters may evolve as long as they preserve forward-equivalence. Leveraging this insight, we introduce FRESCO, which frames asynchronous updates as a constrained optimization problem: it finds the minimal update modification that acts trivially on the activation subspace. We demonstrate that FRESCO consistently outperforms representative asynchronous baselines in LLM pre-training, with the gains most pronounced under deep pipelines and high inconsistency levels: at pipeline depth 32, it reaches the same loss as the strongest baseline with 43% fewer iterations and 30% less wall-clock time, while requiring 8~921$\times$ less memory overhead. Moving beyond the current limitations, FRESCO establishes a new foundation for high-utilization pipeline parallel training, providing a scalable path for large-model development on diverse, resource-constrained systems.

0.31

Third Eye: Beyond Immediate Gains in Recursive Self-Improvement

cs.LG cs.AI ·2026-11-08 ·Sujato Dutta, Padala Indira Sneha, Kypu Mahitha, Shashank Vanga, Neeraj Choudhary
Abstract

Does a useful self-update remain useful after the model updates again? Third Eye separates immediate performance from the outcome of a further native update. We evaluate five 3-8B backbones on math and code: 64 fresh runs, three candidates per run, and three continuations per candidate. Selection and final scoring use disjoint items. In the balanced Qwen3-4B/Llama-3.2-3B comparison, immediate-selected utility is near zero, but continuation reduces it by 1.78 percentage points (simultaneous 95% interval [0.68, 2.88]). Candidate-averaged immediate utility is already negative: selection partly accounts for the apparently neutral first step. Continuation also lowers utility for SmolLM3-3B and Llama-3.1-8B; Qwen3-8B remains inconclusive. Target and out-of-distribution losses dominate retention changes. Matched recipe and recipient-donor controls identify neither a reliable mitigation nor a single mechanism. The results support evaluating persistent self-updates as adaptation transitions, with independent scoring and a no-update baseline, rather than interpreting immediate performance as their full value.

0.31

Target Recoverability under Selective Labels: A Failure Mode in ECG-Based Acute Coronary Syndrome Prediction

cs.LG cs.AI ·2026-11-08 ·Alexander Schubert, Anton Håkansson, Ziad Obermeyer
top venue
Abstract · Accepted to the Proceedings of Machine Learning for Health (ML4H) 2026

Medical machine learning models often learn from diagnoses observed only in patients clinicians chose to test. The biases of such selective labels are well known, but less understood is what happens when a diagnosis includes cases the model's inputs cannot detect. Because testing decisions draw on information the model does not see, these cases can appear predictable among tested patients, producing relationships that fail, and can even reverse, among the untested. We formalize this mechanism and show in simulations that it worsens as testing becomes more selective or undetectable cases more common, while models trained only on detectable cases remain informative. We empirically study electrocardiogram (ECG)-based prediction of acute coronary syndrome (ACS), which is confirmed only by invasive testing and whose subtypes differ in how clearly they appear on the ECG. Three ECG foundation models fine-tuned on revascularization, our ACS proxy, performed well among catheterized patients (AUROC 0.740-0.770) but worse than chance for 30-day major adverse cardiovascular events among untested patients (AUROC 0.381-0.478). Models trained on ECG-visible subtypes retained discrimination (AUROC 0.680-0.709), whereas a model trained on unstable angina, which can occur without ECG changes, was inversely related to risk (AUROC 0.310). Under selective labels, effective learning therefore requires prediction targets that can be inferred from the model's inputs and evaluation in the intended deployment population, which often extends beyond the patients who were tested.

0.31

Weight-Space Programming of Policies and Their Verification via Model Checking

cs.AI cs.LG ·2026-11-08 ·Dennis Gross
Abstract

Large language models (LLMs) can write the weights of small neural networks directly, without gradient descent. This weight-space programming creates policies for sequential decision-making from a task description alone, but nothing guarantees that a written policy behaves as intended. We combine weight-space programming with probabilistic model checking. Given an environment modeled as a Markov decision process and a requirement in probabilistic computation tree logic (PCTL), an LLM writes all weights of a policy network, optionally repairs them on a dataset of optimal actions, and receives model-checking feedback until the requirement holds or a round limit is reached. The model checker Storm computes, rather than estimates from simulations, the probability that the policy satisfies the requirement, so every returned policy is verified. With two LLMs on three environments, the created policy of each of the 18 runs attains the optimal probability, before any dataset or model-checking feedback. Four of six policies programmed for two transporter jobs also complete three jobs with probability 1, so they generalize to a task that the LLM never saw. The pipeline also programs a policy that satisfies a complex safety requirement expressed with the PCTL until operator in all six runs. In one of them, dataset repair raised the accuracy to 98.4\% but broke the requirement, and one round of model-checking feedback restored it.

0.31

Computationally Efficient Offline Policy Optimization with All-Policy Linear Realizability

cs.LG ·2026-11-08 ·Zhijie Zheng
theory
Abstract · 25 pages

We study offline policy optimization in finite-horizon Markov decision processes where every deterministic Markov policy has an action-value function linear in known features, with coefficient norm at most $B$. From independent complete trajectories of one unknown Markov behavior policy, and under all-policy occupancy coverage, we give an algorithm with $\widetilde O(\varepsilon^{-2})$ sample dependence and computational complexity polynomial in $\log(2+B)$. The algorithm performs exactly one convex regression per layer. Its stopped-return targets are the values of a single consistent Markov witness policy. A shared reference design and bounded prediction constraints convert fast clipped-regression error into pointwise confidence for the raw linear predictions. Truncated leverage controls the performance loss without requiring empirical covariance approximation at scale $B^{-2}$. Rounding confidence radii upward preserves the exact witness and yields a polynomial-time bit implementation for rational numerical inputs. The guarantees allow any fixed initial-state distribution; realizability and coverage need only be checked for deterministic policies. We also establish a covariance-coverage variant with uniform approximate realizability and polynomial dependence on $B$. A two-action lower bound shows that the accuracy exponent $2$ is necessary.

0.30

Verify Before You Fetch: Reliable KV Cache Compression for LLM Inference

cs.AI ·2026-11-08 ·Devanshu Jha, Gaurav Kumar, Agniva Chowdhury, Kumar Krishna Agrawal, Aditya Desai
code/data
Abstract · Code: https://github.com/epsilon-prime/verifiedKV Lab Page: https://epsilon-prime.github.io

Long-context applications, long-lived agents, and multi-agent systems require large language models (LLMs) to process and retain extensive information. In these settings, the KV cache can exceed GPU high-bandwidth memory (HBM) capacity, necessitating offloading to host memory, where CPU-to-GPU transfers can dominate inference latency. While KV cache compression can reduce data movement, query-agnostic compression that guarantees accurate attention for arbitrary queries is extremely challenging, if not impossible. We introduce VerKV, a hybrid approach that provides $(ε, δ)$ guarantees for approximating attention on any given query. VerKV maintains a compact set of query-agnostic pivotal tokens on the GPU, together with a token-level sketch of the full KV cache, while retaining the complete cache in CPU memory. We introduce a novel stochastic sampling procedure, PoissonComp, to select the pivotal tokens. At inference time, our proposed KVFetch algorithm uses the pivotal tokens and the sketch to estimate the relative error of the resulting attention computation. If the estimate satisfies a user-specified $(ε, δ)$ tolerance, inference proceeds without accessing host memory. Otherwise, VerKV performs query-aware retrieval to fetch additional tokens from the CPU, refining the attention computation to meet the required tolerance. Thus, VerKV enables attention with bounded approximation error while minimizing CPU-to-GPU transfers. We show that PoissonComp outperforms state-of-the-art approaches to KV cache compression. Combined with KVFetch, it achieves near-dense performance (90-100% of dense quality at 32K context length, averaged over RULER and LOFT), using Qwen3-4B, Qwen3.5-9B and Ministral-3-8B, while reducing the GPU memory footprint by 8-32x and keeping CPU-to-GPU memory fetches within 2-23% of the cache per generated token, depending on the model.

0.30

Seeing Is Not Believing: Counterfactual Lesion Erasure Teaches a Small Medical Vision-Language Agent to Report Only What the Image Shows

cs.AI cs.CLcs.CV ·2026-11-08 ·Bhaskar Gurram
code/data
Abstract · 28 pages, 1 figure, 13 tables. Preprint v1.0; reader study in progress. Code, data recipes, pre-registrations and per-view outputs: https://…

Medical vision-language models (VLMs) often report a finding because the question names it; standard benchmarks miss this. From public segmentation datasets we build counterfactual views without manual annotation: the lesion erased by inpainting, a same-shape control edit elsewhere, and unedited lesion-free images. The evidence sensitivity gap (ESG) is the rate of absent-or-refer answers after erasure minus the rate after the control edit; it is zero for a model that ignores the image or reacts to both edits alike. MedScope-4B (Qwen3.5-4B, LoRA, image tools) was trained on label-derived trajectories, selected by a pre-registered rule and tested once against nine frontier and six open VLMs; four pre-registered checks then probed the mechanism. On 507 internal cases across five imaging types, MedScope-4B reached ESG 0.758 and claimed an erased lesion in 9.6% of cases, against 0.27-0.47 and 41-62% for the frontier models (all p<0.001), without significant loss of accuracy on real lesions. Without counterfactual views ESG fell to 0.279+-0.107 (three seeds); the effect replicated on MedGemma 1.5 4B and two unseen inpainting tools. At four external centres the advantage held but frontier models detected more real lesions; on a fresh external test MedScope-4B detected 0 of 48 pneumothorax films, with net utility below gemini-4-argon's (-0.36 [-0.50, -0.23]). Matching the control edit was not shown to matter; over three seeds no fresh-external ESG advantage over removal-based DPO was established; a paired-view consistency objective did not beat a matched-compute control, while that continuation alone cut erased-lesion claims on unseen tools at every seed, at a cost in sensitivity at one. Training on erased and edited images gives a small model an evidence-gated disposition that prompting does not give frontier models; it does not improve perception, and rare critical findings need coverage.

0.30

Beyond Node Attributes: A Controlled Evaluation and Ablation of Local Interaction Statistics for Ethereum Phishing Detection

cs.LG cs.AI ·2026-11-08 ·Thanh Dat Tran, Duy Loi Duong, Duy Luu Hoang, Minh Huy Tran, Vo Minh Thu Le +1 more
benchmark
Abstract · 13 pages, 4 figures, 3 tables

Methods for detecting phishing accounts on Ethereum are usually evaluated as complete pipelines that differ in features, architecture, and training protocol, which makes the contribution of interaction information hard to isolate. This paper evaluates the Unlabeled Feature Extension (UFE), a block of 12 local interaction statistics computed from each account's incoming and outgoing call and transfer edges without reading class labels. UFE is appended to the 14 source features of HEIG-Phish and tested with six model families: Logistic Regression, Random Forest, LightGBM, Graph Attention Network, GraphSAGE, and Meta-IFD. Within each pair of Baseline and Plus variants, the data partition, model configuration, random seed, decision threshold, and sampled neighborhoods are identical; only the input feature block changes. Across 20 predetermined stratified partitions of the same 2,763 labeled externally owned accounts, UFE increased test Macro-F1 by 1.0 to 9.1 percentage points across the six models. Every pair improved on all 20 partitions, all Nadeau-Bengio corrected 95% confidence intervals lie above zero, and the false-positive rate decreased for all six models. An exact-edge Random Forest ablation over 10 predetermined partitions associates most of the gain with the incoming transfer-edge sum attribute; once this feature is present, an incremental contribution of the remaining 11 UFE columns was not established. The results support reporting feature-augmented baselines alongside graph models, but they show stability under re-partitioning of one static, transductive benchmark rather than temporal or cross-dataset generalization.

0.30

Measuring Generative AI Content in Political Campaigns: Evidence of Rising Adoption and Non-Disclosure

cs.AI ·2026-11-08 ·Anna-Maria Gueorguieva, Nic Weber
Abstract

The use of generative AI in political campaigns has drawn public concern and policy attention. Public concerns range from distrust of candidates to skepticism about the authenticity and accuracy of political advertising, including AI-generated images and video. Nearly half of U.S. states have enacted laws that prohibit, or require disclosure of, "materially deceptive" AI-generated content (AIGC) in elections to guard against these concerns. This study measures how candidates and their campaigns produce and circulate AIGC during an election cycle. We develop two measures that estimate not only wether campaigns use AIGC, but how much they produce (magnitude) and in what forms (modality). We apply these measures to a novel dataset of text and images from the campaign websites, X accounts, and emails of 12,977 candidates for U.S. political office from 2023 to 2025. We report three findings. First, detectable AIGC from campaigns increased significantly across modalities over the observation window, and campaigns that use AIGC frequently do not disclose it. Second, incumbents' campaigns produced significantly less AIGC than challengers' campaigns. Third, candidates in states with AIGC disclosure or prohibition laws do not produce measurably less AIGC than candidates in other states. Our method and findings have implications for transparency and trust in elections, and on state laws that regulate AI use in campaigns. Because current state laws largely target mainly "materially deceptive" content, we identify additional policy mechanisms that could increase electoral transparency and improve trust as AIGC becomes more prevalent.

0.30

Linear Probes to Bayesian Manifolds: Geometry of Arithmetic in Large Language Models

cs.LG cs.CL ·2026-11-08 ·Anshul Kumar, Neha Prasad, Manoj Kumar, Barnabás Póczos, Deeksha Varshney
Abstract · 26 pages, 4 figures, 20 tables. Under review at ICLR 2027

Linear probes are a standard method to study geometric structures, such as helices and circles, from the activations of large language models (LLMs), and such structure is often taken as evidence that the model computes with it. Probe accuracy alone cannot establish this, because it conflates four separate claims. These are (1) that the concept is decodable from the activations, (2) that they form a specific shape, (3) that the shape belongs to the probed concept rather than to a related concept the model also represents, and (4) that the output of the model depends on the shape. We call these claims decodability, shape, ownership, and causal use, and we test them separately with a four-stage Bayesian pipeline. Stage 2 compares candidate shapes and selects the best one only when its evidence passes a predefined threshold. Stage 3 removes a set of related concepts, specified in advance, and tests whether the shape is owned or inherited. Stage 4 intervenes on the shape to test whether the output of the model depends on it. We apply the pipeline to two-digit addition and multiplication in GPT-J 6B, Llama 3.1 8B, and Pythia 6.9B. Nearly every concept is decodable, but the decoded directions overlap substantially across concepts. Every strongly supported comparison selects a ribbon, which is a helix whose radius changes with the value, or a torus. Of 1,330 cells with an ownership verdict, 19% of shapes are owned and 81% are inherited. Owned and inherited shapes move the output about equally often. Intervening on the shape retains approximately 61% of the effect of intervening on the full probe while using at most five dimensions, against a median of nine for the probe. Our results indicate that a probe verdict alone overstates what a model computes with, and we propose that probe results be reported as verdicts on all four claims rather than as a single accuracy.

0.29

When Credit Turns Against Tools: Transition-Aware Credit Assignment in Agentic Reinforcement Learning

cs.LG ·2026-11-08 ·Xiangyu Lu, Zhanke Zhou, Jiazhe Ning, Chentao Cao, Jiangchao Yao +1 more
top venuecode/data
Abstract · Accepted at NeurIPS 2026

Agentic reinforcement learning with verifiable rewards trains language models through multi-turn tool interactions, yet their reasoning performance and tool-calling ability can decline during training. We trace this degradation to three coupled mechanisms: (1) tool-token imbalance produces negative overall tool credit when incorrect trajectories contain more tool tokens per turn than correct ones; (2) aggregation amplification can further strengthen this negative signal as token-mean aggregation gives longer incorrect trajectories greater weight in policy updates; and (3) failure-driven retries prolong trajectories as models tend to call tools again after execution errors, extending potential negative credit to recovery attempts. To counter this suppression, we propose Transition-Aware Credit Assignment (TACA), which targets the transition from reasoning to tool use through supervision of both call initiation and generation. Specifically, TACA uses Advantage Projection to enforce a nonnegative mean advantage on tool-call tokens within each prompt group through a minimal uniform shift. Additionally, TACA guides tool-use decisions with Decision Contrast, comparing rewards for tool-assisted and direct answers to the same prompt. Through Supervised Replay, it reinforces tool-call generation using calls from previously sampled correct trajectories. Across models and tasks, TACA improves performance and sustains more reliable tool interaction during training, raising mean accuracy over MT-GRPO on six reasoning benchmarks by 4.61 percentage points for Qwen2.5-1.5B and 3.61 for Qwen2.5-3B. Our code is available at https://github.com/tmlr-group/TACA.

0.29

StatPrompt-LLM: A Lightweight Statistical Refinement for Foundation-Style Time-Series Forecasting

cs.LG ·2026-11-08 ·Aniket Sinha, Veronica W. Tsou
Abstract · 10 pages, 1 figure, 7 tables

Time-series forecasting has recently seen a paradigm shift with the introduction of foundation-style models that adapt large pretrained language architectures to sequential data. However, effectively exposing fundamental temporal structures, such as trend, scale, and temporal dependence, to these models remains an open challenge, particularly when the available observations are limited. In this paper, we conduct an exploratory empirical study across different generations of forecasting architectures and propose StatPrompt-LLM, a lightweight statistical refinement of Time-LLM. StatPrompt-LLM augments the existing Time-LLM prompting mechanism with dataset-level statistical information and introduces a zero-initialised autoregressive bypass as a complementary linear forecasting pathway. We evaluate our framework on electricity and meteorological time-series benchmarks, including ETTh1, ETTh2, and Weather, under full-prefix and reduced-prefix configurations. Under the 10% reduced-prefix configuration, StatPrompt-LLM achieved lower in-sample prediction errors than Time-LLM, with unweighted average relative reductions across datasets of 40.2% in MSE and 21.4% in MAE. Full-prefix results were more dataset-dependent. Ablation analyses revealed relatively modest and dataset-dependent changes when selected additional statistical descriptor groups were masked within the combined architecture. Computational profiling indicated additional model parameters and runtime overhead relative to Time-LLM.

0.29

Shape-Adaptive Subspace-Constrained Quadratic Matrix Factorization

cs.LG ·2026-11-08 ·Zheng Zhai, Zhuan Liang, Sujun Guo
Abstract

Quadratic manifold fitting provides a flexible framework for modeling nonlinear low-dimensional structures embedded in high-dimensional spaces. However, existing methods typically employ prescribed noise models or fixed loss functions, limiting their adaptability to varying noise characteristics. To address this limitation, we propose an alternating maximum likelihood estimation method for subspace-constrained quadratic matrix factorization, termed AMLE-SQMF, which jointly estimates the quadratic manifold and the underlying noise distribution. Specifically, we model the residuals using a generalized Gaussian distribution (GGD), whose shape parameter controls the tail behavior and determines the associated fitting loss. By exploiting the closed-form maximum likelihood estimator of the scale parameter, we eliminate the scale variable and derive a profile negative log-likelihood. Based on this formulation, we develop an alternating optimization algorithm that iteratively updates the manifold parameters and estimates the noise shape parameter. This enables adaptive loss selection within a unified likelihood-based framework, without requiring prior specification of the noise characteristics. Experiments on synthetic and real-world datasets demonstrate the effectiveness of AMLE-SQMF in quadratic manifold fitting and its robustness to diverse noise distributions.

0.29

When gene summaries reveal the answers: Gene Ontology circularity in gene embedding benchmarks

cs.LG q-bio.GN ·2026-11-08 ·Jiawei Wu, Huishu Wang, Junlin Chen, Hongyi Wang
Abstract · 31 pages, including supplementary material. GeneProv benchmark and leaderboard: https://hwang595.github.io/gene-embedding-exposure/

A gene summary can contain the very annotations used to evaluate an embedding of its text. The Alliance of Genome Resources converts Gene Ontology (GO) annotations into sentences, which NCBI Gene displays and GenePT includes in its input. These summaries covered 6,172 of 19,256 human protein-coding genes (32.1%), including 58% of genes with one to ten linked publications. We embedded the same corpus with and without the generated summaries and refitted a function-prediction probe on each version. Among genes with one to ten publications, removal reduced the reported per-gene area under the receiver operating characteristic curve (AUROC) by 0.078 with a linear probe (95% bootstrap interval 0.072-0.084) and by 0.033 with a multilayer perceptron. These estimates use a historical GO scorer that mishandled ties. The original predictions are unavailable for rescoring. A simple TF-IDF encoder also benefited from the summaries, as did models evaluated only on experimental annotations. Replacing generated summaries with UniProt function descriptions recovered much of the lost performance despite little verbatim overlap. UniProt also curates many of the evaluation labels. In mouse and yeast, we found a similar dependence on the source of the description. We introduce GeneProv to compare encoders on matched text and to examine how representation rankings change across GO recovery, non-GO endpoints and co-essentiality.

0.29

Task presentation and execution paths shape LLM mathematical performance

cs.LG cs.CLcs.SE ·2026-11-08 ·Min Huang, Mingyan Liu
hot direction
Abstract

Mathematical performance depends on how a language model receives information, uses generated text and submits an answer. We study these dependencies through paired comparisons that separate submission requirements, draft reuse and candidate availability. On 65 development problems, an instruction permitting continued derivation produced no valid final answers in 390 short-output calls; a single-line submission instruction produced 390 format-valid outputs, including 389 parseable answers. Under the fixed submission protocol, 198 problems with failed initial submissions were answered correctly in 9.09p of cases with the question alone, 33.84p with its draft, 46.46p with candidates, and 59.60p with both. Drafts added 13.13 percentage points with candidates fixed, and candidates added 25.76 points with drafts fixed; both effects were significant after Holm correction and were subadditive on the accuracy scale. Retaining a trailing excerpt outperformed retaining a leading excerpt of similar average length by 10.10 points Holm-adjusted p 0.027. A second completion model showed the same directions of input effects on the fixed materials. In a cohort with parseable initial submissions, candidates added no correct answers when the previous draft was supplied, but added 27 points when it was absent. These results establish a conditional value of generated text: a failed submission can leave useful material, whose contribution depends on what else is supplied and what is retained.

0.29

FinCutBench: A Continually Updated Benchmark for Measuring the Temporal Boundaries of LLM Financial Knowledge

cs.AI cs.CL ·2026-11-08 ·Siyuan Li, Ke Wan, Boyong Hou, Yujie Deng, Zixuan Yuan
benchmark
Abstract · 40 pages; v2:Improved writing

In financial backtesting and other point-in-time tasks, external inputs can be restricted by date, but model parameters may retain financial facts disclosed later, introducing parametric look-ahead bias. We introduce FinCutBench, a continually updated monthly financial knowledge benchmark for measuring how far a fixed model's financial knowledge extends over calendar time. The current dataset contains 9,000 Chinese and English four-option multiple-choice questions from January 2023 to September 2026, covering Chinese and U.S. markets and four categories of financial information. We construct monthly knowledge profiles using the approximate months in which the tested facts became public. After a profile passes the decline test and knot checks, we define the FinCut Boundary as the first month when the fitted profile reaches its lower level, and report the stability of the estimated date under resampling. Boundary dates and localization precision vary across models. Using the independently measured Transition Start and Boundary as fixed references for downstream sample grouping, we find significant relative declines in U.S. asset-ranking performance and in earnings recall for Sol and Luna. Downstream effects vary across markets and tasks. The FinCut Boundary thus provides an empirical reference for the temporal scope of parametric knowledge in point-in-time financial evaluation, supporting evaluation-window selection and result interpretation.

0.29

Variance-Reduced Stochastic Hessian-Driven Nesterov Acceleration with Adam-Style Preconditioning

cs.LG math.OCmath.NA ·2026-11-08 ·Yaxin Yu, Long Chen, Minfu Feng
theory
Abstract · 27 pages, 5 figures

Stochastic first-order methods use sampled gradients to reduce per-iteration costs, but estimation errors can slow convergence and, in some settings, cause divergence. To improve gradient accuracy while retaining acceleration, we develop scalar and diagonal versions of VR-Adam-SHANG for smooth stochastic convex optimization. The methods combine STORM-type recursive estimation, Adam-style preconditioning, and SHANG's accelerated interpolation and correction, using at most two stochastic gradient evaluations per iteration without full-gradient refreshes. Under unbiased sampling, bounded variance, and mean-square smoothness of gradient noise, polynomial steps with prescribed initialization yield $O(1/K)$ mean-square gradient-estimation error and expected objective-gap bounds combining $O(1/K^2)$ initial-energy decay with an $O(1/\sqrt K)$ stochastic term after $K$ iterations. These guarantees require neither monotonicity of the preconditioner nor a uniform positive lower bound on its eigenvalues. The analysis also establishes polynomial-step convergence of projected nonrecursive Adam-SHANG-s, filling a gap in its original theory. Experiments compare the methods at matched sample-gradient budgets. In every tested setting, at least one of the two versions achieves the lowest mean terminal objective gap among all compared methods.

0.29

Characterizing Scanpath Decoding Pipelines with Perturbation Profiles

cs.AI ·2026-11-08 ·Saleh Mozaffari, Abolfazl Nadi
Abstract

When a scanpath descriptor supports a classification, it is rarely clear what the classification rests on. We propose perturbation profiles. Five transformations each target one stated property of every scanpath (fixation order, the coupling of durations to positions, the duration distribution, orientation and handedness, and fine spatial structure) while preserving a listed set of invariants. The classifier is refit, and the change in participant-held-out AUC is the pipeline's sensitivity to that property. Intervals come from a participant bootstrap that refits the classifier; in simulations they were conservative (99.7% coverage; standard errors about 1.8 times the empirical spread). We profiled nine descriptors on thirteen binary problems from three public datasets, with a within-fold AUC and class-balanced folds and a stated rule for which sensitivities to interpret. Of 565 intervals, 95 excluded zero; 80 were robust to the pooled AUC, 20 fold partitions and longer runs (36 of them within +/-0.05 AUC), and 45 also held with boosted trees. Most came from GazeBase (322 participants). Permuting fixation order raised the AUC of a descriptor based on mean saccade amplitude, because permuted amplitude has as its expectation the mean pairwise distance, a measure of spread. A "comprehension" contrast pooled over two programs was a contrast between their layouts: it vanished among trials of the same program, and synthetic gaze drawn from the layout alone reproduced the program contrast beneath it. For native versus non-native readers, spatial jitter at the scale of text lines reduced recurrence's discrimination. A Python package is released.

0.29

Direct GPU Energy Measurements of Batch-Level Sharing in Mixture-of-Experts and Recurrent Models

cs.LG cs.AI ·2026-11-08 ·Rafael Pina, Varuna De Silva
benchmark
Abstract · 16 pages, 9 figures

Energy claims for efficient neural networks are often inferred from operation counts. We report direct GPU energy measurements on two workloads that batch very differently, recurrent policies in GPU-vectorised multi-agent reinforcement learning (MARL) and mixture-of-experts (MoE) language models, and ask one question of both: can the rows of a batch be made to share computation, across the batch and over time, at a measured saving? On recurrent policies the answer so far is no. Shared activity patterns exist but rarely last one step, masking a frozen policy fails, and trained-in masks succeed only because a fixed mask is equivalent to a narrow GRU, at 11x the energy of a small dense policy in our implementation. Test-time steering adds nothing beyond re-picking on either workload. On MoE models the answer is yes, and granularity decides how much: on Qwen3-30B-A3B, batch-aware routing reads 25-31% fewer expert weights within KL 0.05, with benchmark changes within noise at 500 items, and uses 12-18% less energy per token on a fused forward. We also introduce persistence-aware OEA, a one-parameter rule that keeps the batch-aware cut with about 40% fewer new experts per step than native routing at batch 32-64, though its energy has not yet been measured. These are pilot measurements on one GPU type, and we state their limits.

0.28

Recursive Pricing and Incumbent Subtree Column Generation for Solving Optimal Classification Trees

cs.LG math.OC ·2026-11-08 ·Tengmu Hu, Chen Chen, Theodore T. Allen
benchmark
Abstract · 48 pages, 6 figures, 10 tables

Solving optimal classification tree (OCT) problems is computationally challenging because branching, routing, and leaf-assignment decisions must be optimized jointly. We develop an incumbent subtree column generation algorithm (ISCG-OCT) that exploits the recursive structure of classification trees. We first introduce a mixed set-packing and set-partitioning formulation (MSSF). We then construct a restricted master problem whose columns represent feasible subtree patterns. Unlike conventional dual-based column generation, our incumbent-centered pricing evaluates objective improvements relative to the current solution. We establish a recursive decomposition in which fixing a branching decision separates the remaining pricing problem into independent subtree subproblems, enabling exact dynamic programming. We also develop threshold-based variants that trade optimality guarantees for computational efficiency. Computational experiments on 27 benchmark datasets demonstrate competitive training and testing accuracy and substantial reductions in solution time on many instances. When initialized with a CART solution, ISCG-OCT can certify global optimality faster than warm-started mixed-integer programming solvers on selected instances. These results demonstrate how recursive pricing and structured subtree optimization can improve the computational efficiency of exact classification tree methods.

0.28

Orthogonal Updates for the Win: Towards Accelerated Adaptive Minimax Optimization

cs.LG math.OCstat.ML ·2026-11-08 ·Zhiwei Zhai, Xinyu Wang, Wenjing Yan, Lei Ding, Ying-Jun Angela Zhang
top venuetheory
Abstract · Accepted by NeurIPS 2026

Single-loop minimax methods are appealing for modern machine learning, but standard stochastic gradient updates can be unstable and oscillatory under tightly coupled primal-dual dynamics, making performance notoriously sensitive to stepsizes. To fundamentally improve stability, we develop an orthogonal update framework tailored to minimax optimization, which applies updates along orthogonalized directions to reduce update anisotropy. However, orthogonalization reshapes the optimization geometry and makes coordinating primal-dual stepsizes even more delicate. To overcome this challenge, we develop AdaSGDA, which equips orthogonal updates with an adaptive stepsize rule driven by accumulated gradient norms. This mechanism automatically balances coupled primal-dual progress under orthogonalization, yielding an effectively parameter-free method for nonconvex-strongly-concave minimax optimization. We prove that AdaSGDA avoids problem-dependent tuning while attaining the state-of-the-art $\mathcal{O}\left(T^{-1/4}\right)$ convergence rate. We further propose AdaSGDA-VR, which incorporates a carefully designed variance reduction scheme and achieves the faster $\mathcal{O}\left(T^{-1/3}\right)$ rate. Extensive experiments demonstrate that our methods deliver competitive performance compared to existing minimax optimization methods.

0.28

Detecting Common Golfing Faults Using AI and Video-Derived Biomechanical Features

cs.LG cs.AI ·2026-11-08 ·Oliver Martin, Jeff Clark, Nawid Keshtmand
Abstract

Understanding how an athlete moves can improve both technique and safety, yet outside the laboratory technique is still assessed by eye. We present a pipeline that detects named golf swing faults from ordinary video, without markers or calibration. Instead of learning end to end, it tracks body pose and the clubhead in a body-centered frame, and classifies the biomechanical quantities a coach would measure, such as the height of the clubhead above the swing plane, with shallow boosted trees. We demonstrate the approach on two common faults, coming over-the-top (OTT) and early-extension (EE). On 160 swings and 143 swings from public sources it detects OTT and EE with a cross-validated ROC-AUC of 0.975 and 0.911 respectively. Because every feature is a biomechanical quantity a coach already uses, measured at a known phase of the downswing, each decision can be traced to the signals and the moments of the swing that drove it. The signals the model relies on most are the same ones a coach checks when diagnosing these faults. The pipeline processes phone video on a consumer laptop in seconds, so it could feasibly be developed into an app for golfers of all abilities and extended to other sports in which technique is judged against a biomechanical reference.

0.27

A Unified Framework for License Plate Identification and Character-Recognition

cs.LG cs.AIcs.CV ·2026-11-08 ·Saurabh Kumar, Dewant Katare, Prakriti Jain
Abstract

This paper presents a unified framework for license plate detection and character recognition on low-power edge hardware. The proposed network uses a shared CSP-Darknet and PANet feature encoder with an anchor-free localization head and a RoIAlign-based sequence-recognition branch. Character decoding is performed directly on aligned feature maps, avoiding image-space plate cropping and a second recognition backbone. The model is trained on an AI-assisted dataset of approximately 37,000 manually verified images covering license plate layouts from all 50 United States. The dataset includes variations in illumination, viewpoint, weather, blur, occlusion, and plate design. The framework is converted to an INT8 TensorRT engine and evaluated on an NVIDIA Jetson Orin Nano with 8 GB memory. The INT8 model shows an mAP50 of 97.8% and a full-sequence accuracy of 90.7%. It records 147 FPS at 6.8 ms end-to-end latency, with 315 MB inference memory and 9.4 W average power consumption. Relative to the YOLOv8s+CRNN baseline, the proposed deployment provides a 7.1x throughput increase and a 77.8% reduction in allocated inference memory

0.27

CHANNELS4WORLD: An Environment-Scale Dataset for Wireless Foundation Models

cs.LG cs.ITeess.SP ·2026-11-08 ·Joonkyu Jang, Hyung-Joo Moon, Kwang Soon Kim, Seong-Lyun Kim, Jakob Hoydis +2 more
dataset
Abstract · 21 pages, 7 figures, 7 tables

Wireless foundation models learn reusable representations across tasks, configurations, and deployment environments. Their pretraining requires environmental breadth, yet wireless datasets often sample few sites densely. We introduce Channels4World, a scene-centered dataset of 100,000 distinct outdoor RF scenes and a procedural method for generating them. Each scene links a generated environment and RF material assignments to 32 base-station (BS) positions, 16,384 UE positions, and aligned variable-length multipath and ordered interaction records at 3.5 and 7~GHz. The corpus comprises 52.43 billion physical BS-UE pairs, 104.86 billion frequency-specific evaluations, and a 29.14-TB packaged payload. Scene identity aligns geometry, placements, channels, and interactions across frequencies, allowing environmental breadth and within-scene observation density to vary independently. The public release offers predefined subsets at multiple scales and separate access to corpus components. A deterministic 90K/10K scene-level train-test split keeps both frequencies and all scene components together. We characterize morphology, RF materials, and transceiver placement across three generation profiles. Controlled propagation and path-retention analyses show that diffraction materially affects retained-link availability and multipath structure, while the retention policy reduces storage substantially with little loss in represented aggregate power. Under matched training exposure, models trained on broader scene collections perform better on held-out scenes, although an equal-episode control shows that scene count alone cannot explain the gain. By making environmental breadth explicit and controllable, Channels4World supports the development and evaluation of wireless foundation models for generalization across deployment environments.

0.27

Not Yet in the Profile: Uncertainty-Aware Dynamic User Profiling by Learning from Future Interactions

cs.LG cs.AI ·2026-11-08 ·Jieyong Kim, Neil Shah, Dong Wang
Abstract

On the Web, user profiling builds compact user representations from interaction histories for efficient personalization, and dynamic user profiling updates them as users' behavior and preferences change. However, new user behavior can be ambiguous, and its interpretation differs across users. To address this, we introduce Uncertainty-aware Adaptive Profile Manager (UAPM), which separates the claims in a user profile from interpretations of new behavior that are not yet confirmed, through a Profile, a Buffer, and a Rulebook. UAPM records user-specific interpretation and update rules in the Rulebook, keeps ambiguous evidence in the Buffer outside the Profile as evidence for later decisions, and writes supported general and situational characteristics to the Profile. We train UAPM with Privileged On-Policy Distillation with Dual Specialization (POPD-DS), which learns from recorded future interactions during training without access to them at deployment. Its first stage, Dual Specialization, distills a stronger model into a current-only Student and a future-informed model that becomes the Teacher. In the second stage, privileged on-policy distillation (POPD), only the fixed Teacher receives future interactions and guides the Student's own management decisions alongside task rewards. Experiments on three Web personalization tasks show that UAPM outperforms multiple baselines under the same downstream input budget, improving on the best same-budget baseline by up to 5%.

0.26

The Shape of a Program: Path Signatures for Trace-to-Program Induction

cs.LG cs.AI ·2026-11-08 ·Mohamed Ghanem, Bernd Finkbeiner
Abstract

Synthesizing programs from execution traces is a fundamental challenge in algorithmic interpretability and the reverse engineering of complex systems. We bring a rough-path-theoretic perspective to program induction, treating execution traces as high-dimensional rough paths and encoding them with the path signature transform, a tool from stochastic analysis whose expectation provably characterizes the distribution of traces a program induces. On three program-induction domains, Karel, List, and RASP, our log-signature trace encoder outperforms LSTM, Transformer, and Fourier baselines in almost every setting, with its largest margins when training data is scarce. We further observe that the trace-to-program pipeline is naturally contractive: it often produces valid programs that are shorter than the ground truth. We turn this into a bootstrapping procedure that exposes redundancy in the ground-truth supervision without any external oracle, which we demonstrate on redundancy-injected versions of our datasets. Finally, the log-signature encoder induces a markedly more interpretable latent geometry than the baselines, in which the nearest neighbors of every state lie on its own execution trace.

0.26

An Interpretable Fuzzy Tree for Turbofan Remaining Useful Life with an Auditable Per-Component Diagnostic Layer

cs.AI ·2026-11-08 ·Hugo Henry, Bharadwaj Dogga, Scott Philips, Kelly Cohen
Abstract · 16 pages, 3 figures, 3tables and 6 equations

Data-driven remaining useful life (RUL) estimators for turbofan engines are accurate but opaque, and their internal features typically encode remaining life rather than any specific component's state. This paper presents an interpretable genetic fuzzy tree that predicts RUL from cruise-condition sensor means on N-CMAPSS while exposing a per-component diagnostic layer inspectable rule by rule. The tree follows the engine hierarchy: sensors to component health, then to bounded component damage, spool damage, and RUL, so each level answers a different maintenance question. Training is staged so the prognostic objective cannot reshape the diagnostic layers: health leaves are fitted to physical targets and frozen, a deterministic fuzzy damage map with fleet-referenced thresholds is imposed on each component, and only the aggregation is optimized for RUL. It consumes 13 cruise-mean scalars per cycle rather than a trajectory, which removes any sequence model. On the native split it reaches 8.44 cycles RMSE on the multi-fault subset and 11.51 across all modes with 333 free parameters, 411 total rules, 32 to 12,000 times fewer than compiled baselines. An exhaustive sensor search bounds which components are diagnosable under the tested cruise-mean design, and the rule base is structured to admit formal verification, which this study does not discharge.

0.26

When Sensors Fail: Coverage Auditing with Incomplete Sensor Records

cs.LG stat.AP ·2026-11-08 ·Chung-I Huang, Wen-Yi Chang
Abstract · Concise author preprint; 3 pages, 1 table; no supplementary material

Sensor outages can conceal prediction-interval coverage deficits behind acceptable marginal performance, while unavailable future readings leave some issued forecasts unscorable. We present a measurement-record audit that links issue-time outage duration, original target availability, and time-uniform statistical evidence. The audit retains complete forecast keys, batches all sensors and overlapping horizons at each measurement tick, and projects simultaneous covered-count and scored-count bounds onto a cumulative conditional expected coverage ratio. This target is distinguished from empirical coverage and outage-conditional population coverage. Selected traffic and air-quality results illustrate the gap between pooled and long-outage coverage; in METR-LA, only 7.6% of issued long-gap forecasts have originally observed targets. Of 30 primary sequential outage-state audits, 26 are inconclusive and four lack scored support. This concise presentation gives the core construction and selected aggregate evidence without the full experimental supplement.

0.26

HydroTuring: A physics probe benchmark testing mass, energy, and momentum conservation in hydrologic models

cs.AI physics.geo-phcs.LGphysics.app-ph ·2026-11-08 ·Zhi Li, Changming Li, Siavash Shams, Han Wang, Qingyi Yang +15 more
benchmark
Abstract

Hydrologic models, including deep learning systems now deployed for operational flood forecasting, are evaluated primarily by how closely their simulated hydrographs match observed streamflow. However, high hydrograph skill does not guarantee physical consistency: in a synthetic calibration experiment, a conceptual model that loses 15% of its water input to an unreported sink still reaches a Nash-Sutcliffe efficiency of 0.93, and no shared benchmark tests whether models obey fundamental conservation laws. Here we introduce HydroTuring (v0.1.0), a black-box benchmark of 37 physics probes that test mass (24 probes), energy (7) and momentum (6) conservation in hydrologic models. Each probe runs a controlled experiment on forcing generated at run time from a recorded seed and applies binary pass or fail criteria validated against 18 physically exact and 59 deliberately broken reference models. Across 32 evaluated models (3 exact references, 7 AI-based models and 22 process-based models), only the 3 exact references pass every probe among the 28 models scored on 5 or more probes. Process-based and AI-based models pass similar shares of their scored probes (76% and 74%), and a pre-trained flood-forecasting LSTM fails 6 of the 11 runoff-only probes it can be scored on. Furthermore, 49% of all 1,184 model-probe pairs remain unscored because models omit the required state or flux variables, while the scored failures occur in perturbation responses and state bounds rather than long-term water-budget closure. HydroTuring provides a reproducible smoke test for model development, peer review and operational deployment under an open community protocol.

0.26

BiVerify: Closing the Loop Between Backward and Forward Verification for Mathematical Self-Correction

cs.AI ·2026-11-08 ·Rui Zou, Xin Zhao, Mengqi Wei, Ji-Rong Wen
Abstract · 16 pages, 10 figures

Large language models (LLMs) have made substantial progress in mathematical reasoning but still produce incorrect solutions to complex problems, indicating the need for more reliable self-correction mechanisms. Mathematical self-correction typically identifies errors through verification, which can proceed backward or forward. The two directions provide complementary verification capabilities. Backward verification checks a candidate answer against the original constraints, bypassing complex intermediate reasoning, but cannot ensure correct reasoning or objective completeness. Forward verification follows the problem conditions and current solution, enabling it to locate reasoning errors and assess objective completeness, but may overlook key constraints in complex solution chains. We propose BiVerify, a dynamically scheduled framework for bidirectional mathematical verification. BiVerify begins with backward verification to screen the answer against the original constraints, followed by forward verification of the reasoning process; subsequent direction switching, state resetting, and re-verification are governed by verification outcomes. Upon failure, the system revises the candidate or verification flow until the same candidate passes both directions. We also introduce shared mathematical tools that separate logical reasoning from precise computation. We systematically evaluate BiVerify on five challenging mathematical reasoning benchmarks using multiple mainstream DeepSeek, GPT, and Gemini models. BiVerify exceeds the strongest representative public baseline evaluated in this work by 13.5 percentage points in average accuracy, with the margin reaching 31.2 percentage points on GPT. Further analysis shows that BiVerify's overall performance exceeds that of either verification direction alone.

0.26

Learning to Rank Propagators in Constraint Programming

cs.LG cs.AI ·2026-11-08 ·Matija Lojović, Mladen Nikolić
Abstract

The order in which propagators are executed can substantially affect the performance of constraint programming solvers, yet propagator priorities are typically determined by fixed, manually designed heuristics. In this work, we explore learning instance-dependent propagator priorities from static problem features. We formulate the task as a learning-to-rank problem, addressing the challenge that the quality of a priority assignment can be measured through solver runtime, while the desired priorities of individual propagators are not directly observable. A self-attention-based model jointly scores the propagators present in an instance and is trained using pairwise preferences derived from solver runtimes measured under randomized propagator priority assignments, with a Plackett-Luce-based ranking objective. Predicted rankings are mapped to the discrete priority levels of the Choco solver before search begins, without modifying its propagation or search procedures. We evaluate the approach on XCSP3 instances and demonstrate that propagator ordering can have a substantial impact on solving time. On a test set of 698 instances, picked out as most difficult by problem class, the learned priorities increase the number of solved instances from 634 to 654, reduce median solver runtime by 27.93%, and reduce aggregate penalized runtime (PAR1) by 12.65% compared with Choco's default priorities. We further observe that improvements vary across problem classes and show that a simple policy selecting between learned and default priorities further increases the number of solved instances. These results demonstrate the potential of learning instance-dependent propagator priorities to improve constraint programming performance, particularly on difficult instances.

0.25

The comparison of density estimation for transfer learning in Banach, Hilbert, Euclidean spaces

cs.LG cs.AI ·2026-11-08 ·Maryam Alsadat Kazemian, Azam Bastanfard
theory
Abstract

Domain adaptation or DA methods are indispensable when target labels or computational resources are scarce, yet the domain kernel density and subspace alignment approaches completely assume a Hilbertian geometry. They measure distributional discrepancy with symmetric L2 divergences over Euclidean feature spaces or reproducing kernel Hilbert spaces (RKHS). This preprint summarizes the manuscript Non-Hilbertian Domain Adaptation. Total Variation Alignment via L1 Density Regularization on Feature Manifolds, which proposes Banach Manifold Transfer Learning or BMTL. BMTL makes two coupled design decisions. The first one is that features are embedded on a geodesic manifold rather than treated as points of a flat Euclidean space, and the second one is that the density matching regularizer is formulated in an L1 Banach norm, whose minimization is equivalent, by Scheffe's theorem, to contracting the Total Variation or TV distance between domain densities, a bounded metric that weights all regions of the sample space uniformly. The L2 formulation, by contrast, squares deviations and therefore amplifies errors in the tails, over emphasizing density outliers that do not reflect meaningful distributional shift. Experiments on MNIST-USPS, COIL20, and Office+Caltech10 show consistent gains; the L1 variant beats an otherwise identical L2 ablation on every task, for example 92.44% vs. 89.80% on USPS->MNIST and surpasses the strongest baseline DMDA by +3.68% on average over twelve Office+Caltech transfers. This document situates the contribution with an explicit comparison of Euclidean, Hilbert, and Banach geometries.

0.25

GenISys: Integrated Ship Systems Design Using Subsystem-Conditional Tabular Diffussion

cs.LG ·2026-11-08 ·Noah J. Bagazinski, Faez Ahmed
Abstract

Ship design requires coordinating coupled subsystems to satisfy engineering constraints and balance competing performance and cost objectives. Generative artificial intelligence has supported component design, but integrated ship generation must also account for interactions among hull geometry, structure, propulsion, and operation. This paper presents GenISys, a framework for generating integrated systems, demonstrated through early-stage bulk-carrier design. Its dataset contains 217,999 proposed designs, including 16,247 that satisfy the formulated engineering constraints. The proposed simultaneous conditional denoising diffusion probabilistic model (SCDDPM) generates subsystems together, allowing each subsystem model to respond to changes elsewhere during generation. Experiments compare this approach with monolithic whole-ship generation and sequential subsystem generation across five independent training runs per architecture. In a study targeting lower cost per tonne of cargo transported, designs were generated for a required voyage range of 5,000 nautical miles. SCDDPM produced designs satisfying the formulated engineering constraints and meeting a common low-cost target in 55.70% of generation requests, compared with 51.10% for the monolithic model and 47.54% for the sequential model. These correspond to 9.0% and 17.2% more qualifying designs, respectively, while maintaining similar estimated transport costs. These results show how generative artificial intelligence can coordinate subsystems to generate feasible, high-performing integrated systems.

0.25

PROLA: Principal-Orthogonal Low-rank Adaptation for Predictive Spatiotemporal Weather Downscaling

cs.LG cs.AI ·2026-11-08 ·Minseo Yoon, Minseong Bae, Sojin Lee, Hyunwoo J. Kim
top venue
Abstract · Accepted to NeurIPS 2026

High-resolution weather prediction is important for resolving local atmospheric patterns missed by coarse global forecasts. Large pretrained weather backbones have improved global forecasting, but adapting them to produce future high-resolution fields from recent coarse states remains challenging and costly. We formulate this setting as Predictive Spatiotemporal Weather Downscaling (PSWD), a task that jointly refines spatial and temporal resolution while predicting future high-resolution trajectories. To this end, we propose a framework that decomposes prediction into spatial and temporal stages using a shared pretrained backbone, fixed numerical scaffolds, and lightweight residual heads. For backbone adaptation, we introduce PROLA (PRincipal-Orthogonal Low-rank Adaptation), which allocates a fixed low-rank budget between pretrained weights' principal directions and their orthogonal complement. PROLA further rescales rank-1 optimizer updates using gradient signals, improving adaptation without increasing trainable rank. Experiments in the PSWD setting show that PROLA outperforms representative low-rank adaptation baselines. Further experiments across diverse datasets and tasks support the broader applicability of PROLA, showing that its benefits extend beyond the primary weather downscaling setting.

0.25

The Race Is Decided Early! A Kernel Law Predicts Lock-In in Majority-Vote Test-Time RL

cs.LG ·2026-11-08 ·Jaeho Hwang, Sehong Park, Daejin Choi, Uiwon Hwang
Abstract

Test-time reinforcement learning with majority-vote rewards trains a language model on unlabeled questions by rewarding the samples that agree with its most frequent answer. During training, each question locks into one answer, on hard questions often a wrong one, but which answer it locks into is known only after training. In this paper, we show that this outcome is decided early, in a short race driven by sampling noise between the answers that are close at the start, and we derive the probability of each outcome. With a separate logit for each answer, the runner-up wins with a normal tail probability of the initial gap, scaled by an explicit noise constant. In a network, however, the answer logits depend on shared parameters, and through the kernel of these parameters the updates of the trailing answers also change the gap of the two leaders. This spillover shifts the boundary between the starts won by each leader, and a closed form computes the shifted probability from a kernel probe and the start logits. In LoRA experiments with three models from two families, predictions made before training with the kernel frozen at its start value forecast which answer each question locks into better than a logistic model refitted on the outcomes, although they use no fitted constants. The closed form matches these predictions on the runs in which a check before training finds that the two leaders stay above the other answers, and it explains why a law that ignores the spillover needs a strongly inflated step.

0.23

Probing Transformer Representations Layer by Layer, Module by Module

cs.LG cs.AI ·2026-11-08 ·Ambroise Odonnat, Theo Gnassounou, Laetitia Chapel, Romain Tavenard, Ievgen Redko
Abstract

Which representation should be probed in a transformer-based foundation model? Recent work shows that the standard practice of probing the final layer is often suboptimal, as intermediate layers can provide better features. Yet, representations are still largely studied along depth alone, treating each transformer block as an atomic unit. In this work, we analyze transformer representations layer by layer, module by module. Through more than 144,000 linear probing evaluations of vision foundation models (VFMs) and large language models (LLMs), we show that layer and module choices should be considered jointly: the best layer-module pair yields substantially larger gains over the final-layer residual stream than the best layer with the residual stream fixed. Moreover, module performance can vary substantially across layers, showing that useful representations cannot be characterized by depth alone. Building on these findings, we evaluate simple fixed probing strategies: the final-layer feedforward activation (Act@100) for vision and the second normalization layer at 60% depth (LN2@60) for language. These strategies improve over the probing baseline in a majority of evaluations and recover, on average, 42% and 86% of the gains from probing the best layer-module pair for vision and language, respectively. For VFMs, these gains increase with the transfer difficulty between pretraining and downstream data.

0.23

Chatter Attack: Resource Consumption Attack for Large Language Models

cs.LG cs.AIcs.CRcs.CV+1 ·2026-11-08 ·Yibo Miao, Yichuan Cao, Xiao-Shan Gao, Yinpeng Dong
published
Abstract · NeurIPS 2026

LLMs are powerful but incur substantial inference cost, which can be exploited by malicious users to induce overly long outputs, increasing latency and operational expense. Prior resource consumption attacks either coerce repetition or suppress the end-of-sequence token by lowering its logit value. However, with autoregressive, sampling-based generation, the search space grows exponentially with length and small deviations or unstable trajectories can derail optimization, limiting attack effectiveness and stability. To address this issue, we introduce Chatter Attack, a novel class of resource consumption attacks that optimize adversarial prompts to induce predefined short target phrases, thereby avoiding search-space explosion while still eliciting long responses. Specifically, Chatter Attack triggers self-doubt using short doubt-inducing phrases as the target, and sustains this state throughout generation via inference attention loss and global entropy loss. To extend to black-box settings, we further propose a novel Bayesian Prompt Optimization (BPO) method, which models the objective function globally by constructing a discrete token kernel, efficiently exploring the entire search space using prior information. Extensive experiments across multiple models and datasets show that our method outperforms baselines, with maximum response length increases to 31.5$\times$, and revealing practical resource-depletion risks for real-world LLM inference services.

0.23

EbbFlow: PINN-Inspired Progressive Curriculum for MeanFlow Training

cs.LG ·2026-11-08 ·Mengfan Yuan, Xunheng Kuang, Jianbin Chen, Lei He
Abstract · 20 pages, 6 figures

MeanFlow learns interval-average velocity for few-step generation, but its fixed temporal sampler mixes diagonal and interior supervision throughout training. We interpret temporal coordinates as collocation points: diagonal samples provide a stochastic boundary target, whereas finite intervals supply a differential-consistency target. The diagonal fraction controls how often the consistency component appears, while interval length enters its gradient through an explicit factor. Motivated by this distinction, EbbFlow uses Temporal Supervision Allocation (TSA) to schedule the diagonal fraction and interval geometry. It starts with diagonal supervision, then admits interior samples and expands their maximum interval; a left-anchored interval law preserves coverage across time. The per-sample loss, target, backbone, and inference procedure remain unchanged. In matched class-conditional ImageNet $256\times 256$ runs without classifier-free guidance (DiT-B/2, one training seed per method, 400k updates), EbbFlow achieves 50k-sample FID of 39.65 at one network evaluation and 36.02 at two-evaluation ODE sampling, versus 44.82 and 39.86 for MeanFlow. Gradient probes show a mid-training difference in component geometry. Ablations support allocation scheduling and left-anchored intervals, although a fixed full-horizon variant performs better at one evaluation. These results suggest that temporal sampling is a useful training control whose benefits depend on the inference protocol.

0.23

ClusLoad: Cluster-Guided Model Sharing for Scalable Distribution Transformer Load Forecasting

cs.LG ·2026-11-08 ·Fangzhou Lin, Guoshun He, Yi Wang, S. Supraja
Abstract

Large-scale device-level load forecasting for distribution transformers is challenged by substantial cross-device heterogeneity and the deployment burden of maintaining device-specific forecasting models. Existing approaches typically rely either on device-specific models, which are difficult to scale to large transformer populations, or on global models, which reduce deployment complexity but may overlook heterogeneous load patterns across devices. To address this trade-off, this study develops a trend-clustered forecasting framework that enables model sharing among transformers with similar temporal behaviors. Representation Learning for Time Series using Time-Embeddings (T-Rep) is employed to learn compact temporal representations of transformer load series, which are subsequently grouped using a Gaussian mixture model (GMM). Representative transformers are sampled within each cluster to train cluster-specific Patch Time Series Transformer (PatchTST) models while retaining device-level predictions. The framework is evaluated using real-world 15-min operational data from nearly one thousand distribution transformers. The proposed framework reduces mean absolute error (MAE) by approximately 20% relative to the direct PatchTST baseline, achieves a high-load warning accuracy of 95.66%, while requiring only seven cluster-specific forecasting models. These results demonstrate that behavior-based device organization provides an effective balance between forecasting accuracy, deployment scalability, and risk-oriented operational support for distribution transformer management.

0.23

GARP: Real-Time Generative Agents on Local Hardware in a Game Engine

cs.AI cs.HCcs.MA ·2026-11-08 ·Piero Molino, Patrick John Chia, Federico Bianchi, Jonathan P. Chen, Paul Szerlip
Abstract · 13 pages, 13 figures

Generative agents are game characters whose plans, conversations, and memories are written on the fly by a large language model (LLM). The original Generative Agents demonstration used cloud inference in a custom simulation. We present GARP, a platform for real-time generative agents inside Godot, and describe the Character Behavior System used in GARP 2. The complete loop runs on local hardware using a 3B-parameter open-weight model and no cloud service. Alongside the original planning, memory, and reflection, GARP gives each agent an emotion system driven by LLM appraisal, four graded personality axes drawn from the Big Five that set simulation parameters as well as prompt text, six needs whose urgency grows over time, and pairwise relationships updated after every conversation. The system narrows 50+ candidate actions to typically 3-8, constrains model responses with JSON schemas, and surfaces important memories in decision prompts. A layered optimization strategy (an adjustable game clock, decisions taken ahead of need, multi-tier caching, prompt compression, and fully asynchronous execution) keeps the game responsive while the model thinks. In a nine-scenario evaluation, 25 of 27 runs passed log-based checks. Both failures occurred in the birthday party scenario, where one guest omitted the event from their daily plan.

0.23

Query- and Candidate-Aware Long User Sequence Modeling for Sponsored Product Ranking in Search

cs.LG stat.ML ·2026-11-08 ·Dongyue Xie, Sergey Kolbin, Manavender Malgireddy, Iurii Shcherbak, Weixing Tang +5 more
published
Abstract · Accepted to ACM RecSys 2026 Workshop on Context-Aware Recommender Systems (CARS)

User sequence modeling is widely used in recommendation and advertising. Search ranking presents an additional challenge: the current query expresses immediate intent, while a candidate product determines which parts of a user's long-term history are relevant. We present a production sequence framework for sponsored-product ranking that augments recent behavior with two retrieval channels over long-term history: candidate-nearest-neighbor retrieval and query-nearest-neighbor retrieval. The selected actions are encoded together with dedicated query and candidate context tokens, while candidate-aware early fusion injects the candidate into every history token. Offline ablations show that dual retrieval improves major ranking metrics relative to the production baseline. In an online A/B test on sponsored-product search traffic, the model increased CTR, attributed conversion rate, and orders per impression.

0.23

LiveAsset: Learning Intent Assets from Corrective Interactions for Agent Runtime Control

cs.AI ·2026-11-08 ·Guanming Liu, Weihang Wang, Peng Zhang, Meng Wu, Kainan Tu +6 more
code/data
Abstract · 12 pages

Large language model (LLM) agents acting across web applications and online services often need multi-turn corrections to align their work with users' underspecified goals and preferences. In practice, users may identify problems in the current outcome without specifying the desired result, so learning user intentions requires interpreting each correction against the agent's preceding and revised outcomes. These expectations are often connected to a particular context, making it hard to know when they apply and how to ensure the agent later acts as the user expects. To this end, we propose LiveAsset, an intent harness that learns reusable intent assets from corrective interactions and uses them for runtime control of agent execution. LiveAsset jointly interprets each correction with the agent's original and revised results to infer user intent, incrementally updating the assets while preserving earlier requirements that remain applicable. The learned requirements guide agent execution from task entry and provide criteria for runtime verification, which uses the explanations to assess applicability before each reply is delivered and directs the agent to revise the result to address unmet requirements while preserving those already satisfied. Following a learning phase in which simulated users progressively reveal requirements through multi-turn corrections, we evaluate LiveAsset using same-task replay of the original requests on π-Bench, PERMA, and Vita and cross-task transfer on PAST-Bench and SkillEvolBench. Experiments show that LiveAsset better fulfills requirements learned from corrective interactions and improves performance on these benchmarks compared with memory, skill, and rule-based harness baselines. Code is available at github.com/landian60/LiveAsset.

0.23

When Are Nearby Experts Composable? Cross-Task Model Merging in Neural Thickets

cs.AI cs.CLcs.LG ·2026-11-08 ·Suxin Ji, Hungtao Wan, Mingjun Liu, An Zhang
code/data
Abstract

Random perturbations around a shared checkpoint can produce experts for distinct tasks, but shared origin does not ensure that their gains survive composition. We formulate expert-pair selection for joint base-relative retention, measured by strict joint success and worst-task gain with unconditional abstention accounting. An exact finite-difference decomposition and an arbitrarily small orthogonal counterexample show why proximity and orthogonality alone are insufficient. We instantiate this formulation with quality-constrained spectral selection, a fixed-budget framework combining cross-task screening, layerwise pair geometry, separate validation, and Base fallback. Across six checkpoints, 21 task pairs, and five independent perturbation pools, the procedure improves strict joint success over Top-1 averaging throughout and leads the four-method model-scale comparison on four of six checkpoints. For the five models with at least 1.5B parameters, matched quality-only, global-cosine, and spectral ranking attain 49.1%, 50.5%, and 53.1% joint success. Under this matched protocol, the spectral selector also yields the highest worst-task gain and Weak Pareto rate and the lowest regression rate. Mean worst-task gain reaches +0.61 percentage points, while 17.1% regression and 13.0% abstention show that composability remains conditional. Project page: https://anonymous.4open.science/r/Thickets-Merge/.

0.23

CVaR Deep Optimal Stopping

cs.LG math.OC ·2026-11-08 ·Yechan Jeong, Youngjoon Hong
Abstract · Neurips 2026

Risk-sensitive optimal stopping is useful when poor stopping outcomes carry substantial costs, as in option exercise, sequential medical treatment, and safety-critical control. A natural objective is Conditional Value at Risk (CVaR) of the stopped reward. We develop a theoretical framework and algorithm for CVaR deep optimal stopping. The framework and algorithm address four difficulties in applying risk-sensitive reinforcement learning to optimal stopping: state augmentation, score-function pathology, blindness to success, and dependence on auxiliary networks. We isolate three structural properties of optimal stopping: single-point reward, action-independent dynamics, and sum-zero gradient. We exploit these properties to establish Markov optimality without state augmentation and derive a pathwise gradient estimator. Under bounded policy derivatives, the pathwise estimator has bounded second moments near deterministic policies, avoiding the divergence that can occur with score-function estimators. Building on this framework, we propose DIOS (Distribution-Informed Optimal Stopping) to address the remaining two difficulties. A per-time-step soft penalty combined with tail-level annealing mitigates blindness to success. Estimating the reward quantile directly from each batch removes the need for an auxiliary critic. For a fixed tail level, we establish a finite-time stationarity guarantee under regularity assumptions along the training iterates. Across the 24 main Bermudan max-call configurations, DIOS achieves the highest mean CVaR and the shortest recorded training time among the tested CVaR baselines.

0.22

Hydra: Towards Transferable Multi-Task Learning on Temporal Graphs

cs.LG cs.AI ·2026-11-08 ·Kiarash Shamsi, Farimah Poursafaei, Tran Gia Bao Ngo, Reihaneh Rabbany, Baris Coskunuzer +4 more
Abstract

Real-world evolving networks are naturally modeled as temporal graphs (TGs), where capturing temporal dynamics is essential for predicting future graph properties that support downstream decision-making. Existing temporal graph methods have been developed primarily for single-task prediction, and little is known about their generalization across tasks or transfer to unseen networks. This leaves the challenge of multi-task graph property prediction in TGs largely open. We address this challenge by introducing Hydra, a novel architecture that integrates local connectivity features from temporal GNNs with a spectral learning module that captures global connectivity patterns. This design enables joint learning of local and global information under a multi-task objective. In multi-task classification, Hydra achieves an 8.9% relative gain in AUC over the strongest competitor. In multi-task regression, Hydra achieves competitive results in all three tasks, while obtaining the best results in two tasks with a 8.2% relative gain in MAE compared to the strongest baseline. Moreover, Hydra delivers these gains with a 22% reduction in training time compared to temporal transfer models. These results provide the first systematic evidence that multi-task transferable learning on temporal graphs is effective. By delivering consistent top-ranked performance, Hydra highlights multi-task training on temporal graphs as a promising direction toward adaptable foundation models for temporal graphs.

0.22

A Sequential Monte Carlo Method for Data Assimilation in Subsurface Flow Systems with Hierarchical Uncertainty

cs.LG physics.geo-ph ·2026-11-08 ·Wenchao Teng, Guido Di Federico, Louis J. Durlofsky
Abstract

Subsurface flow systems often involve hierarchical geological uncertainty, meaning both higher-level metaparameters and cell-by-cell properties are uncertain. This leads to challenges for data assimilation because both sets of parameters must be estimated. In this work we develop an adaptive-tempered sequential Monte Carlo (SMC) procedure for joint Bayesian estimation of these quantities. The method uses noncentered geological parameterizations, with geomodel realizations represented by metaparameters and Gaussian latent variables that are independent under the prior. Adaptive likelihood tempering, importance weighting, resampling, and blockwise Markov chain Monte Carlo (MCMC) mutation are applied at each iteration. Flow surrogates are used for all function evaluations. The hierarchical SMC procedure is further extended to treat successive data assimilation. In this case, existing posterior particles are updated in a manner that accounts for correlations in surrogate model errors across observation batches. We consider two distinct 3D systems with hierarchical uncertainty. The first is a faulted carbon storage system containing three stacked aquifers and two extensive faults. Principal component analysis is used to represent target-aquifer heterogeneity, while geological metaparameters characterize uncertain aquifer and fault properties. The second is a channelized oil-water system containing three facies with uncertain properties. A latent diffusion model represents variability in channel geometry and facies realizations. For the four synthetic true models considered, SMC posterior metaparameter distributions and flow predictions agree closely with reference MCMC results. SMC, however, requires fewer function evaluations and provides speedups of 2.5--7.0 for single-stage data assimilation. Even greater speedups are achieved with multiple GPUs.

0.22

Evaluating generative samplers for lattice field theory by scale

cs.LG hep-lat ·2026-11-08 ·Satoshi Iso, Lingxiao Wang
Abstract · 27 pages, 11 figures, comments are welcome!

Generative samplers for lattice field theory are commonly evaluated by their effective sample size (ESS). We show that a small global ESS need not imply inaccurate long-distance observables: errors spread over many short-distance modes can dominate this diagnostic. We introduce a scale-resolved diagnostic based on the overlap between coarse-grained model and target distributions. At the lattice cutoff, this overlap determines the asymptotic ESS fraction; lower cutoffs isolate long-wavelength errors. In the exactly solvable free theory, two samplers with comparable ESS yield a $32\%$ bias and no bias in the susceptibility, respectively. In two-dimensional $φ^4$ theory near criticality, trained diffusion samplers show the same separation of scales: their path-weight ESS fraction falls rapidly over the volumes studied, while low moments of the magnetization remain close to their target values. Varying the reference parameters substantially changes infrared observables, with little effect on the log-weight variance. Given the target magnetization distribution, a marginal Metropolis test maintains an acceptance rate of order one in our tests. Combined with exact local updates, it reduces the remaining mismatch, although residual bias remains at finite relaxation length. In the free field, uniformly bounded relative errors in the proposal mode variances give a volume-independent bound on the number of updates needed for a prescribed accuracy of the chain's stationary mode variances at fixed correlation length. The target marginal must be supplied or computed separately; estimating it from the samplers' own weights can inherit the poor global overlap.

0.22

RealEmbedBench: Benchmarking LLMs on Real-World Application-Layer Embedded Software Tasks

cs.AI cs.SE ·2026-11-08 ·Yujun Huang, Yuwei Zhang, Yi Zhou, Shanshan Wu, Xiang Chen +1 more
benchmark
Abstract

Large language models (LLMs) perform well on general coding tasks, but their capabilities in application-layer embedded software development remain underexplored. Existing embedded benchmarks primarily focus on hardware-in-the-loop validation in simulated environments. We introduce RealEmbedBench, a benchmark of 5,142 samples covering three application-layer tasks: code generation and completion, state machine modeling and implementation, and cross-OS code translation. We evaluate six representative LLMs and find substantial performance differences across tasks and models. Project context, especially the local code window, significantly affects code generation and completion. Models perform reasonably well at state machine modeling but struggle to translate these models into executable code. Cross-OS translation performance varies substantially by direction. Translation from Harmony to RT-Thread is particularly challenging, with an average pass rate of 0.1667. These findings suggest that developers should consider local code context when using LLMs for code generation and completion and select models according to the translation direction. They also motivate the use of more real-world state machine implementation data in LLM training to narrow the gap between modeling and executable code.

0.22

Compositional Conformal Risk Control for Multi-Stage LLM Cascades

cs.LG ·2026-11-08 ·Yuxia Sun, Jiahuang Wang, Haoting Liu, Zhiwei Huang, Tingrui Pei +1 more
Abstract

Multi-stage large language model (LLM) cascades balance accuracy and inference cost, but reliable deployment also requires controlling error risk and identifying its sources across stages. We propose Compositional Conformal Risk Control (CompCRC), a risk-control framework for pre-specified multi-stage LLM cascades. We decompose the risk of accepting incorrect answers into additive stage-wise contributions over the full query population, linking local risk budgets to a system-level guarantee. CompCRC constructs and freezes candidate budget allocations and acceptance thresholds, selects one candidate on independent data, and certifies its stage-wise risk contributions on separate held-out data. Under conditional exchangeability, we establish a finite-sample guarantee: the marginal expected accepted-error risk of the complete deployment procedure does not exceed the global risk budget. At the reported comparison budget of $α=0.45$, all three stages of CompCRC pass their local checks on MMLU, GSM8K, and HellaSwag. Compared with Global-CRC, which passes system-level risk certification at the same budget, CompCRC reduces observed accepted-error risk by 1.33--2.00 percentage points, improves final accuracy by 0.67--2.00 percentage points, and reduces weighted inference cost by 30.37\% on HellaSwag. The selected cost-based budget allocation also improves accuracy per unit cost by 10.6\%--46.7\% over uniform allocation. Further experiments show that stage-wise statistics identify stages that fail local certification checks, providing diagnostic information alongside system-level risk control.

0.22

TEAM: Topology-Aware, Semantic-Sensitive, General Transformer for Multimodal-Attributed Graphs

cs.LG ·2026-11-08 ·Yushuo Li, Xunkai Li, Zhengyu Wu, Ronghua Li, Guoren Wang
Abstract · 10 pages, 6 figures, 5 tables

Multimodal-Attributed Graphs (MAGs) organize heterogeneous data in graphs whose nodes are characterized by modalities such as text and images rather than traditional unimodal attributes. However, many representative MAG methods organize propagation around node-level states and topology-defined neighborhoods, which can obscure fine-grained modality contributions: a bounded neighborhood may contain semantically weak evidence, while compatible context need not be confined to it. To address both limitations, we propose TEAM, a modality-centric framework that constructs and encodes graph context with node-modality tokens rather than treating each node as a single interaction unit. Concretely, TEAM unfolds the local topology into a Modality Encoding Tree and performs tree-anchored semantic evidence acquisition, organizing local structural context while expanding the accessible evidence space outside the constructed local Tree. We further introduce the Modality-Source Context Transformer (MSCT), which encodes the selected node-modality tokens and coordinates interactions among modality-source groups. By retaining textual and visual attributes as separate interaction units until readout, it supports fine-grained cross-node cross-modal exchange without first collapsing them into whole-node features. Across the evaluated baselines, TEAM attains the highest reported mean on all graph-learning metrics, all three G2Text metrics, and both G2Image metrics. On Toys, it surpasses the strongest baseline by 3.37 and 2.93 percentage points in Accuracy and Macro-F1, respectively.

0.21

SGA-Sparse: Training Native SwiGLU Gates for Activation-Sparse and Low-Bit LLMs

cs.LG cs.AIcs.CL ·2026-11-08 ·Ruokai Yin, Steven Li
Abstract

Activation sparsity can reduce FFN weight traffic during single-user LLM decoding, but active channels must be identified early enough to avoid fetching unused weights. In pretrained SwiGLU models, training-free routing with the native gate performs substantially worse than routing with the up activation at high sparsity. We show that this ordering reverses after sparsity-aware training. Based on this insight, we introduce SGA-Sparse, which converts pretrained SwiGLU LLMs into fixed-budget activation-sparse models by training the existing gate projection directly as a top-k router. SGA-Sparse updates all model parameters end-to-end while preserving the original SwiGLU activation and introducing no auxiliary router parameters. After approximately 30B training tokens, LLaMA models from 1B to 8B achieve uniform per-token 75-85% FFN channel sparsity while remaining within 0.2-2.7 pp of their original dense checkpoints in average zero-shot accuracy. Because post-training quantization degrades sparsity-trained checkpoints, we further study two complementary quantization-aware training (QAT) recipes: continued-QAT largely recovers performance at 4-bit weight quantization, while joint-QAT performs better for the more aggressive sub-4-bit target. To compare activation sparsity with dense low-bit quantization, we translate the sparsity-induced reduction in weight traffic into an equivalent bit-width. Under this unified weight-traffic metric, our 2.5- and 1.5-bit-equivalent models establish a new accuracy-traffic Pareto frontier: the 2.5-bit-equivalent models match dense 3-bit QAT accuracy, while the 1.5-bit-equivalent models approach dense 2-bit models and outperform dense ternary models by 1.3 pp on average. These results establish native-gate training as a practical approach for combining high FFN sparsity with low-bit weights.

0.21

TLRS-DPO: Token-Level Risk Scoring for Better Direct Preference Optimization

cs.LG cs.AIcs.CL ·2026-11-08 ·Suxin Ji, Hungtao Wan
Abstract

Direct Preference Optimization (DPO) is a simple and widely used alternative to RLHF for aligning large language models (LLMs), but its sequence-level objective cannot precisely locate or suppress the local patterns that make an otherwise acceptable response unsafe or undesirable. We propose TLRS-DPO (Token-Level Risk Scoring for Better Direct Preference Optimization), which augments DPO with fine-grained token-level control while remaining fully offline and reference-based. TLRS-DPO introduces a token-level robust compensation term that converts the policy-reference log-ratio on rejected responses into localized risk scores, a curriculum-style schedule that gradually increases the strength of this compensation during training, and a risk-adaptive screening mechanism that selectively penalizes only high-risk tokens instead of entire sequences. This design bridges sequence-level preference fitting and token-level behavior control, and can be implemented as a drop-in modification to standard DPO trainers. Empirically, TLRS-DPO improves alignment quality and safety-relevant behavior with minimal impact on helpfulness compared to strong DPO baselines.

0.21

But It Was True When I Checked: Compiling SHOP-Style Method Preconditions into Partially-Ordered HTN Planning

cs.AI ·2026-11-08 ·Harrison Oates
Abstract · 10 pages, 3 figures

Method preconditions specify when a method may be applied in hierarchical task network (HTN) planning, but their meaning differs under two established semantics. The SHOP planners, like the classical HTN formalism of Erol, Hendler and Nau, require the precondition to hold immediately before the method's first action. HDDL, the input language of the International Planning Competition's HTN track, instead introduces an auxiliary action to check it; under partial order, unrelated actions may invalidate the precondition before the method starts. HDDL's designers adopted this weaker semantics because the first-action reading could not easily be compiled. Subsequent work has repeatedly identified its compilation under partial order as an open question, with existing transformations limited to total order. We give a compilation of the first-action semantics into standard partially ordered HTN planning without method preconditions using only two additional facts, thereby resolving a longstanding question in the HTN planning community.

0.21

System Prompts Affect Response Bias More Than Discriminability in LLM Safety

cs.LG cs.AIcs.CLcs.SE ·2026-11-08 ·Mohan W. Gupta, Sandy Tanwisuth
benchmarkposition
Abstract · 23 pages, 4 figures

Large language model safety evaluations often focus on a compliance-refusal tradeoff: models should refuse unsafe requests without refusing safe ones. We argue that this misses a broader form of prompt-conditioned task selection, where the same request can elicit different behaviors depending on the system prompt. We investigate this using a controlled evaluation crossing eight safety-task prompt families with three clarity levels across HarmBench, XSTest, and IHEval. Holding benchmark items fixed, we evaluate 13 models and use signal detection theory to distinguish two possibilities: whether models become better at telling safe from unsafe requests (discriminability), or simply more willing to refuse (response criterion). We find substantial differences in discriminability across models, but system prompts primarily shift the willingness to refuse, with smaller and less consistent effects on discriminability. Increasing prompt clarity makes models more willing to refuse without reliably improving their ability to distinguish safe from unsafe requests. Beyond refusal, prompt families also shift responses among six distinct behavioral policies, including clarification and safe help. Together, these findings show that system prompts primarily change how models respond to safety-relevant requests rather than consistently improving their ability to distinguish safe from unsafe ones.

0.20

DriveFoundry: Co-Evolving Data and Model for Efficient Closed-Loop Iteration in Autonomous Driving

cs.AI ·2026-11-08 ·Ziyang Yao, Yuncheng Jiang, Fang Li, Haochen Liu, Shaoqing Xu +22 more
code/data
Abstract · 57 pages, 31 figures, Project page: https://Xiaomi-EV.github.io/Xiaomi-Drive-o1

A driving foundation model must learn reusable decision knowledge from rare critical events and apply it under strict runtime constraints. We introduce DriveFoundry, a closed-loop framework improving iteration efficiency under data and computation budgets by coupling structured causal data refinement with adaptive inference. The Data Engine builds traceable scene-candidate records linking scene evidence and factor-decision relations to candidate trajectories and outcome assessments. Candidate comparisons and counterfactual annotations clarify how scene factors affect behavioral feasibility and preference. Verified model failures guide VLM-assisted annotation and expert correction, refining training data and annotation systems. The model engine, Xiaomi-Drive-o1, unifies Perception, Causation, Action, World, and Value within a shared vision-language backbone to enable cross-task knowledge transfer. Block-wise latent refill passes candidate-specific evidence and predicted visual consequences between task blocks, connecting scene understanding, planning, and evaluation. A learned value-of-computation head selects tasks and allocates latent tokens based on expected benefit and cost, adapting inference to each decision. On PhysicalAI, the 9B model achieves the lowest ADE@1 across evaluated splits and horizons among tested models. In AlpaSim, it outperforms Alpamayo 1.5 across all metrics and achieves lower close-encounter and off-road rates than the 34B Alpamayo 2 Super, while trailing in progress and AlpaSim scores. For the 0.8B variant, joint training reduces 5 s minADE@6 by about 54% without altering the inference path or token budget. Adaptive inference matches full reasoning performance using only 28% of tokens. Real-vehicle tests confirm runtime feasibility, supporting a unified paradigm for learning from critical experience and dynamically allocating computation.

0.20

Building heterogeneous multistep workflows in automated microscopy: reward design and knowledge transfer for autonomous structure property discovery

cs.LG ·2026-11-08 ·Sergei V. Kalinin
Abstract

Autonomous microscopy can turn fast acquisition into scientific progress when it decides not only where to measure, but also what information to acquire, at what scan rate, and when to recalibrate the instrument. We develop a budget constrained framework for such multistep structure-property workflows. It combines explicit scientific rewards, action costs, instrument observation models, transfer of pre experiment knowledge into the online campaign, and evaluation kept independent of the controller. The transferred knowledge includes predictor accuracy and computational cost, empirical learner maturity functions that weight model-derived rewards, and drift characteristics established before controller training. Using a real electron energy-loss spectroscopy dataset, we run two complementary campaigns. In the first, the controller chooses between imaging and spectroscopy, image size, and scan speed. In the second, the instrument drifts and the controller also decides when to recalibrate. The campaigns show that response discovery and structure-property learning favor different surrogate models and acquisition strategies. They also show that simple acquisition rules form strong, practical baselines. Independent error analysis reveals where model uncertainty is and is not a reliable learning signal. In the drift campaign, explicit registration greatly improves positional accuracy, and the framework measures what this costs in spectroscopy throughput. Together, these results provide a reproducible way to design, compare, and audit autonomous experiments, and they define what microscopes must provide for deployment: independently executable actions, characterized scan and detector responses, timestamped registration telemetry, and complete action records.

0.20

Alchemically Transferable Parameters for Variational Quantum Eigensolvers

cs.LG quant-phphysics.chem-ph ·2026-11-08 ·Jonas Motyl, Jakob S. Kottmann
Abstract · 14 pages, 8 figures

Machine-learned circuit parameters could remove one of the main bottlenecks of variational quantum eigensolvers: the classical optimization loop, with its typical barren plateaus, local minima, and costly energy and gradient evaluations, is replaced by a direct prediction. Prior work showed that machine-learning models paired with a structured circuit design can transfer across system size and structure within hydrogenic systems. The next step: Extending them to organic molecules would seem to require re-training on use-case-specific training sets. In this work, we show that this is not necessary by establishing an alchemical approach. A geometric projection places surrogate hydrogen atoms along the bond directions of the target molecule and so maps its local bonding environment onto small hydrogenic fragments. Existing hydrogenic solvers can then be reused as they are. The only new quantity is a single projection radius per bond type, which is fitted once on a small reference molecule and reused unchanged thereafter. We demonstrate the approach on the separable-pair circuit designs that have been used in prior work and extend the approach to multi-circuit approaches that span effective bases for quantum subspaces.

0.20

ODSA: Orchestrated Discovery and Self-Aggregation

cs.AI cs.CLcs.LGcs.MA ·2026-11-08 ·Macardle Meng, Haomin Zhang
Abstract

Test-time scaling improves the accuracy of a language model's answers by spending more computation at inference time. Established methods do so either sequentially, revising a single solution over several passes, or in parallel, generating several independent solutions and selecting among them. We present ODSA (Orchestrated Discovery and Self-Aggregation), an LLM-directed scaling framework in which an orchestrator LLM is given existing test-time scaling methods as actions --- discovery of a new independent solution, comparison of existing solutions to build a better one, and acceptance of a proposed answer, which triggers its verification --- and directs stateless workers to carry them out. The orchestrator does not solve the problem itself: it allocates computation, feeds the results of earlier actions into later ones, and takes every action from a single context, so that each decision is informed by everything tried before. On Humanity's Last Exam, ODSA improves accuracy by 4.7 to 18.0 points across three models over a single direct call to the model with no test-time scaling. On LiveCodeBench, where a single direct call already solves most problems, it improves accuracy by 0.6 to 14.0 points.

0.20

ConvLUT: Rethinking Convolutional Architectures for Ultra-Low-Latency LUT-Based Inference

cs.LG ·2026-11-08 ·Marta Andronic, Olivia Weng, Danial Zuberi, Ryan Kastner, George A. Constantinides
Abstract

Lookup-table (LUT)-based neural networks (NNs) offer a compelling path to ultra-low-latency inference on FPGAs. However, existing LUT-based neural networks have largely been restricted to MLP-style topologies, leaving structured vision and audio tasks underexplored. We present ConvLUT, a family of fully unrolled LUT-mappable convolutional architectures targeting the MLPerf Tiny benchmarks. It is comprised of a standard convolutional model built around a hierarchical tree of LUTs and a depthwise separable model following the MobileNet design philosophy. These architectures are enabled by four key contributions: a unified LUT-based convolutional framework supporting standard, depthwise, and pointwise convolutions under explicit fan-in constraints; a LUT-tree that addresses the $k^2c$ receptive field dimensionality bottleneck; a channel-partitioned LUT pooling layer that replaces conventional spatial pooling; and a hardware-aware iterative pruning strategy for stable convergence at scale. Evaluated on CIFAR-10, Visual Wake Words, and Keyword Spotting, ConvLUT achieves competitive accuracy with fully unrolled inference latencies below $\sim 50$ ns. On CIFAR-10 it reaches at least a $8844\times$ latency reduction over non-logic-based prior implementations, and outperforms existing MLP-style LUT-based models on the same task by at least $20$ pp in test accuracy.

0.20

ZarrFM: A Canonical Tensor Storage Architecture for EEG Foundation-Model Pretraining

cs.LG ·2026-11-08 ·Dheeraj Chintapalli, Sunkalp Chandra
published
Abstract · 40th Conference on Neural Information Processing Systems (NeurIPS 2026). Workshop: Foundation Models for the Brain and Body

Statistical models for brain-computer interfaces (BCIs) increasingly require large, heterogeneous EEG corpora, yet existing datasets vary in sampling rates, channel layouts, window lengths, and storage formats, creating substantial preprocessing and I/O overhead. We propose ZarrFM, an execution-oriented representation that complements BIDS by storing a corpus as a single Zarr v3 hierarchy with indexed metadata, canonical channel spaces, explicit masks, hash-stable subject splits, verified event tables, and block-aligned sharding. On real PhysioNet recordings and synthetic corpora, we compare ZarrFM with EDF, FIF, NPY, HDF5, and Parquet. Across 200 recordings, its two-level index reduces metadata queries from 601 requests to 14. Preserving native signal codes reduces storage from 42.4 MB/hour in EDF to 22.7 MB/hour, while block-aligned sampling reaches 3,131 windows/s, up to 11 times faster than Parquet. Reads remain bit-identical across independent Zarr implementations, and concurrent failure-injected appends recover without data loss. These results show that ZarrFM can provide a compact, interoperable, and execution-efficient storage layer for large-scale EEG pretraining.

0.19

Multimodal Graph Learning: A Survey of Models, Benchmarks, and Frontiers

cs.LG ·2026-11-08 ·Rui Xue, Tianfu Wu
survey
Abstract

Multimodal attributed graphs (MAGs), in which nodes carry text, images, and other modalities while edges encode relations among entities, underlie e-commerce, social media, content platforms, and creative applications. Learning on MAGs has developed along two complementary lines: graph-centric methods redesign how modalities are fused, propagated, and connected within message passing, while foundation-model-centric methods pretrain transferable multimodal graph encoders or bring graph context into multimodal large language models for prediction and generation. In parallel, several benchmarks have appeared with different datasets and evaluation protocols. This survey takes an algorithm-centric and evaluation-grounded view of the field and makes three contributions. Modeling: we introduce a unified message-passing formulation and organize methods into five integration levels (feature, propagation, topology, learning, and foundation model), covering discriminative and graph-conditioned generative models; within this formulation we derive representative methods as special cases, compare their assumptions, expressiveness, and cost, and identify cross-modal sensitivity as a property that separates otherwise similar architectures. Datasets and evaluation: we review recent benchmarks, summarize their datasets and metrics, define model-independent descriptors (label homophily, average degree, clustering, and edge provenance), and map each evaluation protocol to the baselines it uses. Frontiers: we connect multimodal graph learning to AI agents, covering the multimodal graphs that agents perceive and remember and graph engineering, the practice of organizing agent workflows as explicit graphs; we then discuss open problems, concrete efforts for the community, and research directions, including when graph structure helps in the presence of strong multimodal foundation models.

0.19

Separation of Skilling Context for Standard Operating Procedure-based Agents

cs.AI ·2026-11-08 ·Mitsuhiro Nakayama, Toshiaki Sota, Takehiko Yamaguchi, Michiaki Tatsubori
Abstract

Standard Operating Procedures (SOPs) are kept concise for human readability, so LLM agents that generate executable skills from them lack the tacit knowledge and implementation details needed for automation. Human intervention fills these gaps, but must be repeated as SOPs evolve---a problem we call \textit{SOP-skill drift}. We propose managing SOPs and Contextual Knowledge (CK) separately, together with a \textit{contract-aware} evaluation that distinguishes genuine violations from CK-resolved differences. On eight Indeed-derived workflows (single LLM), it raises the functional-equivalence rate from 12\% to 58\%. Under three SOP-revision perturbations, previously extracted CK already covers 79.5\% of newly introduced automation gaps.

0.19

Regret without Optimal-Arm Feedback under Reward-Dependent Delays

cs.LG ·2026-11-08 ·Zhong Han
Abstract · 45 pages, 3 figures. Code and data are included as ancillary files

We study stochastic bandits with reward-dependent delays, possibly infinite, whose reports omit the originating round. For each $c\in[0,1)$, Balanced Historical Optimism (BHO) has a delay term controlled by the largest suboptimal-arm quantile at level $1-cΔ_i$, without requiring optimal-arm reports. Lower bounds give an inverse-square slack cost in the logarithmic statistical regime and linear minimax regret at $c=1$. A family with growing $K$ forces worst-case regret $Ω(G\log K)$ for every learner, where $G=\max_iΔ_i h_i$ is the largest gap-weighted suboptimal cutoff. Pending-triggered balance preserves BHO's weighted-cutoff bound for $c<1$ and attains the historical-greedy exploration coefficient with deterministic rewards, a unique optimum and fixed bounded delays. Gap-weighted comparisons and experiments show when shared waiting helps and what balanced exploration costs.

0.18

Jointly Optimizing Memory, Computing and Offloading for Energy-Efficient DNN Inference

cs.LG cs.AI ·2026-11-08 ·Yunchu Han, Zhaojun Nan, Sheng Zhou, Zhisheng Niu
top venue
Abstract

In edge intelligence systems, computation offloading and dynamic voltage and frequency scaling (DVFS) are widely used to improve energy efficiency. However, as applications like deep neural network (DNN) inference evolve in complexity, their demands for memory resources have progressively increased, establishing memory as a critical bottleneck that impacts both latency and energy consumption. In this paper, we propose a joint optimization scheme integrating memory, computing and offloading for energy-efficient DNN inference. We formulate the energy minimization problem under deadline constraints as a mixed-integer nonlinear programming (MINLP) problem, where the main challenge arises from the coupling between offloading and resource optimization. To address this issue, we decompose the original problem into two subproblems under a fixed offloading policy. The local subproblem is transformed into an equivalent geometric programming (GP) problem to derive the optimal memory frequency and computing frequency, while the edge subproblem is solved via convex optimization. An efficient coordinate descent (CD) algorithm is further developed to jointly optimize all variables. Simulation results demonstrate that our proposed algorithm achieves near-optimal performance with the gap from the optimal policy being at most 9.3%, and reduces the energy consumption of devices by up to 16.1% compared with benchmarks without memory frequency optimization.

0.18

A Theoretical Reduction From Membership Inference to Reconstruction Attacks Against Learners

cs.LG ·2026-11-08 ·Amnon Levi, Yuval Dagan
theory
Abstract · 23 pages

We study reconstruction attacks, which recover training data from a learned model. In our adversarial model, the adversary designs a hard instance, drawn from a parameterized family of data distributions, and is given the parameter used to generate it. Our main contribution is a general reduction showing that, in this model, membership inference (determining whether a given datapoint was used for training) implies reconstruction: for a training set of size $n$, a membership test with recall at least $q(n)$ and false-positive rate $O(δ^2/n^2)$ yields a reconstruction algorithm that, with probability $1 - O(δ)$, outputs only training points, and recovers $Ω(n \cdot q(n))$ of them in expectation. This reduction applies to any learning algorithm. For stochastic convex optimization (SCO), we make the reduction computationally efficient over exponentially large domains via structured distributions whose support is partitioned into polynomially many categories of exponentially many elements each, so that reconstruction iterates over categories rather than the full support. Combining this with an existing membership inference result for SCO, we prove that any learning algorithm achieving low excess risk over all distributions admits a polynomial-time reconstruction attack recovering $Ω(n)$ training points, even when the underlying distribution has exponentially large support.

0.17

RADE: Risk-Adjusted Deployment Efficiency, A Framework for Routing, Memory, and Escalation in Safe LLM Systems

cs.AI ·2026-11-08 ·Chaymaa Abbas, Kareem Hassani, Mariette Awad
benchmark
Abstract

A deployed language model does not merely answer questions. It spends computation, consumes time, consults evidence, exposes memory, invites or avoids human review, and sometimes makes mistakes whose consequences are radically unequal. Yet LLM systems are still commonly compared as if the relevant object were a single model and the relevant outcome were a single benchmark score. This paper argues that the unit of analysis should instead be the deployment policy: the mechanism that decides when to retrieve, when to remember, when to call a larger model, when to verify, and when not to answer at all. We introduce risk-adjusted deployment efficiency (RADE), a system-level objective that scores the utility a deployment delivers net of inference, retry, oversight, and priced expected failure cost, while treating harms that cannot be priced as hard budgets alongside constraints on latency, privacy, and resources. The framework brings selective routing, retrieval and governed memory, efficient specialization, uncertainty estimation, abstention, and human escalation into a common account. We then examine five deployment regimes in which the same nominal LLM capability leads to different rational architectures, and propose a reporting protocol based on risk-coverage behavior, end-to-end cost, update drift, memory exposure, evidence quality, and escalation load. The central claim is modest but consequential: frontier capability should be treated as a resource to allocate, not a default state to maintain. A safe and sustainable system is not the one that always uses the strongest model; it is the one that spends capability where the reduction in risk or increase in utility justifies the cost.

0.17

Does Predictive Performance Translate to Causal Performance? Evaluating tabular foundation models as nuisance learners for causal effect estimation

cs.AI stat.APcs.LGstat.ME+1 ·2026-11-08 ·Andrew R Weckstein, Issa J Dahabreh, Marinka Zitnik, Sonia Hernández-Díaz, Sebastian Schneeweiss
Abstract

Tabular foundation models based on prior-data fitted networks (PFNs) achieve strong predictive performance, but their value for causal inference is unclear. We evaluated whether PFNs, which predict through in-context learning, could match or outperform conventional nuisance learners in supporting accurate causal estimation and reliable uncertainty quantification (inference). Learners modeled propensity scores and outcome regressions within doubly robust estimators of average causal effects, with and without cross-fitting, across synthetic and semi-synthetic settings with varying confounding complexity and effect heterogeneity. Under simple parametric data-generating processes, the PFNs were competitive with correctly specified parametric regressions and ensembles containing them. In more complex settings, they outperformed all conventional learners evaluated, including richly specified Super Learner ensembles, for both accuracy and uncertainty quantification. As sample size increased, only the two PFN models (TabPFN and Mitra) consistently approached nominal confidence interval coverage, whereas under-coverage persisted for conventional learners. In some settings, PFNs achieved near-nominal coverage and well-calibrated standard errors even without cross-fitting. At the smallest sample size (n=200) and in the higher-dimensional semi-synthetic settings, PFNs and other flexible learners exhibited poor coverage without cross-fitting. For PFNs, this appeared to be driven by propensity score overfitting consistent with in-sample memorization of treatment labels. These findings highlight the promise of prior-data fitted networks as in-context nuisance learners for causal estimation, while underscoring the need for further evaluation in challenging settings.

0.17

MOMENT-DP: Momentum Aware Evaluation of Correlated Noise for Differentially Private Training

cs.LG cs.AIcs.DC ·2026-11-08 ·Sunny Gupta, Raja Viswanadha
published
Abstract · 18 pages, 1 figure, 4 tables. Accepted to the Privacy in the Era of Large Opaque Models (PriLOM) Workshop at NeurIPS 2026

Temporally correlated noise can substantially improve the utility of differentially private optimization, yet candidate mechanisms are commonly designed or evaluated using only partial information about the optimization process, such as workload structure, learning-rate schedules, or local curvature. We study whether persistent optimizer state provides additional information that can change the relative utility of privacy-valid correlated-noise mechanisms. We introduce MOMENT-DP, an optimizer-state-aware framework that models how temporally correlated privacy noise propagates through momentum-based optimization. For locally quadratic objectives, MOMENT-DP expresses privacy-induced final loss through an optimizer transfer operator that jointly captures curvature, learning-rate scheduling, momentum, and temporal noise covariance. NoiseCurve's curvature-aware quadratic criterion is recovered exactly as the zero-momentum, constant-learning-rate special case. Changing only momentum while holding curvature, schedule, privacy calibration, and candidate mechanisms fixed induces genuine mechanism-ranking reversals. The effect persists outside exact quadratic dynamics in per-example-clipped private logistic regression, while remaining unresolved in the neural regime we study. These results establish optimizer state as an under-modeled, decision-relevant axis in evaluating temporally correlated privacy noise.

0.16

LoMASA: A Local Multi-Agent Sales Assistant with Confidence-Gated Delegation for B2B Quoting

cs.AI cs.MA ·2026-11-08 ·Yu Liu, Xingbang Du
Abstract

Frontline sales staff spend much of their day re-coordinating quotes after customers change quantities, ask for discounts or new payment terms, or costs move. We present LoMASA (Local Multi-Agent Sales Assistant), a training-free multi-agent system that runs entirely on one desktop computer. A 35B mixture-of-experts model (System-1) makes each decision in a single forward pass and reads its confidence from the probabilities of the answer options. A 27B dense model (System-2) writes customer replies and escalation briefs and is consulted when System-1 is unsure. Code computes every price, margin and option, and hard pricing rules cannot be overridden. On a test split of 500 simulated episodes with 1512 decisions built from public supply-chain data, the salesperson has to decide only 19.5% of decisions: 38.0% are settled automatically, with a median latency of 2.67 seconds, and 42.5% are escalated to a manager by the hard rules. On the automatically settled decisions LoMASA is as accurate as a rule baseline (0.894 vs.\ 0.894) and makes fewer commitments whose margin later falls below the floor (2.61% vs 4.53%); the price is additional escalations to the manager. Option probabilities are well calibrated without any training, unlike confidence stated in words by the same model. We also report where the design does not help: the 27B model rarely changes a decision.

0.16

Understanding Spike Encoding across IoT Sensing Modalities: An Empirical Study

cs.NE ·2026-11-08 ·Yuxuan Hou, Di Yu, Linshan Jiang, Jin Zhang
Abstract · 9 pages, 7 figures, 5 tables; includes supplementary material

Spiking neural networks (SNNs) offer a path to energy-efficient sensing on Internet of Things (IoT) devices. The input encoder shapes that opportunity by changing the cues available for recognition and the work required to process them. We study these choices across wireless, inertial, surface electromyographic, and tactile sensing. For wireless activity recognition, interleaving reduces estimated energy for arithmetic in the network layers, latency from the prepared input window to returned scores, and energy on the graphics processing unit and system-on-chip (GPU/SoC) rail, compared with repeating the full window on Jetson AGX Orin. With the same input size at each SNN time step, repeating a compact input can yield higher mean development accuracy than interleaving. The inertial experiments show that higher accuracy can come at higher device cost. Direct input can also be faster and more accurate than an event encoder while using less energy on the measured rail, even when its estimated energy for arithmetic in the network layers is higher. Level crossings improve grasp recognition on a later day for previously observed users. In tactile slip detection, level crossings shift the balance between missed slips and false alarms, and checkpoint selection can reverse which input offers higher slip recall. We use these findings to outline how edge sensing systems can profile encoders against the signal structure, the errors that matter to the task, and the cost of the full inference pipeline.

0.16

Data Manipulation Attacks Against Machine Learning Integrated Cyber-Physical Systems

cs.LG ·2026-11-08 ·Soyon Choi, Scott Alfeld, Meiyi Ma
Abstract

There is a well-established body of work demonstrating that individual machine learning (ML) algorithms can be manipulated or deceived. However, in many real-world systems, multiple ML algorithms are composed to enable automated decision-making, particularly in Cyber-Physical Systems (CPS). Often such systems use a composition of vision, planning, and other ML models to affect their internal state and ultimate take actions in the physical world. While there is a huge literature on the (in)security of ML algorithms, it is not immediately clear how an attacker might leverage their ability to trick individual learners when attacking the composition as a whole. We develop methods for creating an attack policy (a mapping from state to attack) for a data-manipulating attacker affecting a CPS or other state-machine-based AI compositions. We describe our framework, a basic threat model and corresponding methods for computing an optimal attack policy, as well as planned extensions. The goal of this work-in-progress paper is to stimulate conversation and forge new collaborations regarding the security of compositional learning.

0.15

Branch-and-Commit for Frozen RL Agents: When and Where to Spend Test-Time Compute

cs.LG ·2026-11-08 ·Sahar Admoni, Assaf Hallak, Yftah Ziser, Ofra Amir
Abstract

Reinforcement learning agents are often deployed with policies that cannot be retrained or modified. In digital environments whose states can be copied and resumed, an operator has another option: sample several continuations from the current state, evaluate them, and continue from the endpoint of the selected branch. We call this procedure branch-and-commit. We study when it improves return, what produces the improvement, and where its computational budget should be spent. Across four environments, two frozen policies, five continuation verifiers, and a model-free selection rule, we find that the benefit of branching depends on dispersion: the difference between the best candidate return and the mean candidate return. Branching provides no benefit in low-dispersion settings, while increasing return by 12.8% in our main Craftax-Medium setting. Most of this gain comes not from selecting better actions, but from selecting among different stochastic realizations of the same action. Several competent verifiers achieve similar episode returns, suggesting that verifier quality is not the main constraint at our operating budget. Instead, computation is better allocated to states where useful dispersion is likely. On held-out seeds, gating branching using predicted dispersion retains 69% of the improvement from uniform branching while using 48% as many branching calls. These results characterize when branch-and-commit is useful and clarify what its gains represent.

0.15

OPSRL-SSP: Optimistic Posterior Sampling for Stochastic Shortest Path with Minimax-Optimal Regret

cs.LG ·2026-11-08 ·Yiming Xu, Xian Wei, Cheng Chen
theory
Abstract

We study online reinforcement learning for the stochastic shortest path (SSP) problem with unknown transition dynamics. Although existing posterior-sampling methods perform well empirically on SSPs, their regret guarantees do not achieve the minimax-optimal order. We address this limitation by proposing OPSRL-SSP, an optimistic posterior-sampling algorithm for the SSP setting. The algorithm operates in epochs and uses a logarithmic, time-dependent schedule for the number of posterior samples. To induce optimism, we augment the transition model with a pseudo-state assigned an optimistic boundary value. We establish a high-probability regret bound of $\tilde{O}(B_\star \sqrt{SAK} + B_\star S^2 A)$, where $B_\star$ is an upper bound on the expected cost of the optimal policy, $S$ and $A$ are the numbers of states and actions, respectively, and $K$ is the number of episodes. In addition, we extend the analysis to general zero-cost SSPs using a universal cost-perturbation argument. The leading term matches the minimax lower bound $Ω(B_\star \sqrt{SAK})$, thereby addressing the open problem raised by Jafarnia-Jahromi et al. [19] for posterior sampling in the SSP setting.

0.15

D3Tree: Learning to Prune Block-Diffusion Draft Trees for Efficient Verification

cs.LG cs.AI ·2026-11-08 ·Hong Nie, Zhangchi Gao, Jun Yu, Bo Wang, Jianqing Sun +1 more
Abstract

Block-diffusion drafters propose a whole block of tokens in one forward pass, and draft trees built from their position-wise marginals increase the expected accepted length of speculative decoding. Every tree node occupies a query row in target verification, so under concurrent serving low-yield branches compete with other requests for verification capacity. Existing tree methods select verified nodes by draft probability under a fixed or cost-selected per-request budget, and thus always retain a prefix of the drafter's ranking. We present D3Tree, which learns which nodes of a block-diffusion draft tree are worth verifying. A tree-size head with 2H+2 parameters predicts each candidate's conditional survival from the draft state, the candidate embedding, and its draft log-probability. Thresholding the product of these scores along parent chains yields an ancestor-closed subtree whose size adapts to each request and decoding round. The head is trained on a frozen drafter with conditional labels on the full candidate tree; an optional joint mode also steers the drafter toward the retained reference path. At serving time, D3Tree verifies whole trees under light load and, under heavy load, allocates one batch-wide node budget by predicted survival, sized to the target's matrix-multiplication tiles. Retained trees share one ancestor-masked target pass, while construction, pruning, and scheduling run on the GPU and overlap with the host. Across six vision-language verifiers and 17 text and multimodal benchmarks, D3Tree is fastest in 70 of 72 load settings, with 3.2-4.0x speedup over autoregressive decoding for one request and 1.3-1.8x at 64 concurrent requests, where fixed-budget trees fall below autoregressive speed. Under a time-varying arrival trace, it has the lowest mean end-to-end latency in every arrival phase.

0.14

GOPR-CAD: Graph-of-Operations Progressive Reasoning for CAD Generation

cs.AI ·2026-11-08 ·Yitian Pang, Yifan Li, Xiaowei Yue, Li Zheng
benchmark
Abstract

Generating accurate CAD models requires substantial engineering expertise and careful control of geometry, topology, and mechanical constraints. Text-to-CAD aims to simplify this process, but current LLM-based methods often generate CAD programs directly. Language models remain unreliable when producing long code while coordinating coupled geometric, topological, and dimensional constraints. Consequently, executable programs may still produce structurally incorrect models, which execution rate and rendered appearance may fail to reveal. To address these limitations, we propose GOPR-CAD, an agentic Text-to-CAD pipeline based on Graph-of-Operations Progressive Reasoning. A language model acts as a decision agent over encapsulated CAD operations and progressively constructs an executable operation graph, which a deterministic compiler converts into the final CAD model. This separation requires no additional training and supports different language models. To evaluate structural fidelity, we construct MHCAD, a benchmark of 300 standard mechanical part requirements derived from Machinery's Handbook. Their explicit structures enable reference-free STEP/B-rep evaluation. GOPR-CAD achieves 82.7% Structural Yield on MHCAD, and 87.2% Structural Yield on an external CADPrompt subset.

0.14

Statopia: A Prover--Library Ecosystem for Automated Formalization and Proof Discovery in Statistics

cs.AI stat.CO ·2026-11-08 ·Ziyuan Chen, Tong Gan, Wei Zhang, Ruofei Fang, Boyue Tang +2 more
theory
Abstract

Theoretical results in statistics and machine learning rest on long, technically delicate proofs, and fluent arguments generated by large language models are not necessarily correct. Automated formalization makes such arguments mechanically checkable, and automated proving goes further by discovering machine-checked proofs for statements whose proofs are incomplete or missing altogether. Both require reconstructing assumptions from source materials, accessing specialized mathematical infrastructure and managing long proof developments. We present Statopia, an ecosystem that jointly develops Statopia Prover, an independent agent specialized for formalizing and proving results in statistics and machine learning, and Statlib, a Lean~4 library built with the prover and organized for automated use. The prover controls its own execution without relying on external coding agents. It is built for formal proof discovery and checking existing proofs. A natural-language proof, when available, guides its search but is not required: from a statement alone, the prover develops the argument itself and constructs a Lean proof. Its workflow integrates source preparation, statement review, library supported planning and proof construction. Dynamic proof dependency graphs and task groups support parallel, resumable proof development. Statlib's shared definitions and theorem interfaces are organized for retrieval and composition by agents, and its coverage of statistics keeps expanding. Verified results and usage feedback from the prover inform reviewed library updates, so that proof development drives this expansion. A cross-platform application provides an interactive workspace for uploading source files and running tasks locally or in the cloud.

0.13

Primitives and Paths: Predicting What Post-Training Can Do from What the Base Model Already Has

cs.LG cs.AI ·2026-11-08 ·Anqi Qu
position
Abstract · 10 pages, 1 figure

Does reinforcement learning teach language models new capabilities, or merely amplify ones acquired during pre-training? The literature offers evidence for both: pass@$k$ studies suggest RL reweights latent behaviours while some controlled composition studies demonstrate gains beyond simple amplification. We argue that this question has no answer independent of the capability state of the base model, and propose a two-dimensional state space for post-trainability. The first axis, \emph{effective full-solution support}, measures whether successful trajectories can be sampled under a feasible exploration budget. The second axis, \emph{atomic primitive coverage}, measures whether the constituent skills of a solution can be executed when the burden of composition is removed. Together, these dimensions motivate three regimes -- amplification, composition-limited, and acquisition-limited -- each predicting a distinct empirical signature and calling for a distinct intervention. We propose a prospective measurement protocol and investigate the framework using synthetic composition tasks across three small language models. We find that primitive mastery alone does not guarantee that RL can learn a composition. Checkpoints that execute every primitive reliably can fail to improve when complete solutions are never sampled during training, whereas modest composition supervision that establishes nonzero support enables substantial RL gains. The framework reframes mid-training as preparation for post-training: its role is to move models into states from which RL can operate effectively.

0.11

Same Error, Different Recovery: How Delivery Changes LLM Agent Behavior

cs.AI cs.AR ·2026-11-08 ·Sunghyeun Woo, Javier Conde, Pedro Reviriego
Abstract

A tool-using language model must decide whether to retry a failed call, correct it, or stop and tell the user. The error text alone may not be enough: the same feedback can lead to different actions depending on where it appears in the conversation. We isolate this delivery problem by varying placement, guidance, retained history and turn structure. Two studies with plans fixed before data collection, across 13 open models from 3B to 70B parameters, compare the next action and subsequent task completion, with exact deterministic reruns. A new user turn that retains the failed exchange avoids the severe stopping failure observed in Llama models when that exchange is removed. It is non-inferior to history rewriting in pooled synthetic comparisons, but is not best for every model. We implement this form as an opt-in recovery layer within a deterministic inference engine and evaluate it on public benchmarks and a coding agent. On ToolMaze, it raises success on explicit errors by 14.2 points with 7-8B models and 10.5 points with 32-70B models; Loud/Quiet, 24 multi-turn tasks with visible failures and silent corruption, scored by final-state agreement, shows no measurable gain. A client given the exact fault kind performs better still. The findings support delivery as a distinct factor in recovery, not an engine-exclusive solution or a universal retry policy.

0.11

Self-Indexing Latent Memories for Access-Aware Continual Instruction Tuning of LLMs

cs.AI ·2026-11-08 ·Dhruv Dixit, Paritosh Pandey, Snehalkumar 'Neil' S. Gaikwad
Abstract · 9 Pages

Continual instruction tuning often preserves old skills with task-private modules. Inference still requires selecting the right module from the prompt alone. We introduce Task-Isolated Latent Adapter Continual Training (TILA-CT), a replay-free method that makes each task module self-indexing. Each task receives a latent memory and adapter bank in a frozen LLM. During adaptation, the memory gathers causal-prefix information and injects task-conditioned context. For access, its persistent state provides the routing prototype used by a shared prompt-only router. On the 8-task CT-Pipeline stream, TILA-CT achieves 65.19 acquisition and 63.98 final macro score, with 1.38 points of average forgetting. A middle-reversed order gives 65.38 acquisition and 64.17 final macro score. Across 6 backbones, final scores remain within 1.5 points of acquisition. Ablations support memory-derived routing keys and causal gathering. Final macro routing accuracy is 93.9\%, and oracle routing recovers 1.21 final macro points. These results separate preserved task computation from learned access to it.

0.11

How Efficient Is Your Agent? Online Predictive Bounds under Budget Constraints

cs.LG ·2026-11-08 ·Roy Maor Lotan, Shai Feldman, Yaniv Romano
theory
Abstract

We evaluate AI agents in an online setting where tasks arrive sequentially and, before running each task, we construct an upper predictive bound on its \emph{time-to-success}---the number of interactions until the task is completed successfully. Under a limited budget, some runs are stopped early, so their time-to-success is unobserved (right-censored), and the correctness of the bound may be unknown. We introduce \emph{\textbf{PR}edictive Bounds for \textbf{O}nline \textbf{B}udget \textbf{E}fficiency} (\texttt{PROBE}), a method for constructing calibrated upper predictive bounds while adaptively allocating a limited budget across agent tasks to reduce the anticipated coverage deficit. Our method calibrates the bounds online using inverse-probability-weighted feedback together with a correction designed to minimize coverage deficit while remaining unbiased. We prove that \texttt{PROBE} provides a coverage guarantee for any finite time horizon while satisfying the budget constraint. Experiments on software engineering, retrieval-augmented generation, and instruction following show that \texttt{PROBE} attains a coverage rate closer to the target level with tighter predictive bounds than a calibrated baseline.

0.10

JoMoHa: Joint Model--Harness Optimization with Reinforcement Learning

cs.LG cs.AI ·2026-11-08 ·Xudong Wu, Guangchen Lan, Mingqi Yuan, Jiayu Chen
Abstract · 10 pages

The performance of language agents depends jointly on model weights and the harness. Optimizing either alone can be constrained by the other component, limiting overall performance gains. Our analysis of naive alternating training reveals progressive incompatibility between the model and harness, which increasingly interferes with task execution and may also undermine model training, leading to performance degradation. To coordinate the evolution of models and harnesses, we introduce JoMoHa, to our knowledge the first framework to jointly train an executable harness generation policy and a task execution policy through reinforcement learning. JoMoHa outperforms the evaluated weight-only and harness-only baselines and achieves the highest success rates among the evaluated methods on ALFWorld, WebShop, and DBBench, averaging 70.3% and reaching 97.0% on ALFWorld. By learning both task execution and runtime improvement from its own generated trajectories, JoMoHa performs recursive self-improvement (RSI).

0.09

Learning through Capacity Awareness to Minimize Hospital Readmissions

cs.LG cs.CYstat.APstat.ML ·2026-11-08 ·Lydia T. Liu, Kara Schechtman, Pengyi Shi, Angela Zhou
theory
Abstract

Predictive algorithms are increasingly used to guide hospital discharge decisions, yet these models can exhibit censoring bias for vulnerable patients due to insufficient historical data or past discharge decisions. When biased readmission risk predictors are used to inform discharges, the training data generated by discharge decisions may perpetuate similar biases when predictors are re-trained, leading to a vicious cycle. We extend a queueing-based Markov decision process for hospital discharge to incorporate periodically retrained readmission risk predictors. Our key insight is that cost-optimal, "capacity-aware" operational decision policies can naturally diversify the lengths of stay at which patient outcomes are observed, enabling consistent estimation of readmission risk without explicit fairness constraints. We prove that repeated retraining under discharge-up-to policies converges to unbiased risk estimates under well-specification. In synthetic experiments and calibrated simulations based on patient-level data from a partner hospital, we find that where commonly used threshold-on-risk policies can remain persistently biased, discharge-up-to policies reduce learning error over status quo decision policies (such as thresholding on risk) by 24%-25%. These improvements to estimation also mitigate some forms of bias, reducing readmission cost disparities (adjusted for baseline risk differences between demographic groups) by 38%-63%. Our results demonstrate how capacity-aware decision policies can improve both predictive performance and fairness by addressing censoring bias.

0.09

Actions Should EARN Their Credit: Test-Time Adaptation and Route Consolidation for Frozen LLM Agents

cs.LG ·2026-11-08 ·Bian Sun, Kevin Zhai, Mubarak Shah, Song Wang
Abstract

LLM agents can use feedback from repeated attempts to improve their decisions without updating model parameters. However, positive action credits can continue to favor actions that repeatedly achieve the same score, even when higher scores are achievable. Under a fixed interaction budget, trying new actions in the final attempts can also lower performance. We introduce EARN, a method that uses past experience to guide further progress and retain earlier performance through conditional reuse of effective action sequences. EARN reduces action credits for reaching scores repeatedly attained in earlier episodes and assigns larger credits to action segments that reach higher scores. These credits adjust the probabilities of LLM-proposed actions. During the final episodes, EARN directly replays stored actions when the current context matches the recorded context and the actions remain available, bypassing LLM candidate generation. We evaluate performance at the end of the interaction budget using the average score over the final four episodes. Across Jericho, ScienceWorld, and Crafter, EARN achieves the best score on this metric in 31 of 33 task--budget settings among the evaluated methods, including ties. Across the evaluated Jericho budgets, replay bypasses 78.9--86.3\% of candidate-generation calls in the final four episodes, averaged across games.

0.09

Nonasymptotic causal structure learning in generalized linear DAGs with continuous optimization

cs.LG stat.MEmath.STstat.ML ·2026-11-08 ·Chang Deng, Inbeom Lee, Bryon Aragam
Abstract

We study the problem of learning directed acyclic graphs from observational data in nonlinear models and provide finite-sample, nonasymptotic statistical guarantees in these models. Although widely studied, statistical guarantees for gradient-based methods in nonlinear models are missing from the literature. Under a generalized linear model for each node, we analyze a global nonsmooth objective combining an exact acyclicity constraint with a quasi-minimax concave penalty, which can be optimized via standard gradient-based solvers. We establish finite-sample guarantees under this model, including the most widely used logistic and Poisson models. Experiments corroborate structure recovery accuracy and computational performance, including both standardized data and negative controls as baselines.

0.08

MADSNet: A Multi-Scale Axial-Attentive Dense Spiking Network for Spike Camera Image Reconstruction

cs.LG cs.AI ·2026-11-08 ·Sirui Li, Rui Zhao, Siqi Cai, Haizhou Li, Tiejun Huang +1 more
Abstract

Spike cameras encode dynamic scenes into high-temporal-resolution binary spike streams for efficient high-speed vision. Although spiking neural networks (SNNs) can process such signals with sparse and accumulate-only computation, existing SNN-based reconstruction methods are typically shallow and still lag behind ANN-based models in reconstruction fidelity. This paper proposes MADSNet, a Multi-scale Axial-attentive Dense Spiking Network for spike camera image reconstruction. MADSNet adopts a hierarchical spiking encoder--decoder, where a multi-scale spike stem extracts early features, axial spike attention captures long-range dependencies, a dense spiking bottleneck enhances feature reuse, and a spike-to-continuous decoder reconstructs continuous intensity images from spike-domain representations. Experiments on gray-scale and color spike reconstruction show that MADSNet outperforms SSIR by 4.3 dB PSNR and Spk2ImgNet by 1.94 dB on SREDS. Compared with Spk2ImgNet, it achieves about 15$\times$ lower latency and uses only about 1/40 of the estimated energy, demonstrating that carefully designed SNN architectures can improve reconstruction quality while preserving computational efficiency.

0.08

Prune to Probe: Causal Decision-Stability Probes for Efficient Reasoning

cs.LG ·2026-11-08 ·Yufei Qi, Chonghe Jiang
Abstract

Reasoning models often continue generating after producing an answer, yet stopping prematurely may interrupt useful verification or revision. In this work, we investigate whether targeted interventions in a model's internal computation can provide a reliable signal for when to stop reasoning. We introduce Prune-to-Probe, an inference-time controller that uses causal activation interventions to probe the stability of a model's current answer. Specifically, we identify late-MLP coordinates that contribute to post-answer continuation while excluding those important for answer completion. Once an answer is stated, Prune-to-Probe temporarily suppresses the selected coordinates and measures the resulting stopping response relative to matched random interventions. Crucially, these interventions serve only as probes: the model either returns its existing answer or resumes generation from an untouched cache. On MATH, Prune-to-Probe reduces reasoning-token forwards by 2.60%, preserving all naturally correct trajectories while repairing six errors. On MMLU-Pro, it achieves 1.145% pooled savings with no observed harm. Further analysis shows that the intervention captures decision stability rather than correctness (AUROC 0.502). These results demonstrate the potential of causal activation probes for efficient inference-time control without permanent model modification.

0.07

Data-informed Adaptation of Foundation Models for Causal Effect Estimation

cs.LG cs.AI ·2026-11-08 ·Chunwei Ma, Russell D Wolfinger
benchmark
Abstract · Preprint

Due to the fundamental problem of causal inference -- the impossibility of observing counterfactual outcomes -- there is always a gap between the training goal (outcome, propensity) and the inference target (CATE, ATE). Traditional causal effect estimation revolves around setting assumptions, selecting identification strategies, and choosing causal estimators, while more recently, Causal Foundation Models (CFMs), built on Prior-Data-Fitted Networks (PFNs), ease this process by pretraining on massive synthetic prior datasets. Despite their success, such amortization makes causal inference less transparent, obscuring which assumptions the model implicitly relies on for a given dataset. The high diversity of real-world data-generating processes (DGPs) and Structural Causal Models (SCMs) makes it hard to know, a priori, which causal inference strategy is compatible with a given dataset. In this work, we focus on this compatibility between an observational dataset and method-specific assumptions, and find that geometric characteristics of data embeddings -- specifically, the variogram -- provide a useful signal of whether a given causal estimation strategy is appropriate. We develop MoE-Learner, a mixture-of-causal-experts framework that embeds a dataset under multiple assumption-specific views and uses the variogram as a counterfactual-free router to assess each expert's validity. Rather than eliminating causal assumptions, this shifts their validation from a fixed, a priori choice to a criterion that can be checked directly from data geometry. MoE-Learner achieves the best average rank on benchmark datasets (IHDP, ACIC, Lalonde). Being orthogonal to (causal) foundation models, our method builds a bridge from data characteristics to causal estimation strategy rather than proposing another CFM.

0.06

Neural Activity Synchronization-Based Fast Local Learnable Spiking Neural Network Architecture

cs.NE cs.AI ·2026-11-08 ·Pranith Dutta, Sk Wasim Rajjak, Siddhanta Ray, Santlal Prajapati, Manobendra Nath Mondal
Abstract

Implementing intelligent learning systems on low- power edge devices is challenging due to the high compu- tational complexity, training latency, and energy demands of traditional Artificial Neural Networks (ANNs). Spiking Neural Networks (SNNs), biologically plausible and spike-driven, come as an alternative. However, SNN training commonly relies on backpropagation through time (BPTT), which is computationally expensive, inherently global, and unsuitable for battery-driven devices. This work proposes a novel local learning framework for multi-layered SNNs inspired by Neural Activity Synchronization (NAS). Each layer updates its own weights using locally com- puted gradients. No gradient backpropagation is needed. This significantly reduces computational complexity, training latency, and energy consumption. The proposed SNN model is trained and evaluated on the MNIST dataset, achieving a classification accuracy of 97.78%. The model is also analyzed with Gaussian and salt-and-pepper noise, demonstrating robust performance across different noise levels.

0.06

Where Should a Gate Look? Input vs. State Gating in Diagonal State Space Models

cs.LG cs.AI ·2026-11-08 ·Riccardo Zattra, Giacomo Baggio, Leonardo De Marchi, Augusto Ferrante, Francesco Ticozzi +2 more
Abstract

Modern state-space models owe much of their computational efficiency to their linear, often diagonal, dynamics. Classical recurrent neural networks instead allow nonlinear state-to-state interactions, which can be beneficial when memory updates need to depend on the past context. In this paper, we ask whether state feedback through a simple gate regulating memory updates can enhance the expressivity of diagonal state-space models. To investigate this question, we compare gates that depend on the current input, the recurrent state, or both. We establish a strict expressivity separation: for scalar models, state gating solves induction-head tasks for arbitrary vocabulary size and sequence length, whereas input gating fails on a finite instance under standard stability assumptions on the state transition matrix. Experiments show that state gating retains its advantage on induction-head tasks in higher-dimensional models, whereas input gating performs better on selective copying tasks. Together, our results show that even a simple, diagonal form of state-feedback gating can unlock capabilities unavailable to input-controlled models, while input and state gating offer complementary strengths and achieve the strongest overall performance when combined.

0.06

Language, Vision & Media 105

An Adaptive, Context-Aware, and Stacked Attention Network-Based Recommendation System to Capture Users' Temporal Preference

cs.IR cs.LG ·2026-11-08 ·Jung-Hsien Chiang, Chung-Yao Ma, Chi-Shiang Wang, Pei-Yi Hao
top venuecode/datahighly cited
Abstract · 12 pages. Accepted version of the article published in IEEE Transactions on Knowledge and Data Engineering. Code: https://github.com/solitud…

Recommendation systems have become more important since the widespread use of the Internet. The large amount of information means that it is difficult for users to discover they really need. However, users' preferences can change over time because of the age and the impact of social networks. Recommendation systems must capture users' preferences and recommend suitable items, which is a difficult challenge. This study proposes a novel recommendation system that adapts to the changing preferences of users. By learning and adapting to users' changing preferences better recommendations are provided. This study uses context factors as additional information to model users' preferences more precisely. For the proposed recommendation system, the context is based on users' interaction items within each defined time interval. This study proposes two novel attention mechanisms for the recommendation system. The contextual item attention module captures contextual information, changing pattern in users' preference and the importance of items. The multi-head attention module extends the diversity of users' preferences and adapts to changing preferences. The recommendation performance is improved using additional item's temporal information to model the contextual item's representation. Experiments compare the proposed algorithm with several state-of-the-art recommendation methods using three real-world datasets. The experimental results demonstrate that the proposed context-aware recommendation model outperforms traditional methods and demonstrate the effectiveness with which contextual information is captured by the attention mechanism.

0.79

MICE: Multi-animal Interaction Context Encoder - A Hierarchical Foundation Model for Mouse Behavior

cs.CV cs.LGq-bio.NC ·2026-11-08 ·Yen-Shuo Su, Weihan Li, Anqi Wu
top venuecode/data
Abstract · NeurIPS 2026 (Spotlight). 25 pages, 4 figures. Code: https://github.com/eric0708/MICE

Quantifying mouse behavior from pose is a foundational tool in neuroscience, ethology, drug discovery, and animal welfare. Self-supervised foundation models are a natural fit, but a foundation model for mouse behavior must handle social behavior - many behaviors that matter (aggression, courtship, mounting, allogrooming) are inherently multi-animal. Existing self-supervised pose models fall short: they encode each animal independently and discard interaction signal, hard-code the animal count, and pretrain one dataset at a time. We introduce MICE (Multi-animal Interaction Context Encoder), a hierarchical foundation model for mouse behavior that pretrains jointly on six mouse-pose corpora spanning one to four mice and 7 to 27 keypoints. MICE combines a hierarchical individual encoder capturing per-mouse kinematics at multiple temporal scales with a Perceiver-style social encoder whose learned latent codebook decouples parameter count from animal count, so a single checkpoint serves any group size without retraining. Across four evaluation protocols spanning six datasets, MICE matches or exceeds prior keypoint-based baselines, and layer-wise probing shows the social stage incrementally improves discriminability beyond the individual encoder. Code, weights, and data are available at https://github.com/eric0708/MICE.

0.56

Saidzoda-CA-EmbeddingGemma-300M: One Small Embedder for Tajik, Kazakh, Uzbek and Kyrgyz

cs.CL ·2026-11-08 ·Tohir Saidzoda, Nekruz Maxudov, Bedil Karimov, Abubakr Abdullozoda, Amin Jalilov +3 more
code/data
Abstract · 12 pages, 6 tables, 2 figures. Technical report. Model: https://huggingface.co/Tohirju/Saidzoda-CA-EmbeddingGemma-300M

The small open embedder EmbeddingGemma-300M, used as released, finds the right passage first for only 51.7% to 60.8% of Belebele questions in Tajik, Kazakh, Uzbek and Kyrgyz. We present Saidzoda-CA-EmbeddingGemma-300M, one model for all four languages. It is our Tajik fine-tune of EmbeddingGemma-300M, trained for one more epoch on 279,904 pairs: questions in Kazakh, Uzbek and Kyrgyz that our own language model wrote about source passages, a Tajik refresher, and English and Russian replay. On Belebele, recall@1 rises by 14.4 to 26.0 points in each language, to between 75.2% and 77.7%. Against single-language models trained with the same settings, the one model stays within 2.8 points at recall@1 in Kazakh, Uzbek and Tajik, with overlapping recall@10 intervals, though it also had more training. It keeps the Tajik law recall@10 of its starting point (92.2 against 92.4) but loses 4.4 points of law recall@1. A model trained on one of these languages also improves the others on Belebele, but less than training on them. The model still trails the 560M XLM-RoBERTa embedders on Kazakh, Uzbek and Kyrgyz Belebele: the better of multilingual-e5-large-instruct and BGE-M3, used as released, is 9.4 to 10.2 points ahead at recall@1. Our scores of those released models agree with the published MTEB results within 2.4 points. Russian and English recall@1 fall by 3.0 and 4.2 points.

0.56

Martine: Benchmarking Multi-View 3D Surface Reconstruction Across Viewpoint Coverage, Resolution, and Lighting

cs.CV ·2026-11-08 ·Robin Bruneau, Baptiste Brument, Frederic Giraud, Bastian Sigrist, Sascha Jecklin +3 more
top venuebenchmarkcode/data
Abstract · Accepted at the 40th Conference on Neural Information Processing Systems (NeurIPS 2026), Track on Evaluations and Datasets. 23 pages, 14 fig…

Controlled acquisition setups are widely used for high-accuracy multi-view 3D surface reconstruction, as they allow sensor configuration, image resolution, viewpoint distribution, and illumination to be specified. Yet benchmarking remains difficult, since reconstruction accuracy depends jointly on the reconstruction paradigm, object properties, and acquisition regime. Existing benchmarks often restrict viewpoint coverage, resolution, or illumination, which can bias comparisons towards methods suited to the chosen setting. Others lack accurate 3D ground truth or quantified image-to-geometry alignment, limiting their use for metric surface evaluation. We introduce Martine, a benchmark for controlled 3D object surface reconstruction. Martine comprises 28 objects, including transparent materials and fine structures, captured at 9568 x 6376 pixels from 84 views over a spherical cap under both single-light and multi-light regimes. It provides object-independent camera calibration, accurate 3D ground-truth scans, calibration and image-to-geometry accuracy quantified on test images, and 3D property masks for analysing errors by material properties, visibility, curvature and fine structures. Experiments across photogrammetry, classical multi-view stereo, neural implicit surfaces, Gaussian-splatting-based reconstruction, and multi-view photometric stereo show that Martine enables regime-aware evaluation with explicit control over reference-data accuracy. The benchmark resolves sub-millimetric reconstruction differences, with a best observed Chamfer distance of 0.058 mm over the evaluated visible surface.

0.55

ScrapeBench: Evaluating Legal Compliance of AI Agents in Website Scraping

cs.CL cs.AI ·2026-11-08 ·Joseph Marvin Imperial, Daniel Slate, Noam Kolt
top venuebenchmark
Abstract · Camera-Ready for NeurIPS 2026 (Evaluations and Datasets Track)

AI agents deployed on the open web are becoming more capable of performing actions that may breach contractual provisions and violate computer security laws. In this paper, we introduce ScrapeBench, the first agentic benchmark for empirically evaluating the legal compliance of AI agents with website scraping policies, namely terms of service (ToS) and robots$.$txt directives. Across widely used closed-weight and open-weight AI agents, we find that most agents exhibit low compliance rates (below 5%) when instructed to scrape websites in our sample of 1,626 popular websites that explicitly prohibit scraping. In addition, we observe that even where AI agents actively check the applicable ToS and robots$.$txt, they continue to exhibit low compliance rates. Conversely, in our sample of 269 websites that do not prohibit scraping, we observe that agents sometimes refuse to scrape despite it being lawful, such as Claude Opus 4.6 exhibiting an over-refusal rate of 98.5%. Our human baseline experiments (n = 180 participants) suggest that humans exhibit compliance rates comparable to AI agents but over-refuse at much higher rates.

0.51

Not All Layers Are Equal in Image-to-Video Transfer

cs.CV ·2026-11-08 ·Thinesh Thiyakesan Ponbagavathi, Constantin Seibold, Alina Roitberg
top venuecode/data
Abstract · Accepted to NeurIPS 2026 (Main Track)

Parameter-efficient image-to-video transfer adapts frozen image foundation models to video via lightweight trainable adapters. Existing methods insert the same adapter at every transformer layer, implicitly assuming all layers require equal temporal modeling. A layer-wise sensitivity analysis across four video-native model families and two pretraining paradigms reveals that this assumption is wrong: temporal sensitivity increases with depth, and the pretraining paradigm determines its growth profile. Vision-only models exhibit a gradual rise, while vision-language models show early suppression followed by late recovery. This finding motivates STRAP, a training-free adapter placement rule that derives a spatial-to-temporal transition boundary from a paradigm-matched video model. Spatial adapters are assigned before this boundary and temporal adapters after it, without modifying the adapter architecture itself. On four fine-grained action recognition benchmarks, STRAP improves accuracy by up to 11.8% over uniform placement, reduces trainable parameters by 29 to 48%, and lowers inference time by up to 10%. On standard benchmarks, it matches accuracy with fewer parameters. Code is available at https://github.com/th-nesh/STRAP.

0.50

Rethinking Geometric Depth in Monocular 3D Object Detection: A Projection-Consistent Reformulation

cs.CV ·2026-11-08 ·Zhihao Zhang, Abhinav Kumar, Huaizhi Qu, Tianlong Chen, Xiaoming Liu
top venuebenchmarkcode/data
Abstract · Accepted to NeurIPS 2026

Monocular 3D detection (Mono3D) hinges on recovering depth from a single image, for which the geometric formulation $z \approx f \cdot H / h_{2D}$ is widely adopted due to its simplicity and direct grounding in projective geometry. We show that this formulation harbors a systematic, projection-induced bias: substituting the 2D bounding-box height $h_{2D}$ recovers the depth of an object's visible front surface rather than its center, causing size-dependent underestimation that persists even under ground-truth inputs. We address this front-surface bias with MonoPCA, a unified framework built on a projection-consistent reformulation. At its core, PCDepth retains the measurable $h_{2D}$ and closes the gap to center depth through an orientation- and size-aware correction derived from perspective geometry, eliminating the bias without introducing ill-posed regression targets. Because PCDepth couples depth to predicted object size and orientation, attribute perception becomes essential. We therefore introduce GeoAlign, a teacher-student framework that transfers geometric representations from Vision Foundation Models into a lightweight student, strengthening attribute perception at no inference cost. Across four challenging benchmarks, MonoPCA achieves SoTA performance, including a +1.44% AP$_{3D}$ gain on the KITTI leaderboard and +2.80% on Omni3D$_{OUT}$. Code is available at https://github.com/alanzhangcs/MonoPCA.

0.49

DuSS-PET: Self-Supervised Low-Dose PET Reconstruction Using Pseudo-Dual-Domain Consistency Without Paired Data

cs.CV ·2026-11-08 ·Shuijin Lin, Yuzhu He, Yongbo Wang, Jianhua Ma, Fan Wang
top venuecode/data
Abstract · Author accepted manuscript. Published in IEEE Transactions on Radiation and Plasma Medical Sciences (Early Access, 2026)

Positron emission tomography (PET) is pivotal in medicine and healthcare, but requires high radiation doses for quality imaging. Although deep learning has shown promise for low-dose PET reconstruction/enhancement, existing methods primarily rely on fully supervised learning, which requires paired low- and full-dose scans that are clinically impractical to obtain. To overcome this fundamental limitation, we propose DuSS-PET, a novel self-supervised PET reconstruction framework that operates without any paired low- and full-dose data. Our method is built upon a computational Noisier2Noise (Nr2N) paradigm, featuring a Bernoulli corruption strategy that serves as an image-space approximation to emulate low-count noise patterns in image-derived pseudo-sinogram representations. The DuSS-PET framework integrates three core components: an adaptive Swin transformer for pseudo-sinogram restoration, a plug-and-play module for image enhancement, and a consistency-constrained SSL strategy that enforces agreement between the pseudo-sinogram and image representations. Extensive experiments on multi-center datasets demonstrate that DuSS-PET achieves superior reconstruction performance, outperforming state-of-the-art SSL methods and achieving comparable or even better results than fully supervised approaches across various dose levels. In particular, it exhibits exceptional generalization across different scanner vendors. This work is among the first self-supervised low-dose PET reconstruction framework that enforces consistency between reconstructed images and pseudo-sinograms, where the pseudo-sinogram generated in the image space serves as a solid computational consistency constraint for enhanced imaging without full-dose training scans. The source code will be publicly available at https://github.com/ladderlab-xjtu/DuSS-PET.

0.48

IntentLens: Grounding Underspecified Multimodal Queries for Recommendation via Tool-Augmented Reasoning

cs.IR ·2026-11-08 ·Xiao Chen, Dong Fang, Haitao Li, Qing Li
published
Abstract · NeurIPS 2026

As intelligent personal assistants are increasingly deployed in real-world scenarios, recommendation paradigms are shifting from passive preference matching to inferring user intent from underspecified multimodal inputs. In practical interactions, users typically issue a reference image paired with a vague natural language query, keeping key visual attributes, personalized preferences, and item-side constraints implicit. Existing recommendation methods either leverage historical user interactions augmented with multimodal item features or assume fully articulated textual queries; both paradigms struggle to resolve underspecified multimodal queries for accurate item matching. To address this limitation, we propose IntentLens, a tool-augmented recommendation framework that grounds ambiguous multimodal queries into query-relevant evidence prior to user-item matching. IntentLens follows a ground-then-rank pipeline: a shared multimodal LLM leverages visual tools, user memory modules, and item attribute retrieval tools to (i) infer latent user intent from image-text inputs and (ii) augment each candidate item with fine-grained, query-conditioned evidence. Subsequently, a lightweight ranker scores candidates based on the grounded representations. To facilitate research in this novel setting, we construct two benchmarks with underspecified multimodal queries, namely \textsc{GoogleReview-MMTool} and \textsc{Yelp-MMTool}. Extensive experiments on these benchmarks demonstrate that IntentLens outperforms competitive MLLM and retrieval baselines, and exhibits strong generalization to unseen categories and datasets.

0.48

Virtual Task Prompting for Multi-Task Scene Understanding

cs.CV ·2026-11-08 ·Zedong Wang, Yucheng Wang, Dan Xu
top venuecode/data
Abstract · Accepted at NeurIPS 2026. 24 pages, 8 figures, 5 tables

Multi-task scene understanding models typically rely on modified vision encoders or complex decoders for cross-task interaction, which limits their flexibility and compatibility with off-the-shelf vision foundation models (VFMs). We introduce Virtual Task Prompting (VTP), which frames multi-task prompting as retrieving task-adaptive compositions from a shared virtual state, where both the state and routing process are optimized in an end-to-end manner. Concretely, VTP maintains an expanded set of virtual tasks alongside the backbone. At each layer, learnable routing matrices read task-specific compositions as input to interact with image tokens and write the processed results back, forming explicit pathways that capture the intricacies among tasks outside the encoder. Since VTP keeps the backbone clean, it integrates seamlessly with pretrained VFMs, advancing the state of the art on Pascal-Context and NYUD-v2. Stress test on Taskonomy with 13 heterogeneous tasks further shows VTP's robustness in extreme settings. Ablations and analysis indicate that VTP's gains stem from adaptive routing rather than increased capacity. The project page is available at https://jacky1128.github.io/VTP/.

0.47

Conditional Diffusion-Based Proxy Representation Recovery for BCVA Prediction with Missing SLO Modality

cs.CV ·2026-11-08 ·Chuyi He, Hua Zou, Fei Luo, Zixian Yang
publishedhot direction
Abstract · 8 pages, 3 figures, 3 tables. Accepted at IEEE BIBM 2026

Preoperative prediction of postoperative best-corrected visual acuity (BCVA) can support prognosis assessment and individualized decision-making in cataract surgery. Multimodal learning can integrate preoperative optical coherence tomography (OCT) images, scanning laser ophthalmoscopy (SLO) images, and clinical text to characterize patient status. However, SLO images are often missing in real-world clinical data due to acquisition or data-management issues, resulting in incomplete multimodal inputs. To address this problem, this paper proposes a conditional diffusion-based method for SLO proxy representation recovery and fusion prediction. Instead of reconstructing raw SLO images, the proposed method formulates SLO missingness as proxy representation recovery in the latent feature space. Specifically, OCT image features and clinical text features are first fused to construct an OCT-text conditional representation, which guides the reverse denoising process of the diffusion model to estimate sample-specific SLO proxy representations. The recovered SLO proxy representation is then used as a substitute for the unavailable SLO modality and incorporated into the downstream multimodal fusion prediction network for postoperative BCVA prediction. Experiments were conducted on 2,141 preoperative multimodal cataract samples, with an SLO missing rate of 58.80%. The proposed method achieved an accuracy of 0.9714 under the criterion of an absolute prediction error within 0.1 LogMAR and a mean absolute error of 0.0423, outperforming representative postoperative visual acuity prediction and missing-modality learning methods. These results indicate that representation-level proxy recovery can alleviate the impact of SLO modality missingness on postoperative BCVA prediction.

0.44

What Travels Across Languages? Reading, Transfer and Erasure of Cross-Lingual Concept Subspaces in Bangla Idioms

cs.CL cs.LGstat.ML ·2026-11-08 ·Syed Rifat Raiyan, Yousuf Ibne Kamal Niloy
code/data
Abstract · 73 pages, 21 figures, 52 tables. Code, frozen analysis plans and aggregate results: https://github.com/Starscream-11813/bangla-idiom-subspac…

Cross-lingual concept analysis requires distinguishing the information that a subspace retains from the classifiers and erasers that transfer through it. We study these operations in six frozen encoders using 10,359 Bangla (Bengali) idiom senses, eight input conditions, and figurative-class and sentiment labels. Class-mean directions provide the most consistent evidence of shared coarse meaning: in post hoc analyses, they outperform random-subspace, permuted-label and covariance-matched nulls in all 12 model-label tests under balanced accuracy. A prospective internal re-split supports idiom-alone sharing in 10 of 12 tests, spanning five encoders. The original confirmatory logistic-regression family rejects both planned nulls in 4 of 12 tests, where English directions gain only a few accuracy points above the majority-class floor (3.0-7.3). Classifier transfer depends on the input map and regularization: English-trained class-mean readouts beat both zero-shot nulls in 8 of 12 discovery tests, and stronger source-selected penalties improve transfer through logistic and LEACE directions. These readouts use Bangla-validation-selected layers and unlabeled target preprocessing. Input controls show greater coarse-class sharing for explicit glosses than for idiom counterparts. Reading, however, does not imply effective erasure: English-fitted LEACE removes less than one percentage point of Bangla figurative-class accuracy, compared with 4-11 points for Bangla-fitted LEACE; INLP shows a similar asymmetry. Pattern-filter analysis, covariance-weighted input maps and a binary Gaussian overlap bound explain how these operations can diverge. The analogous multiclass ordering is an empirical diagnostic, observed in 11 of 12 cells. The results support an estimator-sensitive account of cross-lingual concept sharing and motivate evaluating reading, classifier transfer, erasure and overlap separately.

0.43

AgentCT: A Vision-Language Agent for Universal CT Imaging

cs.CV ·2026-11-08 ·Zhihao Chen, Qi Gao, Chuang Niu, Qiegen Liu, Yi Zhang +1 more
code/datahot direction
Abstract · 14 pages, 5 figures

Computed tomography (CT) restoration methods often use a single network for low-dose CT (LDCT), sparse-view CT (SVCT), and metal artifact reduction (MAR), despite the distinct physics of these degradations. We propose AgentCT, a vision-language agent that identifies the degradation and routes images to task-specific restoration models. A LoRA-adapted Qwen3-VL-4B controller predicts degradation type and severity, selects a specialist, and reassesses the restored image. The LDCT specialist uses an image-domain diffusion model, whereas the SVCT and MAR specialists combine projection-domain correction, data consistency, differentiable filtered back-projection, and image refinement. We evaluated the controller and specialists separately on patient-disjoint Mayo 2020 abdominal CT data across three severity levels for each degradation. The controller achieved 95.7% accuracy for degradation type and 93.9% for severity, with a 100% JSON parse rate. Using ground-truth degradation labels for routing, AgentCT achieved an overall PSNR/SSIM of 46.18 dB/0.984, compared with 44.93 dB/0.976 for AMIR and 44.29 dB/0.979 for TAMP, and attained the highest PSNR in all nine settings. For SVCT, its dual-domain specialist achieved 46.04 dB with 4.2 million parameters, compared with 44.43 dB for single-task AMIR and 43.57 dB for all-in-one AMIR with 23.5 million parameters. These results support degradation-specific CT restoration with controller-based model selection; evaluation under automatic routing and on independent clinical data remains necessary. Code is available at https://github.com/hao1635/AgentCT.

0.42

Beyond Hard Labels: LLM Probabilities Improve Human-Assisted Regression

cs.CL ·2026-11-08 ·Xinyue Wang, Qunhua Li
code/data
Abstract · 22 pages, including references and appendices

Large language models (LLMs) are increasingly used for text annotation, but their outputs are often reduced to a single categorical label before downstream analysis. This hardening step discards the model's relative support across categories, which may contain useful information when category boundaries are ambiguous. We study whether retaining the full probability vector improves downstream regression when LLM annotations are combined with a limited amount of human annotation. Using matched soft and hard representations from three LLMs across three social-science text datasets, we compare direct regression and human-assisted correction. Soft inputs are not uniformly better when used directly as outcomes, but after human correction, using the probability vector consistently improves regression efficiency across models and datasets. Analyses of repeated human judgments and controlled simulations further show that these gains arise when probability variation beyond the hard category carries useful information about the human-coded outcome. Our results identify output representation as an important design choice in LLM-assisted measurement and support preserving probability outputs for downstream statistical analysis. Code and reproducibility materials are available at https://anonymous.4open.science/r/soft-label-LLM-D201/

0.42

Disentangling Where and When:Factored Spatio-Temporal Explanations for Video Action Recognition

cs.CV cs.AIcs.LG ·2026-11-08 ·Bhavan Vasu, Giuseppe Raffa, Prasad Tadepalli
top venue
Abstract · Accepted at The Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS 2026)

Post-hoc explanation methods for video action recognition typically conflate spatial appearance and temporal dynamics into a single saliency or importance map. We propose a perturbation-based framework that disentangles these contributions into a spatial appearance mask ($M_a$) and a temporal motion mask ($M_m$). The two masks are jointly optimized against a frozen action recognition model through factored perturbation architecture, with nearest frame replacement to identify important temporal dynamics. A separation penalty discourages the appearance mask from spending budget on frames that the motion mask has already discarded. A ReLU-based area constraint penalizes explanations larger than an assigned budget. To evaluate these explanations, we introduce a separation protocol that sweeps spatial pixels and temporal frames independently, backed by proofs characterizing the proposed method. Across EPIC-Kitchens-55, EGTEA Gaze+, and Something-Something V2, our method attains the lowest deletion AUC against leading baselines while producing masks roughly five times sparser.

0.39

Dropping Tokens Without Dropping Them: Causal Attribution for Diabetic Retinopathy

cs.CV ·2026-11-08 ·Pramath Girish Haritz, Chandra Sekhar Seelamantula
code/data
Abstract · 4 pages, 2 figures. Submitted to ISBI 2027

In this work, we investigate whether attention-based attribution maps in a vision transformer provide causal explanations for diabetic retinopathy (DR) classification. We introduce a principled decomposition of attention into two attribution maps: the contribution ($c_t$), which satisfies the completeness axiom in explainable artificial intelligence (xAI), and the pull ($p_t$), which quantifies causal explanations. We define an identity relating $c_t$ and $p_t$ that characterizes when $c_t$ fails to capture causality. We test whether each map is causally faithful on the MAPLES-DR dataset by dropping/inserting tokens, and report Necessity ($N$), Sufficiency ($Su$), and Faithfulness Correlation ($FC$). The pull $p_t$ achieves the highest causal alignment ($FC = 0.53$, $N = 0.71\%$, $Su = 3.0\%$), outperforming Attention Rollout, Grad-CAM, and AttnLRP. Occlusion retains better sufficiency ($Su = 1.6\%$) but is about 800 times slower. Furthermore, on MAPLES-DR, $p_t$ achieves $70\%$ of the average precision (AP) of HRDecoder, a state-of-the-art supervised lesion segmentation model, despite the absence of pixel-level supervision. Together, these findings show that the pull $p_t$ and the contribution $c_t$ form computationally efficient attribution maps for reliable causal interpretation in safety-critical medical diagnostics. The code and demo are available at https://github.com/pramath-haritz/dr-xai.

0.39

BiRefNet-V2: Rethinking Bilateral Reference for High-Resolution Dichotomous Image Segmentation

cs.CV ·2026-11-08 ·Peng Zheng, Deng-Ping Fan, Binglu Wang, Nicu Sebe
code/data
Abstract · Submission version

Dichotomous image segmentation requires pixel-accurate foreground-background separation while preserving fine structures such as thin filaments, sharp edges, and intricate textures. BiRefNet addressed this challenge through bilateral references, combining image patches and gradient information, but its inward reference loses local spatial structure and lacks semantic guidance, while its outward reference relies on single-channel gradient attention. We recast this paradigm as controlled information recovery - locality determines what spatial information survives, semantic selectivity what information enters, and focused boundary supervision where refinement occurs - and propose BiRefNet-V2 to implement it: (1) locality-preserving and semantically guided detail injection through space-to-depth transformation, overlapping patches, and a Semantically Guided Gate; (2) a learned boundary stream with focused supervision that generates full-channel residual attention without requiring boundary annotations; and (3) a dual-branch decoder integrating depth-wise boundary convolutions with large-kernel attention. A modernized framework - LayerNorm, GELU, the Lion optimizer, and a DINOv3 backbone - cuts training from 600 to 50 epochs. Extensive experiments demonstrate state-of-the-art performance on DIS5K, camouflaged object detection, and high-resolution salient object detection across all 13 evaluation sets. Code is available at https://github.com/ZhengPeng7/BiRefNet-V2.

0.37

THEATRON: A Benchmark for Evaluating Persona Persistence in Long-Horizon LLM Decision-Making

cs.CL ·2026-11-08 ·Yuyan Wang, Yuechen Jiang, Zhiwei Liu, Jimin Huang, Sophia Ananiadou
benchmarkcode/data
Abstract

Existing evaluations often test persona expression rather than whether persona information persists in extended decision-making. We introduce THEATRON, a controlled benchmark that uses financial replay to evaluate persona persistence in long-horizon sequential decision-making. In these environments, models conditioned on fixed behavioral-finance personas observe shared daily market evidence and trajectory-specific account-state feedback and make sequential decisions over multi-month trajectories. We evaluate Zero-Shot, One-Shot, and Chain-of-Thought persona injection across four LLMs, using cumulative-return and action trajectories to diagnose within-persona stability, cross-persona separability, and action collapse. Persona conditioning produces systematic behavioral shifts beyond a no-persona baseline, and persona-level separation generally exceeds the variation induced by semantically equivalent descriptions. Nevertheless, these shifts do not reliably form coherent persona-specific policies: apparently stable separation can arise from stereotyped action collapse, while diverse policies can exhibit noisy trajectory switching. These findings expose a gap between observable persona-induced behavior and persistent, context-sensitive persona-conditioned decision policies. Code and benchmark data are publicly available at: https://github.com/yuyan12138/theatron

0.37

Beyond Similarity Entropy: A Rank-Guided Data-Free Quantization Framework for Vision Transformers

cs.CV ·2026-11-08 ·Runqing Jiang, Ye Zhang, Pengpeng Yu, Zhuo Song, Changjun Li +2 more
code/data
Abstract

Vision Transformers (ViTs) have demonstrated remarkable performance, yet their computational and memory demands pose significant challenges for deployment on mobile devices. Data-free quantization (DFQ) enables the compression of ViTs without access to their original training data. Existing DFQ methods often synthesize calibration samples by maximizing the entropy of patch similarity. However, similarity entropy describes the dispersion of similarity values without ensuring that the generated samples cover diverse feature directions, potentially limiting their effectiveness for quantization calibration. To address this limitation, we propose RankQ-ViT, a rank-guided DFQ framework for ViTs. We first observe that patch similarity matrices derived from Gaussian noise tend to have lower ranks than those derived from real images. Motivated by this observation, we introduce a block-aware self-contrastive objective that promotes feature diversity during synthetic data generation without explicitly computing matrix rank at each optimization step. We then select calibration samples based on their rank and semantic consistency and use them for block-wise reconstruction of the quantized model. Experiments on ImageNet, COCO, ModelNet40, and ShapeNet demonstrate the effectiveness of RankQ-ViT across multiple architectures and tasks, particularly under low-bit quantization. The code is available at https://github.com/Runqing-forMost/rankq-vit.git

0.36

Graphene: A Lightweight Transformer with Ising Mean-Field Refinement for Image Super-Resolution

cs.CV ·2026-11-08 ·Seungki Hong, Taekwang Jang
code/data
Abstract

This paper proposes Graphene, a lightweight image super-resolution architecture with two complementary feature refinement mechanisms: Ising-model-inspired local feature refinement using an Ising Refinement Block (IRB) and stage-history-based feature refinement using a History Attention (HistAttn) module. The IRB reuses attention output features and pre-softmax attention logits to construct signed pairwise token couplings within each local window, applies unrolled mean-field updates, and refines features through a gated residual correction. These signed couplings allow alignment and opposition between latent spins to modulate local feature refinement. HistAttn attends over the current stage output and earlier representations at the same token position and adaptively blends the current stage output with a history-weighted feature aggregate without adding spatial pairwise attention. On Urban100 at $\times4$, Graphene and Graphene-light improve over IET and IET-light by 0.24 dB and 0.38 dB, respectively, with 34.3%/13.7% fewer parameters, 44.8%/44.4% fewer FLOPs, 78.4%/71.5% lower latency, and 76.0%/85.5% lower peak memory. Real-world SR comparisons further show that Graphene recovers fine details and preserves textures across diverse degradations. Code and reference checkpoints are available at https://github.com/skethz/Graphene

0.36

How to "train" a surgeon? Benchmarking generalist agents in surgical scene understanding

cs.CV cs.AI ·2026-11-08 ·Gengyuan Zhang, Xiao Han, Shijie Zhou, Abdullah Koukash, Jingpei Wu +5 more
top venuebenchmarkcode/data
Abstract · Accepted at NeurIPS 2026, Evaluations and Datasets Track. 64 pages including appendices. Project page: https://vitagent.github.io/

Evaluating general-purpose multimodal large language models (MLLMs) for surgical scene understanding remains limited due to scarce surgical data and fragmented evaluation protocols. We address these challenges through a two-stage evaluation framework for surgical scene understanding. In Stage I, we introduce SurgKnowBench, a diagnostic benchmark that harmonizes public surgical datasets across procedures and tasks and evaluates proprietary and open-source MLLMs to map the boundary of pretrained surgical competence. SurgKnowBench reveals that current models retain partial visual priors but remain weak on relational and procedural understanding and workflow assessment. Building on this failure frontier and motivated by the scaffolded nature of surgical training, we introduce SurgAgentGym, an agentic environment where general-purpose MLLM agents can use bounded knowledge search over SurgWiki and general-purpose visual grounding for surgical tasks. SurgAgentGym evaluates whether generalist agents can improve surgical understanding at inference time through tool-mediated evidence use, without additional surgical training. Our results show that agentic inference narrows the gap mainly for perception-heavy tasks, while fine-grained localization and expert-style assessment remain persistent bottlenecks. This opens a path toward future studies of agentic surgical understanding.

0.36

From Sound to Occupancy: Privacy-Aware Audio Sensing for Buildings

cs.SD cs.LG ·2026-11-08 ·Myeongwon Chae, Seonghyuk Son
code/data
Abstract · 19 pages, 4 figures, 7 tables. Code: https://github.com/AaronChae/occupancy-audio-privacy

Occupancy-based controls regulate lighting and HVAC according to whether a space is in use. Passive infrared (PIR) sensors, the most common choice, respond to motion and can report an occupied room as vacant when its occupants remain still. Sound offers an alternative cue, since seated occupants type, turn pages and cough, but microphones raise privacy concerns. We therefore study a two-stage design: an on-device encoder maps each second of audio to a short embedding, and a server-side temporal model turns the stream of embeddings into an occupancy state. To our knowledge, no publicly available dataset pairs office audio with occupancy labels, so we evaluate the two stages separately on public data. The encoder is trained on synthetic scenes that mix empty-room recordings, room impulse responses, office sounds and read speech, and tested on unseen rooms, recordings and speakers for whether a second contains a sound made by a person. On 11 real office recordings (DCASE 2013) it reaches an AUC of 0.87, against 0.76 for log-mel statistics. Attacks on the transmitted embeddings recover at most 1% of the spoken words, against 94% from log-mel frames, but still identify the speaker among 40 in 69 to 77% of the utterances. Noise-floor normalization with adversarial training, an 8-dimensional embedding, or sending only class probabilities lowers this to 38 to 55% without hurting detection. On HPDmobile, which pairs home occupancy labels with an amplitude envelope of the audio, a causal temporal model pretrained with a hierarchical contrastive loss matches hand-crafted features (AUC 0.75) and does better when only one labelled day per home is available. These results make the two-stage design a practical basis for privacy-aware acoustic occupancy sensing, with speaker identity as the one leak still to close.

0.36

GlobalTap: Globally Diverse Beat Tracking Benchmark via Crowdsourced Tapping

cs.SD eess.AS ·2026-11-08 ·William T. Botticelli-Wells, Manuel Anglada-Tort, Elif Celen, Rena Far, Harin Lee +2 more
publishedbenchmarkcode/data
Abstract · Accepted at ISMIR 2026. 13 pages, includes supplementary appendix

Existing beat-tracking benchmarks draw from a narrow range of musical cultures, predominantly Western, and are difficult to scale due to reliance on expert annotation. We address both limits with GlobalTap, a benchmark of 200 popular-music excerpts spanning 59 countries, annotated by 746 participants recruited across ten countries. The benchmark is created using a scalable online tapping pipeline that records taps via audio for high temporal precision, aggregates pooled tap times into consensus beat annotations, and computes intrinsic quality measures that flag perceptually difficult excerpts without requiring a reference. In a separate study on a different 240-excerpt validation set, tapped by an additional 733 participants, an independent group of 130 listeners rated how well the resulting beat tracks lined up with the music, and we detected no overall difference between our annotations and existing reference annotations. Crowd-derived annotations occasionally diverged from the references in phase and metrical interpretation, yet listeners did not consistently prefer the reference versions, suggesting that crowd-derived and existing reference annotations can reflect different perceptual or structural interpretations. The disagreements were often systematic rather than random, and metrical-level choices differed across the ten recruitment countries. We release GlobalTap together with open-source pipeline code as a listener-derived resource for evaluating beat-tracking systems on culturally diverse music.

0.36

Language model capabilities fail in a shared, predictable order under compression and perturbation

cs.CL cs.AIcs.LG ·2026-11-08 ·Zhixiang Wang, Ulas Bagci
code/data
Abstract · 45 pages (34 main text with Methods and Extended Data, 11 Supplementary Information), 4 figures, 6 Extended Data figures. Code, protocols an…

Compression can preserve a language model's average accuracy while selectively degrading individual capabilities. Whether these losses follow method-specific patterns or a transferable order is unclear. We applied seven perturbation and compression methods to eight instruction-tuned models, scoring 16 tasks in four answer formats against preregistered predictions. Short-answer capabilities failed in nearly the same order under four random perturbations with different mechanisms (median correlation 0.90) and, within each model, under pruning and quantization (0.875). Intact answer margin, answer frequency and answer format predicted this order. Fragility profiles measured under random perturbation predicted which tasks and facts compression removes first and, after four-task calibration, held-out task accuracy (median absolute error 0.11-0.12, against 0.20-0.21 for uniform decline); profiles borrowed from other model families did almost as well. The order did not transfer across models under GPTQ quantization or hold for long-form tasks. Within short answers, what compression removes is predictable before deployment.

0.35

DynOLAT: OLAT Video Prediction for Controllable Dynamic Full-Body Relighting

cs.CV ·2026-11-08 ·Taoming Liu, Yinji ShenTu, Guangzhao He, Yudong Jin, Yukang Ye +5 more
top venue
Abstract · Accepted to SIGGRAPH Asia 2026 (Journal Track). Project page: https://zju3dv.github.io/DynOLAT

Accurate lighting response is essential for realistic dynamic human relighting, because shadows, highlights, and color changes should follow the target illumination in both direction and relative strength. Current learning-based relighting methods can generate visually plausible videos, but they often exhibit imprecise lighting responses: shadows and highlights may drift away from the target direction, and the output brightness may not faithfully follow the relative intensities in the target environment, making the result closer to appearance harmonization than controllable relighting. To address this issue, we propose DynOLAT, a generalizable OLAT video prediction framework for dynamic full-body humans. Instead of directly synthesizing a single relit appearance, our method predicts temporally consistent One-Light-At-a-Time (OLAT) basis videos from flat-lit inputs, and then linearly combines the predicted bases according to the target environment map. This OLAT-based formulation makes the directional response explicit in the predicted bases, while the relative lighting strength is determined by the linear weights derived from the environment map. We further introduce a dynamic capture pipeline and the first large-scale dynamic full-body OLAT dataset, comprising over 24 million frames from 224 identity-outfit combinations, where different outfits of the same person define separate combinations. Experiments show that DynOLAT responds more faithfully to variations in target lighting direction and relative intensity than the tested baselines, enabling controllable video relighting while preserving fine appearance details and temporal coherence. The dataset and processing tools will be publicly released to foster future research.

0.35

Room to Reconsider: Reply Length Decides Whether Pushback Makes Language Models Sycophantic or Self-Correcting

cs.CL cs.AI ·2026-11-08 ·Ook Lee
Abstract · 10 pages, 5 figures, 2 tables. Code: https://doi.org/10.5281/zenodo.24045308

Language models often abandon a correct answer when a user pushes back without giving a reason, and this behavior is usually studied as sycophancy to be suppressed. The same pushback, however, also leads models to revisit and sometimes repair a wrong answer. We measure both effects on multiple-choice questions from MMLU and ARC-Challenge with four open models from two families (Qwen3 1.7B, 4B and 8B, and OLMo-2 7B), reading the final answer exactly from option logits after a free-form reply. The reply budget changes the verdict. When Qwen3-4B may answer the challenge in at most 96 tokens, pushback leaves accuracy unchanged or slightly lower; with 1,024 tokens, it corrects 43% of its wrong answers and pushback raises accuracy by 5.7 points. Whether pushback helps overall follows a simple break-even rule in the cave and correction rates. Inside the model, caving can be partly predicted before the challenge: hidden states add about 0.04-0.06 AUROC over turn-1 confidence in three of our four probing runs. A probe trained only on self-correction cases predicts caving beyond confidence when replies are truncated, but not when the model can finish its reasoning. Activation steering never reduced caving selectively: where it worked, it suppressed self-correction as well. We conclude that sycophancy evaluations should allow full replies and report self-correction alongside caving.

0.34

NuJEV: Structured Visual Judgments for Consistent Nuclei Quantification

cs.CV ·2026-11-08 ·Yonghuang Wu, Pengfei Wang, Zhenyang Liang, Yize Ma, Pengfei Song +2 more
Abstract

Nuclear quantification requires locating and typing nuclei, beyond generating a count. Inspired by Jev's structured decisions in place of free-form generation, we introduce NuJEV for nuclear perception in Qwen3-VL-2B. We jointly train visual LoRA and a parallel head to predict nuclear presence, centers, and types. One image encoding feeds both the nuclear head and original language decoder. The decoder interprets region and type requests, with counts computed from the predicted nuclei. On 2,732 PanNuke test images, NuJEV achieves a five-query mean absolute error ($\mathrm{MAE}_5$) of 1.844 overall and 5.268 on dense images, with localization/typed F1 of 0.816/0.682. A matched encoder ablation shows 41.0\% lower dense error with visual adaptation. For language requests, NuJEV reduces request MAE by 76.2\% relative to direct generation. Our code will be released as open source.

0.34

SurgTAP-3D: A Large-Scale Cross-Surgery Video Dataset with Contact-Aware 3D Motion Tracking

cs.CV cs.RO ·2026-11-08 ·Jiaxin Guo, Wenzhen Dong, Taoyu Wu, Haomin Kuang, Mengya Xu +7 more
publishedbenchmark
Abstract · Accepted to IEEE Transactions on Robotics (T-RO)

Understanding 3D scene dynamics in minimally invasive surgery (MIS) is a fundamental step toward physically intelligent and autonomous surgical systems. However, existing public datasets offer limited scalability and diversity in real-world clinical scenarios. Moreover, annotations typically focus on either tissue or instruments alone, overlooking tool-tissue interactions. This gap limits applications such as augmented reality guidance and autonomous surgery driven by vision-language-action (VLA) models or world models, which demand large-scale, high-quality 3D supervision. In this work, we construct SurgTAP-3D, a large-scale in-vivo pseudo-3D surgical motion tracking dataset and benchmark, comprising 520K frames, 4,549 action clips, and 1.4M 2D/3D point trajectories from diverse robotic and manual procedures under unified action categories. To make annotation scalable, we design a unified framework for surgical 3D point tracking that converts uncalibrated monocular surgical videos into structured 4D annotations through contact-region prompting, active-tool segmentation, dynamic bundle adjustment, and global 4D refinement. This pipeline adapts foundation priors to surgical videos and produces temporally consistent trajectories, pseudo-scale geometry, camera poses, and instrument masks. The dataset captures realistic challenges inherent to surgical environments, including occlusions, smoke, bleeding, and rapid instrument motion. We establish a comprehensive benchmark on SurgPose, SurgT, and STIR, covering both instrument and tissue tracking. We demonstrate instrument displacement estimation, deformable surface reconstruction, and foundation-model fine-tuning. Extensive experiments and analysis demonstrate the effectiveness of our dataset and annotation framework for 3D surgical point tracking, reconstruction, and surgical action understanding. Homepage: https://wrld.github.io/surgtap-3d/

0.34

Attribute-Evidence Learning with Multimodal Prompt for Generalized Category Discovery

cs.CV cs.LG ·2026-11-08 ·Wei Zhang, Zhu Teng, Zhixiang Chi, Junnan Zou, Wenxin Luo +3 more
top venue
Abstract · IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)

Generalized category discovery (GCD) aims to discover novel categories from unlabeled data by leveraging limited labeled samples from old classes. Existing methods that rely solely on visual cues often neglect the multimodal nature of human perception, prompting recent approaches to incorporate textual information such as category tags and attributes. However, these approaches typically entangle global tags and local attributes within a shared global image representation, rendering attributes unreliable and potentially providing misleading cues that can drive the model toward incorrect predictions. To this end, we propose AptM-GCD, a novel multimodal framework that exploits apt textual guidance through attribute evidence learning and multimodal prompt reasoning. Specifically, we design an Attribute-Evidence Learning (AEL) mechanism that treats attributes as localized visual evidence rather than global category descriptors. By interacting attribute embeddings with patch-level visual tokens, AEL identifies reliable attribute-related regions and extracts discriminative local evidence, thereby reducing the impact of noisy or irrelevant attribute cues. Building upon the extracted evidence, we further develop a Multimodal Prompt Learning (MPL) mechanism to bridge local attribute semantics and global visual understanding. MPL converts fine-grained textual evidence into visual prompts and injects them into a multimodal prompt transformer, enabling attribute information to enhance category discrimination through contextual reasoning without enforcing direct global alignment between modalities. Extensive experiments on six datasets demonstrate that AptM-GCD consistently achieves SoTA performance across both fine-grained and generic benchmarks. Moreover, the proposed framework exhibits strong generalization ability, delivering consistent gains under unimodal and class-imbalanced GCD settings.

0.34

Vietnamese Legal Ontology: Every legal answer, traced to the article that grounds it

cs.CL cs.AI ·2026-11-08 ·Khac-Hoai Nam Bui, Quan Tran Minh
Abstract

A legal answer is only as good as the article it rests on. Vietnam's largest open legal corpus pairs 109,543 published questions and answers with 573,851 legal documents, yet its citations are free text: "Khoan 3 Dieu 101 cua Bo luat nay" ("Clause 3, Article 101 of this Code") names neither the code nor its version. Following the five stages of NVIDIA Auto Ontology, Discover, Ground, Validate, Expose and Learn, we turn four open datasets into a knowledge graph of 11.5 M nodes and 16.0 M edges in which every node and edge carries its source record and method, with no duplicate keys, dangling edges or schema violations. Of 67,868 cited provisions, 61,487 resolve to a dated law version, 51,694 to a parsed article, clause or point, and 19,012 are confirmed word for word. The ontology then shows what flat records hide. 36% of the 78,361 answers whose citations resolve now cite a law that is no longer in force; 2,525 verified citations point to an article number that says something else in the current law; 2,520 citations were re-versioned because their words match a different version than their date suggests; and 5,575 document numbers are shared by several local authorities, one of them by 72. Every number in this paper is computed from the graph.

0.33

MySign: A High-Fidelity Motion-Capture Dataset for 3D Sign Generation in Bahasa Isyarat Malaysia

cs.CV ·2026-11-08 ·Jiayu Shen, Kalin Stefanov, Lay-Ki Soon, Vee Yee Chong, Alpha Agape Gopalai +1 more
benchmarkdatasetcode/data
Abstract

Progress in 3D sign generation is constrained by the scarcity of high-fidelity 3D sign language datasets, a limitation that is particularly severe for under-resourced languages. We introduce MySign, the first motion-capture dataset for isolated signs in Bahasa Isyarat Malaysia (BIM). Its 5,000 samples cover 1,000 lexical items (glosses) and were produced by five Deaf native signers. Prior datasets that rely on SMPL-X parameters regressed from video provide pseudo-labels, which are often affected by depth ambiguity, inaccurate finger poses, and temporal jitter. MySign instead provides ground-truth SMPL-X sequences retargeted from recordings made with a motion-capture setup tailored for sign language; these sequences offer higher-fidelity motion, particularly in fine-grained hand and finger articulation. A controlled comparison on the same glosses and signers indicates that training on these sequences improves generation over training on such pseudo-labels. We establish MySign as a challenging benchmark for gloss-conditioned 3D sign generation: state-of-the-art models produce visually plausible outputs but fall short of the ground truth, most notably in fine-grained articulation, revealing persistent gaps in modeling articulated sign motion. In contrast, recognition baselines consistently achieve high lexical accuracy, indicating that the dataset reliably supports supervised learning. The dataset is available at https://huggingface.co/datasets/mysigner/MySign

0.33

Kinematic Total-Body PET Video Reconstruction via Hierarchical List-Mode EM and Validation-Adaptive Space-Time Regularisation

eess.IV ·2026-11-08 ·Andrew J. Reader, Gary J. R. Cook, Paul K. Marsden, Sarah-May Gould, Alexander Hammers +1 more
Abstract · 10 pages, 11 figures

Kinematic PET is introduced as an imaging task focused on visualisation of early radiotracer delivery and rapid multi-organ motion without gating. The proposed kinematic video reconstruction method uses 33-ms frames (~30 frames per second), initialised with 3D expectation maximisation (EM) time of flight (TOF) list-mode reconstruction, followed by cycles of a 2-stage update. These 2 stages are i) a hierarchical EM (HEM) update using multiple temporal pooling levels, and ii) a validation-adaptive space-time (VAST) regularisation update. HEM combines frame-specific EM corrections with corrections estimated from higher-count temporal blocks, slowing noise amplification for the ultrashort frames. VAST mixes each 3D target patch in a frame with source patches sampled nearby in space and time. It selects the candidate that best matches the mean of the corresponding patches in the two adjacent frames, with the unchanged patch also eligible. Optional temporal block sum correction uses the initial EM temporal sum as a reference to reduce loss of spatial detail. HEM+VAST was evaluated with a dynamic digital phantom, and demonstrated on human FDG data acquired on a Siemens Biograph Vision Quadra (28 - 58 s post-injection). At a background spatial coefficient of variation of 10%, the temporal full width at half maximum of the fastest phantom bolus was 151 ms with HEM+VAST, compared with 86 ms for the ground truth. The video frame rate FDG images reveal vascular and organ detail, radiotracer bolus passage and respiration-related multi-organ motion. Kinematic PET could complement conventional dynamic PET by supporting the development of image-derived physiological biomarkers and motion estimation for quantitative reconstruction. Future work should compare HEM+VAST with prior 4D PET reconstruction methods at the same 33-ms frame duration and matched noise levels.

0.32

What You Show Is Not What You Get: Rethinking Demonstration Following in Visual In-Context Segmentation

cs.CV ·2026-11-08 ·Xiao Hu, Ilia Sucholutsky
Abstract

Visual in-context segmentation uses visual demonstrations to guide predictions for a new image. In SegGPT, we identify background color mapping deviation, where non-target regions sharing a background label receive different output colors, and part-to-parent expansion, where predictions extend beyond the requested part into its parent object. These deviations can occur even when the relevant regions remain distinguishable in internal features. Controlled changes to demonstration colors and requested targets support an account in which demonstrations guide role and color assignments over familiar semantic regions, influenced by learned semantic and color preferences. Building on these findings, we introduce Semantic Contrast Prompt Selection (SCPS). Its semantic contrast encoding assigns separate colors to the requested part and the rest of its object while keeping the remaining scene black. SCPS combines existing retrieval scores with quality estimates from labeled development images to select demonstrations and their binary or semantic contrast encodings. Averaged across seven retrieval methods on three benchmarks, SCPS improves target IoU by 0.054 and reduces leakage into the rest of the object by 22.7% relative to the original binary selectors. Our findings offer a new perspective on understanding and improving demonstration following in visual in-context segmentation.

0.32

Virtual-Flow: Virtual Microphone-based Speech Enhancement via Semi-supervised Flow Matching

eess.AS ·2026-11-08 ·Dongheon Lee, Ashutosh Pandey, Sanjeel Parekh, Zhaoheng Ni, Daniel Wong +3 more
top venue
Abstract · Accepted to NeurIPS 2026

Neural network-based virtual microphone estimation (neural-VME) synthesizes microphone signals at unobserved spatial locations from a limited set of real microphone (RM) physical recordings. This enables high-resolution arrays that improve spatial processing for downstream tasks such as speech enhancement. However, existing methods rely on supervised training with ground-truth signals captured at virtual locations, requiring datasets that scale with the number of virtual microphones (VM) and leaving open the question of how to design optimal large-array configurations. We propose Virtual-Flow, a flow-matching framework that implicitly generates VM signals conditioned on real recordings, without requiring ground-truth VM signals. We introduce a semi-supervised training strategy based on pseudo-VM targets constructed from time-delayed superpositions of real recordings. Virtual-Flow shows improved magnitude estimation in high-frequency bands, yielding richer spatial cues for beamforming. Experiments demonstrate that the proposed Virtual-Flow, combined with reflow distillation, outperforms existing neural-VME baselines in VM-based beamforming and achieves competitive end-to-end speech enhancement performance with low computational cost.

0.32

AOP-Smart-Finder: A High-Throughput Human-AI Hybrid Framework for AOP Discovery and Evidence Mapping from Large-Scale Literature

cs.CL q-bio.QM ·2026-11-08 ·Qinjiang Niu, Lu Yan
Abstract

The rapid growth of scientific literature and the high heterogeneity of biological terminology and mechanistic descriptions make it difficult for researchers to integrate scattered research evidence, identify connections among biological events, and construct mechanistic networks. Here, we developed a high-throughput human-AI hybrid framework, AOP-Smart-Finder, for biological event discovery, relationship mining, and Adverse Outcome Pathway (AOP) construction and evidence mapping from large-scale literature. This framework integrates Large Language Models (LLMs), AOP-Smart retrieval-augmented generation (RAG) technology, and Python-based automation, supporting two analysis modes: one based on a single compound and the other based on a predefined set of biological events. In the di(2-ethylhexyl) phthalate (DEHP) case, the system processed 12,983 articles within 30 min and included 1,093 of them, from which it extracted 1,589 biological events (including 458 official key events [KEs] and 1,131 candidate KEs) and 2,519 key event relationships (KERs). In the amyotrophic lateral sclerosis (ALS) case, the system identified 55 stressors from 437 articles and extracted 278 biological events and 548 event relationships. Following manual curation and Weight of Evidence (WoE) evaluation, a mechanistic network comprising 16 KEs and 23 KERs was ultimately constructed. This framework can either use existing AOPs as knowledge guidance for mechanistic expansion and evidence supplementation, or discover novel event connections and construct mechanistic networks around user-specified research topics. It provides a high-throughput, traceable, human-AI hybrid methodology for complex biological mechanism elucidation and AOP discovery driven by large-scale literature.

0.32

G-STAR: Group-Scaled Token Capacity for Deeply Compressed Autoregressive Generation

cs.CV ·2026-11-08 ·Xu Huang, Ye Huang, Xiaotong Li, Zhengrong Yue, Jialv Zou +6 more
community pick
Abstract

Discrete autoregressive models offer a scalable framework for image generation. However, they face a fundamental trade-off: long token sequences preserve representation quality but slow generation, while aggressive spatial compression reduces representation capacity. Based on product quantization, we show that, under a fixed nominal information budget, group size should scale quadratically with the per-axis spatial compression factor. Building on this principle, we propose GSTAR, a spatial--group co-scaling framework that enables \(32\times\) and even effective \(64\times\) compression while preserving strong reconstruction, understanding, and generation performance. Since larger groups introduce intra-group dependencies that must be modeled efficiently, we analyze their impact on inference latency and generation quality and introduce a semi-autoregressive decoder that combines parallel feature computation with lightweight causal modeling. With these designs, our highly compressed models perform competitively with leading models across five representative text-to-image benchmarks (GenEval, DPG-Bench, PRISM, LongText-Bench, and Qwen-Image-Bench). At \(1024\times1024\) resolution, our model delivers a \(20\times\) speedup over prior discrete autoregressive models, generating each image in 2.5 seconds on a single NVIDIA H200. Together, these results show that highly compressed discrete representations can enable high-quality autoregressive generation at low latency.

0.31

Dhara: Evidence-Oriented Retrieval for Indian Knowledge Systems

cs.IR cs.CL ·2026-11-08 ·Avinash Holenarasipur Nagendra, Anand Kaniyaar Sriranga Prasad, Rahul Vishwakarma, Trupti Gavit, Anish Amol Tipnis +4 more
Abstract · 19 pages. Supplementary material (24 pages) is provided as a separate ancillary PDF

Relevant evidence in Indian Knowledge Systems (IKS) collections may be distributed across passages in different parts of a work, including verses, translations, and commentaries. Locating it requires matching natural-language questions and partially remembered verse fragments to source passages. We present Dhara, a retrieval engine combining hybrid search, chapter- and book-level Entity Notes (EN), Fuzzy Verse Retrieval (FVR), and neural reranking. Five comparisons on the 325 questions in Dhara Bench examine retrieval representations against BM25, EN's contribution, reranker quality and computation time, FVR, and Gemini. BGE-M3 single-vector retrieval exceeds BM25 in nDCG@10, Recall@20, and MRR@20 across all categories. Its scores are broadly comparable to ColBERTv2, with mean retrieval times of 0.108 s and 0.962 s under the same hardware configuration and a smaller memory footprint for chunk storage on disk. Adding EN to a direct branch with a matched candidate count improves nDCG@10 in all twelve categories with Qwen and ten with BGE; the combined ranking is capped at the same candidate count. Qwen improves nDCG@10, Precision@5, and MRR@20 over BGE in every category, at approximately 6.72 times the overall estimated serial computation time. FVR increases Verse-category nDCG@10 by approximately 20 percentage points with either reranker. Gemini is evaluated after source mapping and manual review with relevance annotations visible. The Qwen pipeline with EN and FVR has higher Recall@20 than Gemini in ten categories and stronger complete-evidence retrieval in Multi-hop questions, while Gemini retains early-relevance advantages in several categories. The results inform retrieval choices according to evidence requirements and resource budgets.

0.31

CrossLiDAR: Controllable Multi-Sensor LiDAR Sequence Generation

cs.CV cs.AIcs.RO ·2026-11-08 ·Khanh Nguyen, Thomas Walker, Hichem Telli, Carlos de la Torre-Ortiz, Stefano Fiorini +4 more
published
Abstract · WACV 2027

High-fidelity LiDAR simulation remains a bottleneck for scalable development and validation of autonomous systems. Existing generators can synthesize realistic scans, but they are typically constrained to a single sensor configuration, offer limited control over structure and geometry, and drift in long-horizon 4D generation. We present CrossLiDAR, a unified transformer-based framework capable of generating controllable and spatio-temporally consistent multi-sensor LiDAR sequences. Our approach leverages a rectified flow objective with a Multi-Modal Diffusion Transformer (MM-DiT) backbone, within a compact latent space of range images. To support multiple LiDAR sensors, an Inter-sensor Module enables interaction across sensors, while a ControlNet-style branch provides structural conditioning (e.g., BEV priors or 3D bounding boxes). For modeling the temporal consistency, CrossLiDAR generates short $N$-frame sequences and extends them autoregressively to improve long-horizon coherence. Our model significantly outperforms state-of-the-art baselines, achieving an average improvement of 29.56\% in JSD and 15.36\% in MMD evaluated on nuScenes and the Waymo Dataset. CrossLiDAR bridges the gap to modern autonomous-driving simulation by enabling calibration-consistent multi-sensor LiDAR rollouts over long horizons for controllable, scenario-based testing.

0.31

Modeling individual and item differences in atypicality inferences with ACT-R

cs.CL ·2026-11-08 ·Guifu Liu, Vera Demberg
published
Abstract · Accepted as a conference paper at CogSci 2026

Humans engage in pragmatic reasoning when encountering mentions of highly predictable events. Such mentions may appear redundant, leading comprehenders to infer that the event is atypical, and accommodate this inference with an explanation for why it is atypical. However, these inferences are not drawn uniformly across individuals, and experimental items vary in inference rates due to differences in event typicality. We show that these patterns can be predicted with a resource-rational cognitive model in which comprehenders select interpretation strategies based on individual exploration tendencies and item-level predispositions. These parameters are estimated independently from participants' Raven's Matrices score, event typicality rating and language model based accommodation likelihood. Taken together with recent findings from reference games, our results motivate a new perspective on the mechanisms that support pragmatic inferences.

0.31

Capacity Allocation at the Source: Sparse Target Optimization for LLM Knowledge Editing

cs.CL ·2026-11-08 ·Dongyao Chen, Jiayun Lei, Zhiyuan Ji, Zhiying Deng, Wei Liu +1 more
benchmarkcode/data
Abstract · neurips2026

Knowledge editing has emerged as a promising approach for efficiently updating embedded knowledge in large language models (LLMs). It first computes an ideal target hidden state that steers the LLM toward the new output, and then edits the model parameters so that the original hidden state is mapped to this target. While most existing methods focus on the second stage of parameter editing in preserving other knowledge, we find that the upstream construction of the target hidden state also plays an important role in allocating the model's limited capacity space. We first establish a theoretical connection between interference in the parameter space and interference between target hidden states, showing that the two are positively correlated. Building on this insight, we propose a simple yet effective $\ell_1$ regularization method that prunes away unnecessary ``tentacles'' of each knowledge vector, retaining only the essential components. With fewer extraneous tentacles, each edit affects less other knowledge, enabling better coexistence. Extensive experiments across multiple LLMs and benchmark datasets verify the effectiveness of our method. Code: https://github.com/wintersdragon-c/STO.

0.30

Partition, Refine, and Detect: Preserving Novel Category Capacity in Open-Vocabulary Object Detection

cs.CV cs.LG ·2026-11-08 ·Wei Zhang, Zhu Teng, Zhixiang Chi, Junnan Zou, Yi Liu +1 more
top venue
Abstract · IEEE Transactions on Multimedia (TMM)

Open-vocabulary object detection (OVD) expects the model trained on base classes to recognize both base and novel classes at inference. Typical methods rely on CLIP-style vision-language embeddings, yet the learned space is often biased toward base categories, with limited effective capacity preserved for novel concepts. Such a base-dominated representation can compress novel semantics, leading to semantic collapse and mis-calibrated confidence at inference. To address this issue, we present PR-OVD, a partition-and-refine framework that explicitly reserves capacity for novel categories and strengthens category separation, while being seamlessly integrated into different detector architectures. Specifically, PR-OVD first performs coarse partitioning to build global divisions for base, novel, and an explicit background partition in both visual and textual modalities, preventing object-irrelevant embeddings from leaking into category-specific representations. It then applies category-level refinement within each partition to sharpen decision boundaries and alleviate the lack of fine-grained region-level pretraining in CLIP. To further provide reliable visual anchors for novel categories, we design a cognitively inspired Structure Prompt (SP) that encodes both egocentric attributes and allocentric context, facilitating stable partition-and-refine learning. Extensive experiments on two OVD benchmarks show that PR-OVD delivers an average 6.5\% proportional improvement on novel categories over the state-of-the-art methods. Notably, PR-OVD also benefits other open-vocabulary tasks, including segmentation and classification, further confirming its effectiveness and generality.

0.30

What Does Multiple Choice Actually Measure? Disentangling Video Understanding from Option Exploitation

cs.CV ·2026-11-08 ·Elliott Piercy, Justin Chan, Sepand Dyanatkar
published
Abstract · Accepted to TAE (Trust-AI-Eval) Workshop at NeurIPS 2026 https://neurips.cc/virtual/2026/loc/sydney/178197

Video understanding evaluations are intended to measure a model's proficiency in video perception and task solving. However, current VideoQA benchmarks struggle to isolate these capabilities due to their reliance on multiple-choice (MC) answer guidance, language priors, and fragile scoring pipelines. Disentangling these confounding factors reveals how much of a model's score is supplied by the evaluation context rather than earned through faithful perception. Across six VideoQA benchmarks of varying length, we find that accuracy is significantly above chance without any visual input. This is caused by shortcutting and option-prior reliance. We express these effects in the language of faithfulness: option-exploitation and answer-conditioned reasoning traces are unfaithful even when they are correct. Together, our results show that a VideoQA score is jointly determined by language priors, option timing, and how outputs and labels are converted into correctness. We produce a novel Option-Delayed Replay (ODR) mechanism that discloses options at controlled stages of an open-ended (OE) trajectory, to obtain objective MC scores from faithful option-free inference. Our work demonstrates that agentic methods with ODR and faithful inference can match option-exploiting performance. Finally, we formalize our diagnostic collection and ODR as a reproducible evaluation protocol, and we release a human-relabelled edition of LVBench (correcting $20.5\%$ of keys) to remove annotation noise as a confounding variable.

0.30

YTShorts-92K: An Interpretable Multimodal Dataset of YouTube Shorts

cs.MM ·2026-11-08 ·Celian Charrin, Sidonie Abuzakuk, Artúr Vlaszov, Nándor Birher, Roland Molontay
dataset
Abstract · 10 pages, 3 figures, 3 tables

Short-form video has become one of the dominant formats of online content, yet large-scale datasets enabling the analysis of what makes these videos successful remain limited. Existing resources often focus on post-publication popularity signals or rely on non-interpretable representations such as high-dimensional embeddings. Here, we introduce YTShorts-92K, a dataset of 92,654 YouTube Shorts collected from 345 creators and spanning 12 manually defined content categories. Each video is enriched with an interpretable multimodal feature set covering metadata, textual information, visual characteristics from frames and thumbnails, and cross-modal consistency measures. We describe the collection methodology, data cleaning process, and feature extraction pipeline, followed by descriptive analyses of the diversity of creators, topics, and extracted features. This dataset offers a reusable resource for studying short-form video content, online attention, and multimodal content analysis.

0.30

ADC-RANet: Improving Resolution Adaptive Neural Networks with Attention, Distillation and Calibration

cs.CV ·2026-11-08 ·Jeferson Severino de Araujo, Lilian Hollard, Eric Granger, Marco Pedersoli
code/data
Abstract

Dynamic neural networks are a class of adaptive models in which the computation assigned to each input sample is determined dynamically, potentially reducing computational cost while maintaining similar performance. For image classification, the Resolution Adaptive Network (RANet) is a strong approach that progressively allocates computational resources across input scales based on sample difficulty. In this work, we introduce RANet with Attention, Distillation and Calibration (ADC-RANet), an enhanced version of RANet that improves performance across all computational budgets. Our approach combines three tightly connected components: bottom-up channel attention, top-down cascaded knowledge distillation, and overconfidence correction using MixUp, in a synergistic design that coordinates both pathways. Experiments on CIFAR-100 and ImageNet datasets show consistent gains while preserving the efficiency of RANet. Our code is available: https://github.com/jefersonseverino/ADC-RANet

0.30

Evidence-Grounded Iterative Reasoning for the Diagnosis of Diffuse Cystic Lung Diseases

cs.CV ·2026-11-08 ·Haoqing Li, Jun Shi, Tou Li, Xianmeng Chen, Rui Wang +7 more
Abstract

Overlapping imaging manifestations complicate the differential diagnosis of Diffuse Cystic Lung Diseases (DCLDs), while limited annotated data constrain supervised model development. Multimodal Large Language Models (MLLMs) can integrate imaging and clinical information, but their diagnostic reasoning remains limited by incomplete volumetric perception and insufficient grounding in disease-specific evidence. We propose DCLDs-EIR, an evidence-grounded iterative reasoning framework coupling diagnostic hypothesis refinement with multimodal evidence acquisition and verification. Multiview renderings of predicted lung and cyst segmentations characterize global cyst distributions and initialize the differential diagnosis. Competing hypotheses guide axial CT inspection to identify local findings that, together with clinical context, inform retrieval from a disease-specific knowledge graph. Accumulated patient observations and retrieved knowledge are jointly assessed for diagnostic support and refutation to revise the diagnostic state. Updated hypotheses and unresolved distinctions guide subsequent inspection, forming a feedback loop between evidence acquisition and hypothesis verification. Multicenter evaluation demonstrates that DCLDs-EIR consistently improves diagnostic accuracy over baselines and enables traceable differential diagnosis through explicit links between diagnostic hypotheses, patient observations, and disease-specific knowledge.

0.29

Does Merge Form a Hopf Algebra? Coproducts, Traces, and the Prospects of Algebraic Minimalism

cs.CL math.QA ·2026-11-08 ·Daisuke Bekki
theoryunusual
Abstract · 15 pages

Marcolli, Chomsky, and Berwick (2025; henceforth MCB) formulate the Merge operation of Minimalism on a Hopf algebra of Connes-Kreimer type, and claim that language and physics obey the same algebraic structure. This paper separates their program into the working hypothesis H, that Merge on workspaces can be described as a Hopf algebra, and the claim P placed on top of it, that language and physics obey the same principle, and examines the validity of H. MCB employs two coproducts in formulating Merge, corresponding to two different translations of movement (Internal Merge): $Δ_d$, which deletes the lower copy, and $Δ_c$, which leaves a trace. MCB uses the former in syntax and the latter at the syntax-semantics interface, but $Δ_d$ is not coassociative, and $Δ_c$ is not connected under the gradings MCB gives and, moreover, not coassociative as long as the labels of traces are retained. Consequently, MCB's algebra is not a Hopf algebra, and the Birkhoff factorization is not a consequence of the Hopf-algebra theorem but of a weaker statement that requires neither coassociativity nor connectedness. There are four ways to recover a Hopf algebra, but none of them can distinguish different traces, and hence none is compatible with Heim-Kratzer-style formal semantics, which reads traces as indexed variables. Thus H does not hold at present, and under the current formulation, making Minimalism a Hopf algebra and preserving its connection with formal semantics are mutually exclusive. Finally, we indicate three directions in which algebraic Minimalism may proceed.

0.29

Selective Surgical Action-Center Localization:Retaining Localization Gains through Supervised Scoring

cs.CV ·2026-11-08 ·Yanchong Xue, Wanting Zhang, Lingtao Yu, Guochao Zhang
hot direction
Abstract

Surgical action-center localization requires both an accurate reference point and a decision about whether to report it. When frames with and without a reportable center are mixed, the benefit of improved localization depends on which correct candidates survive selection. We present a selective localization pipeline combining heatmap and coordinate supervision with a logistic-regression scorer. The scorer learns to identify predictions with a reportable center and a localization error within a specified tolerance. The internal evaluation uses 5,000 frames from 50 surgical sources, five source-grouped folds and two random seeds. Matched locator and scoring comparisons track correct candidates and their rejection. At an 80% acceptance quota per source and a tolerance of 0.10 times the image diagonal, our method retains 46.08 correct outputs per 100 inputs. The baseline locator with the same scoring approach retains 44.20. The gain is 1.88 (descriptive 95% source-paired bootstrap interval, 0.94 to 2.79). Under entropy-based selection, increased rejection of correct predictions offsets the higher supply of correct predictions on average. For the coordinate-supervised locator, supervised scoring reduces these rejections from 6.15 to 2.99 per 100 inputs. Evaluating retained correct outputs reveals whether a selection rule preserves or erodes an improvement in localization.

0.29

When Normality Shifts: Multi-Level WavLM Modeling for Open-World Audio Deepfake Detection

cs.SD cs.LG ·2026-11-08 ·Daniel Niekurzak, Daniel Kozuch, Weronika Folwarska, Khan Muhammad, Rafał Scherer
published
Abstract · Accpeted for the OWAD Workshop at ICDM 2026

Audio deepfake detectors trained with known synthesis attacks can learn generator-specific shortcuts that fail as synthesis systems and acoustic conditions evolve. We investigate an attack-free alternative: a normality model trained exclusively on bona fide speech, which scores synthetic audio by its deviation from authentic representations. We propose MLN-WavLM, a method that applies shrinkage-Mahalanobis modeling to pooled hidden states from a frozen large-scale, pre-trained speech processing WavLM Base+ architecture. Evaluated on ASVspoof 2019 LA, our best attack-free variant achieves an 8.45\% Equal Error Rate (EER), closely tracking a fully supervised WavLM-HM-Conformer baseline (7.86%). However, when subjected to the ASVspoof 2021 DF domain shift, performance drops to 22.12% EER, exposing vulnerabilities in the bona-fide-calibrated operating point. Layer-wise analysis reveals that early WavLM representations dominate the ensemble. To test real-world viability, we simulated deployment streams using unlabeled threshold adaptation. While robust median/MAD recalibration easily handles in-domain contamination up to 10%, recalibrating under the 2021 DF domain shift triggers a total collapse of low-FPR recall. We conclude that while frozen source-domain normality can effectively rank deepfakes, it does not yield a stable open-world decision rule, and downstream threshold adaptation cannot rescue upstream ranking degradation.

0.29

SegDis: Foreground Segmentation Followed by Type Discrimination for Salient and Camouflaged Object Detection

cs.CV ·2026-11-08 ·Dingning Luo, Shuo Dou, Tong Jiang, Zhenhua Duan, Jiachen Zhu +1 more
Abstract · 10pages, unincludes supplementary material

Salient and camouflaged object detection (SOD and COD) both recover foreground support, yet most joint methods output only a binary mask and therefore cannot distinguish the type. USC12K distinguishes the two types, but its baseline USCNet still binds foreground existence to the type within a single normalization, so the failure source cannot be attributed. Our cross-task study shows that foreground localization transfers between SOD and COD, whereas type discrimination does not: a COD-only model recovers foreground on SOD images but its type judgement follows the training distribution rather than the foreground-background relation, so the two decisions require separate mechanisms. SegDis factorizes the output into an explicit foreground probability and a conditional type probability given foreground: FLR provides high-quality task-agnostic support, and RTA refines the SOD/COD assignment from the foreground-background relation without a task identifier. Because support is explicit, each ground-truth foreground pixel is classified as correct-type, type-swapped, or missed, so foreground-recovery and type-assignment errors are diagnosed separately. With PVT-v2-B2, SegDis attains 68.98% FG-mIoU and 78.35% mIoU on USC12K at about one-eighth of USCNet's parameters; CSCS is 5.75.

0.29

Music Inpainting With Similarity-Guided Latent Diffusion

cs.SD cs.LGeess.AS ·2026-11-08 ·Sean Turland, Eloi Moliner, Vesa Välimäki
published
Abstract · Accepted to ISMIR 2026. 8 pages, 3 figures, 1 table. Code and audio examples: https://s-turland.github.io/SimDPSInpainting/

We propose a music inpainting technique combining similarity matching and a diffusion-based generative model to replace large segments of audio recordings with suitable alternatives. The proposed hybrid method, Similarity-Guided Diffusion Posterior Sampling (SimDPS), retrieves candidate replacement segments from the wider signal based on contextual similarity and uses them to guide the diffusion process toward coherent reconstructions. Being model-agnostic, the framework is compatible with any pre-trained diffusion model operating in either the waveform or latent domain. Objective evaluations demonstrate that SimDPS outperforms widely adopted RePaint and DPS methods across acoustic features and neural-embedding distance metrics. Subjective evaluations on multi-genre music datasets confirm that SimDPS enhances perceptual plausibility compared to previous diffusion-based methods and demonstrates robustness to poorly matched similarity candidates. These findings pave the way for improved music restoration and context-aware audio editing.

0.29

Guidance at Pure Noise Erodes Speaker Similarity in Few-Step Token-Conditioned Flow-Matching TTS

cs.SD cs.AI ·2026-11-08 ·Jaeho Lee, Han-Lim Choi
code/data
Abstract · 13 pages, 8 figures. Submitted to IEEE Transactions on Audio, Speech and Language Processing. Code: https://github.com/16jhlee/FSU

Text-to-speech systems built from a language model and a flow-matching mel decoder now run the decoder with two to four Euler steps instead of the official ten or more. At two steps the word error rate (WER) is unchanged, but speaker similarity (SIM) drops by one to six points. This paper shows that most of this loss comes from classifier-free guidance on the first step. At pure noise the network can only return the conditional mean, so guidance shifts every trajectory by nearly the same vector. Later steps do not remove this shift, and its size grows with the length of the first step, so it is negligible with ten or more steps and accounts for most of the loss with two. Two training-free samplers are derived from this analysis. First-step unguided sampling (FSU) leaves the first step unguided and recovers $84$--$91\%$ of the two-step SIM loss on CosyVoice~2/3 with no significant WER increase. FSU-Mid replaces the two Euler steps by one midpoint step with an unguided first evaluation, recovers $77$--$85\%$ on IndexTTS-2 and matches the ten-step teacher on CosyVoice~2 and GLM-TTS. Both use three network function evaluations, one fewer than the official two-step sampler, and FSU-Mid gives the highest mean recovery among eight published guidance rules re-run at two steps on four decoders.

0.29

TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption

cs.CV cs.AIcs.MM ·2026-11-08 ·Qiming Huang, Peixi Liu, Jianbo Jiao
published
Abstract · Accpeted to NeurIPS2026

Multimodal (e.g. vision-language, in this work) attribution aims to interpret the models without access to ground-truth explanatory supervision. Existing multimodal attribution methods typically rely on paired modalities as proxy supervision, implicitly assuming that image-text pairs are semantically aligned. Information bottleneck (IB)-based attribution methods follow this general paradigm and instantiate it through a global sufficiency-compression objective. In realistic multimodal datasets, however, this assumption is frequently violated due to partial semantic misalignment and unreliable cross-modal correspondences. Through theoretical analysis, we show that, under such misalignment, information-bottleneck-based attribution objectives exhibit a structural limitation: by assuming equal reliability across all proxy supervision signals, they induce posterior over-contraction and forced explanations. To address this issue, we propose TIB, a sample-wise Tempered Information Bottleneck framework that adaptively modulates attribution strength according to cross-modal reliability, without assuming semantic alignment. Extensive experiments on controlled benchmarks and large-scale datasets show the effectiveness of the proposed approach, while also validating the identified failure mode.

0.28

SurfScaff3D: Bridging Geometric and Semantic Representations for LLM-based 3D Object Generation

cs.CV ·2026-11-08 ·Anay Loya, Rishi Gupta, Arnav Samal, Anirban Chakraborty
publishedposition
Abstract · Accepted to WACV 2027

Creating high-quality 3D content has traditionally required specialist expertise and experience with complex tools. The growing ecosystem of AI-assisted creativity demands accessible and intuitive 3D generation systems. Recent advances in Large Language Models (LLMs) offer promising directions, yet extending their generative capabilities to 3D remains fundamentally challenging because existing 3D representations are not naturally compatible with language-based reasoning. Current approaches either depend on expensive fine-tuning or impose restrictive structural priors that sacrifice expressiveness and exclude organic shapes. We argue that the mismatch between LLM-native reasoning and conventional 3D parameterizations is a core bottleneck of LLM-driven 3D generation. To address this, we introduce the Hierarchical Patch Scaffold (HPS), a representation designed around semantically meaningful, interpretable part-level abstractions that are naturally understood by both humans and LLMs, enabling expressive, editable, and open-domain 3D generation without imposing restrictive geometric priors or expensive fine-tuning. Building upon HPS, we introduce SurfScaff3D, an interactive framework for text and image-to-3D generation. HPS inherits the flexibility of B-spline patches, while exposing an interface that leverages open-world capabilities of foundation models to reliably generate 3D. Its hierarchical part-level semantic labels enable targeted, natural-language edits at any level of granularity without re-running generation. Further, HPS offers explicit per-patch control over surface detail, enabling more expressive geometry capable of representing arbitrary objects. Extensive experiments against state-of-the-art LLM-based 3D generation methods demonstrate strong geometric fidelity, semantic alignment, and flexible editing across diverse object categories.

0.28

NightSteps: A Night-Time Dataset and Benchmark for Pedestrian Crossing Intention Prediction

cs.CV ·2026-11-08 ·Sjoerd Takken, Shaodi You
benchmark
Abstract

Autonomous vehicles use pedestrian crossing intention prediction (PCIP) to participate safely in traffic. Existing models are largely trained under favourable daytime conditions with sufficient daylight. Little is known about the robustness of these models under night-time and low-light conditions, due to the lack of datasets containing night-time footage. This paper investigates how a dataset, named NightSteps, can be constructed and evaluated as a baseline benchmark. The dataset is based on nighttime dashcam footage collected from six different cities around the world, capturing a wide variety of road conditions. The dataset consists of dashcam-recorded point-of-view footage, raw data of pedestrians' locations with corresponding 'track ids' and ground-truth labels containing accurate information about pedestrian crossing behaviour, resulting in a diverse dataset. Gemini, ChatGPT, and human assessment are used to demonstrate the usability of the dataset by predicting crossings in clips from the dataset and comparing the results with the initial ground-truth labels. These results show that human input achieves an accuracy of 88.17% and an F1-score of 80.21%. Predictions from Gemini and ChatGPT perform lower, with accuracies of 60% and 79.13% respectively, and F1-scores of 42.50% and 63.64%. The evaluation demonstrates that the NightSteps dataset can be used for future model evaluation to test the performance of PCIP, with the aim of improving road safety at night.

0.28

Not All Branches Are Equal: Adaptive Semantic ID Construction for Generative Recommendation

cs.IR cs.LG ·2026-11-08 ·Guy Hadad, Haggai Roitman, Bracha Shapira
top venueposition
Abstract · RecSys '26

Generative recommendation models represent items as sequences of discrete tokens known as semantic IDs. Existing approaches derive these identifiers using residual or hierarchical vector quantization, assuming that tokens carry independent semantic meaning across quantization levels. Revisiting this assumption, we argue that semantic IDs should behave as hierarchical paths where each token's meaning is conditioned on its prefix. This highlights a mismatch between conventional fixed-codebook quantization and the hierarchical structure implicitly learned by generative recommenders. Motivated by this insight, we propose Adaptive Silhouette Tree (AST), a top-down divisive clustering method that constructs semantic IDs as paths within a tree structure. Unlike prior methods with fixed branching factors, AST dynamically determines the number of branches at each node by maximizing the silhouette coefficient, enabling finer partitions in dense regions and compact groupings in homogeneous ones. We further introduce constrained variants that enforce minimum probability mass during splitting to improve supervision for long-tail items. Extensive experiments on five Amazon Review datasets of varying sizes and domains show that AST significantly outperforms strong baselines. These results demonstrate that aligning semantic ID construction with the inherent hierarchical structure of generative recommenders leads to more effective recommendations.

0.28

Real-Time Arbitrary Resampling for Phase-Continuous Microtuning in Audio Streams

cs.SD ·2026-11-08 ·Andreas Falkenberg
Abstract · 6 pages, 1 figure

This paper develops an output-driven resampling formulation for fine pitch adjustment of recorded signals. A fractional input-position increment combines the desired frequency ratio with independently specified input and output sample rates. Each output sample is evaluated from neighboring input samples, and gradual rate changes preserve the current input position. Musical interval mapping and direct sine synthesis are treated as distinct layers. The formulation separates tuning resolution, numerical accuracy, interpolation quality, and streaming correctness. Optional rational-factor representations use freely chosen approximation parameters; numerical illustrations do not define constants required by the algorithm. Linear interpolation provides a low-cost evaluation kernel but does not establish a universal signal-quality or frequency-accuracy guarantee.

0.27

Beyond Ranking: Are Decision Models Calibrated Relevance Scorers?

cs.IR ·2026-11-08 ·Shengyao Zhuang, Xueguang Ma, Matthias Petri, Jimmy Lin
Abstract

Decision models such as Jev are a new kind of relevance scorer. Their typed score primitive takes a question, an ordered set of textual relevance levels and a state, and returns a probability for each level directly, rather than token likelihoods or generated labels. Jev's vendor promises that these probabilities are calibrated. This motivates revisiting 'Beyond Yes and No', which improved zero-shot pointwise LLM reranking using likelihood-derived scores over fine-grained relevance labels but left calibration against human assessors as future work. In this paper, we first reproduce its open-model FLAN-UL2 setting on eight BEIR collections, confirming its ranking trends, and then run Jev and two open-weight decision models with the same protocol. Jev and the 27B pplx-decider rerank ahead of FLAN-UL2 and three cross-encoders on average. We then compare Jev's native probabilities with the human assessors' labels and find them overconfident: of the DL21 pairs it scores above 0.9, only 66% are relevant. Since this is mostly an error of scale, a two-parameter calibration fitted on 2,257 DL19 pairs removes much of it and transfers to DL21. After calibration, Jev has a lower DL21 Brier score than FLAN-UL2 and the three cross-encoders, mainly because it separates relevant from non-relevant pairs better. Finally, with calibrated probabilities, choosing each query's cut-off by expected F1 needs no parameter of its own, beats a fixed top-10 cut and matches a cut-off threshold tuned on DL19.

0.26

LAION-Audio-XL: A Web-Scale Corpus and Scaling Studies for Contrastive Language-Audio Pretraining

cs.SD cs.CL ·2026-11-08 ·Gijs Wijngaard, Felix Förster, Ross Wightman, Marianna Nezhurina, Christoph Schuhmann +2 more
datasetcode/data
Abstract · Project page: https://projects.laion.ai/clap2/

Recent progress on contrastive audio-language models can be attributed mainly to the scale and quality of their training data. Larger training datasets expose models to a broader range of sounds and corresponding audio captions. To exploit this, we collect 1.48 billion audio samples from academic datasets, internet datasets, web collections, Common Crawl, and YouTube, and process them with a multi-stage curation pipeline. Furthermore, we generate short and long captions for samples in our corpus that lack them. We use our dataset to study the scaling behavior of contrastive audio-language models based on the NaFlex architecture for the audio encoder. Our experiments identify trade-offs between model size and training data for a given compute budget. Moreover, our resulting models, CLAP2, outperform prior audio-text models of comparable size on the MAEB and HEAR benchmarks. We release our dataset, captions, model checkpoints, and code.

0.26

Building and Auditing Chemical Datasets with LLMs: A General LLM-as-a-Judge Framework Applied to Homogeneous Catalytic PET Depolymerisation

cs.CL cs.AIcs.DL ·2026-11-08 ·Muhammad bin Javaid, Karim K. Ben Hicham, Enkhnyam Battulga, Alexander Mitsos, Martin Grohe
datasetcode/data
Abstract

Large language models can rapidly extract structured multivariate experimental data from the chemistry literature, yet verifying thousands of generated records without ground truth remains difficult. We present a general, citation-grounded extraction and verification methodology centered on an independent "LLM-as-a-judge", applied to the literature on homogeneous catalytic poly(ethylene terephthalate) (PET) depolymerisation. In this pipeline, an extractor LLM generates schema-enforced records from research papers with citations to document chunks tagged with universally unique identifiers (UUIDs); an independent LLM judge then reads the full paper alongside each record, flags unsupported fields, and proposes cited corrections. Prompt ablations show that requiring chunk-level citations not only provides provenance down to specific paragraphs and tables but also improves extraction accuracy. Before deploying the judge at scale, we benchmarked it against manual evaluation on extractions from a hand-curated reference set of 295 ionic-liquid-catalysed PET glycolysis experiments. The judge agreed with human evaluation on 85.7% of reviewed records and was significantly more effective than a heuristic string-matching baseline at catching errors across related experiments. Deployed across 3,942 extracted records from 561 papers, the framework accepted 67.5% of them as written. Its proposed corrections concentrated on quantitative fields, where the extractor often conflated related experiments. We release the extracted dataset with the complete LLM judge audit logs, as well as the Chemistry Data Extractor Toolkit, an open-source application implementing this end-to-end pipeline in a code-free, browser-based environment for document parsing, customizable schema and prompt design, model-agnostic extraction and auditing, interactive provenance-highlighted review, and reproducible dataset bundling.

0.26

The Average Patient: Do Simulated Psychiatric Patients Answer Like Real People?

cs.CL ·2026-11-08 ·Weixiang Sun, Zehong Wang, Kaiwen Shi, Hong Huang, Colby Nelson +2 more
Abstract

Language models are used as simulated psychiatric patients to train clinicians and evaluate AI therapists, and are judged by plausibility or total scores. We give a framework for simulated respondents: an information budget bounds how much simulated patients can differ, and an average-patient hypothesis, that a model answers each item at the level typical of the named condition and severity, implies little variation, a full profile of the condition's typical symptoms and totals at or above the top of the stated severity band. We examine these implications at matched totals on item-level data from eight human samples, including a clinical and an instructed-feigning sample. Across 17 models, simulated patients vary 1.5 to 11 times less than real respondents with the same total and on most questionnaires endorse nearly every symptom at a middling level; on the PHQ-9 their symptoms covary less like real ones. People, honest or feigning, answer "not at all" about as often as genuine patients with the same totals, and more often than every model. Single answers are highly concentrated, and prompt changes, published personas, larger models and reasoning reduce the pattern without removing it. In interviews, patients told a severity are rated as having nearly every symptom, so an assessment agent's recognition of absent symptoms goes untested, and scored against the stated severity the agent places a quarter of them in their band, against about half of patients given a real symptom profile. Having the model word such a profile restores half to two thirds of the variation.

0.26

MoT3DVG: A Benchmark for Outdoor 3D Visual Grounding with Motion-Aware Descriptions and Temporal Cues

cs.CV cs.AI ·2026-11-08 ·Shijia Zhao, Xusheng Guo, Qiming Xia, Xun Huang, Kezheng Xiong +2 more
publishedbenchmark
Abstract · NeurIPS 2026 Evaluations and Datasets

3D visual grounding (3DVG) localizes language-referred objects in 3D scenes, with broad applicability in both indoor and outdoor environments. However, existing 3DVG research focuses on static indoor objects, limiting its use in dynamic outdoor autonomous driving. Motion descriptions, which capture the temporal dynamics of moving objects, offer a promising way to address this challenge. We therefore introduce MoT3DVG, a large-scale dataset for dynamic-aware outdoor 3DVG with temporally evolving motion descriptions. It contains 850 scenarios with 31,128 frames from nuScenes dataset, and provides 144,568 language prompts with motion-aware descriptions for dynamic objects across time. We further propose DynaVG, a novel framework that effectively leverages temporal cues for outdoor 3DVG. In addition to the standard modality-specific encoders, DynaVG introduces a Short-Long Integrated Dynamic Encoding (SLIDE) module before the language-point cloud alignment. SLIDE models both local motion cues and global temporal context via short-term window encoding and long-term window shifting. Extensive experiments demonstrate its state-of-the-art performance on MoT3DVG, validating the effectiveness of motion-aware encoding. Our work reveals open challenges and promising directions for future research in outdoor 3DVG.

0.25

Capability-Guided Iterative Online Optimization of Cascaded Spoken-Dialogue Modules

eess.AS ·2026-11-08 ·Yuhang Qiu, Bohan Li, Wenbin Huang, Zhihan Li, Yiwei Guo +3 more
published
Abstract · 15 pages, 4 figures. Pre-review version of a paper accepted at NCMMSC 2026

Cascaded spoken dialogue systems comprise perception-and-comprehension, decision-making, and response-generation modules. We study iterative optimization of the first and last modules. Their reciprocal speech-content interfaces enable one-sided data to provide cross-module supervision: acoustic observations, such as waveforms or feature sequences, can be pseudo-labeled to train generation, whereas content-only samples can be rendered into speech to train comprehension. Because the reliable teacher-student direction varies across domains, we propose a two-stage, capability-guided pipeline for iterative online optimization. Stage 1 compares both modules through a shared intelligibility probe and obtains a domain-wise capability prior. Stage 2 activates the compatible one-sided-data route, constructs and filters supervision with the current teacher, updates the selected student, and refreshes the relevant checkpoints across cycles. We instantiate comprehension with Whisper large-v3 and generation with CosyVoice2. After only 200 steps for Whisper and 600 for CosyVoice2 of our iterative online optimization, we observe that, on the Chinese MAGICDATA-RAMC dataset, Whisper reduces CER from 14.18% to 10.77%, while CosyVoice2 reduces it from 4.78% to 4.04%. On the Japanese Common Voice dataset, Whisper reduces CER from 15.23% to 15.10%, while CosyVoice2 reduces it from 14.85% to 11.18%. These results demonstrate the feasibility of capability-guided iterative optimization of complementary spoken-dialogue modules.

0.25

Unlearning at Speed: Rapid Activation Dampening for Efficient Machine Unlearning in Real-World Face Recognition

cs.CV ·2026-11-08 ·Rouqaiah Al-Refai, Leo Scheubel, Philipp Terhörst
Abstract

Machine unlearning has become increasingly important in face recognition, where regulations like the GDPR empower users to request the removal of their personal data from deployed models. However, most existing unlearning methods are designed for closed-set classification with small-scale models and training data, and therefore do not generalize to real-world face recognition systems, which operate in an open-set setting with large-scale models and data. In this work, we propose Rapid Activation Dampening (RAD), a fast and targeted unlearning method tailored specifically for large-scale open-set face recognition. RAD approximates differences in average neuron activations between forget and retain identities, identifying and selectively suppressing neurons that are most associated with the targeted identity through a sensitivity-controlled dampening mechanism. We further propose the Forget Score (FS), which addresses the mismatch between existing unlearning metrics and open-set problems by providing a probabilistic interpretation of forgetting directly in the representation space. Experiments across several recent large-scale face recognition architectures and datasets show that RAD achieves strong forgetting while preserving overall model utility and substantially improving unlearning speed. These findings establish RAD as a scalable and practical approach to machine unlearning in face recognition systems.

0.25

Saidzoda-Tajik-EmbeddingGemma-300M: Knowing Every Letter Is Not Reading Tajik

cs.CL ·2026-11-08 ·Tohir Saidzoda, Nekruz Maxudov, Bedil Karimov, Abubakr Abdullozoda, Amin Jalilov +3 more
code/data
Abstract · 10 pages, 4 tables, 1 figure. Technical report. Model: https://huggingface.co/Tohirju/Saidzoda-Tajik-EmbeddingGemma-300M

EmbeddingGemma-300M is a small open embedder whose tokenizer represents every Tajik letter, and as released it is weak at Tajik retrieval: on our Tajik law test the right passage is in its top ten results for 49.3% of questions, on tax documents for 45.9%, and on the human-written Belebele Tajik set for 69.8%, against 98.6% on the Russian version of the same set. We present Saidzoda-Tajik-EmbeddingGemma-300M, the same model fully fine-tuned for one epoch on 235,140 question and passage pairs written by our own Tajik language model for laws, tax documents, Wikipedia and web text, with the same mined hard negatives, English and Russian replay, loss, global batch, learning rate and epoch as our earlier Saidzoda-Tajik-E5-Large (arXiv:2611.00592) and BGE-M3 models. After training, law reaches 92.4% and tax 91.7%; on Belebele, the one test written by people, recall@10 rises from 69.8% to 92.6% and the first result is right for 76.6% of questions instead of 51.7%. A checkpoint at step 100 of 466 already holds 88% of the law gain at recall@10. At recall@10 the result is no more than 2.8 points below the two 560M models trained on the same pairs on law, tax and Wikipedia, within the intervals, with 768-dimensional vectors and about half their parameters; at recall@1, for which we have no intervals, it stays 7.0 points behind the E5 model on Belebele. The cost is a drop of 3.0 and 2.8 points in Russian and English recall@1, larger than for the two earlier models. We also report that, according to our run notes, a bfloat16 test run produced a non-finite loss at step 3 which the training framework logged as zero, so training was done in fp32. Our generated test questions come from the same model as the training questions, so we treat those scores as an upper limit until questions written by Tajik speakers exist.

0.25

Let the Gallery Learn Itself: Training-Free Iterative Side-View Face Recognition

cs.CV ·2026-11-08 ·Zijun Huang, Suncheng Xiang
Abstract · Work in progress

What if a face recognition system could learn from every encounter without a single gradient update? We present Gallery Self-Evolution (GSE), a training-free framework where the gallery itself adapts at test time. The core mechanism---dedup-FIFO curation---archives novel appearance samples after successful recognition while permanently locking the enrollment template, bounded at 30 embeddings per identity. On CFP-FP and IJB-C benchmarks, GSE closes the side-view gap by +23 points at 45° yaw and boosts TAR@FAR=0.001 by +9.4 points, all at CPU-level cost. No retraining. No GPU. Just a gallery that grows smarter every time it meets you.

0.24

LongCap: Hour-Scale Timestamped Audio Captioning via Chunk-wise Processing

cs.CL cs.SD ·2026-11-08 ·Jinchuan Tian, Sen Chen, Wenbin Huang, Chao Weng, Zhengwei Huang +5 more
benchmarkcode/data
Abstract · 26 pages. Project page: https://jctian98.github.io/longcap/

Describing hours of audio requires accurate event timing and sustained descriptive detail throughout the recording. We present LongCap, a framework for timestamped audio captioning (TAC), with the human-annotated LongCapBench and three diarization-inspired metrics: Timeline Error Rate (TER), Granularity Error Rate (GER), and Caption Similarity (tcpCapSim). LongCap interleaves audio chunks with structured event captions, using relative timestamps and retained context to support length extrapolation within a 1M-token inference context window. Event-level reannotation restores caption detail, while systematic studies examine model scaling, dependency supervision, and self-distillation. LongCap-Diar achieves strong benchmark results and extrapolates to 16 hours of concatenated audio despite training on recordings no longer than 40 minutes. Our TAC models are competitive with Qwen3.5-Omni-Plus on short clips, outperform evaluated baselines in caption similarity and timeline accuracy in segment-level comparisons on LongCapBench-TAC-Long, and caption full recordings approaching two hours. On LongCapBench-TAC, 9B and 35B-A3B offer favorable size-performance tradeoffs, while 122B-A10B substantially improves event-dependency recovery. These findings support long-horizon general audio understanding and provide a foundation for always-on voice agents.

0.24

FastGRT: Accelerating Training of 3D Gaussian Ray Tracing

cs.CV cs.LG ·2026-11-08 ·Meet Banthia, Xiaoteng Liu, Hexu Zhao, Aurojit Panda, Jinyang Li
code/data
Abstract

Ray Tracing enables rendering capabilities that are difficult to achieve with rasterization-based approaches such as 3D Gaussian Splatting (3DGS). However, the training cost of Gaussian-based ray tracing (GRT) is much higher than that of 3DGS. In this paper, we study how ray tracing using Gaussians executes on GPUs and show that a considerable fraction of the training cost arises from how the computation is organized. We develop FastGRT, which decouples ray traversal from blending and reuses work done in the forward pass for the backward pass. FastGRT further incorporates optimizations such as fast gradient accumulation and a hybrid BVH update strategy to tackle other major bottlenecks in the training pipeline. We implement FastGRT in Vulkan. To the best of our knowledge, it is the first Vulkan-based GRT implementation that supports both training and rendering. Compared to 3DGRT, FastGRT improves training time by 1.9$\times$ to 4.97$\times$, while preserving the original rendering formulation. Our code is open source.

0.23

SoftPAT3D: Physics-Augmented Text-to-3D Scene Generation with Deformables

cs.CV cs.GRcs.RO ·2026-11-08 ·Lei Shu, Alberta Longhini, Leonidas Guibas, Gordon Wetzstein, Jeannette Bohg
Abstract · 33 pages, 13 figures

We present SoftPAT3D, a physics-augmented framework for text-to-3D scene generation in which deformable objects, such as cloth, stuffed toys and ropes, are generated as physically valid, simulation-ready bodies and placed in intersection-free contact with the furniture that supports them, with one another and with themselves. Existing physics-augmented pipelines simulate rigid bodies only and cannot accommodate a generated deformable object, which text-to-3D generation returns as a geometrically invalid mesh, sometimes self-intersecting, that has no rest shape, no material model and no admissible initial state. These are the inputs a deformable simulation requires, and we construct each from information already in the pipeline. The rest shape comes from how the object is manufactured rather than from the generated physically inconsistent deformed configuration, and a material library assigns each deformable object its initial elastic parameters, which can then be optimized. The occluded support is regenerated and converted into a watertight collision mesh, each object is released from an optimizable placement above the supporting part that the prompt specifies, and the complete scene is driven to static equilibrium under one unified contact model, with completion verified by geometric checks and a vision-language model. When a scene fails the prompt, a grid search over placements corrects it, and the adjoint of a differentiable simulation of the same contact energy reaches the targets the grid does not, places several bodies jointly and recovers and optimizes material parameters. Multi-body scenes equilibrate in a single simulation, and the pipeline also runs from the text prompt alone. On all sixty-five prompts, SoftPAT3D produces no undesired interpenetration of any kind, whereas every other system does, and its scenes directly support scene editing and robotic manipulation.

0.23

InfCoiL: Coordinated Planner-Controller Learning for Closed-Loop Physics-Based Human-Object Interaction

cs.CV cs.LG ·2026-11-08 ·Yude Zou, Yifei Yao, Yiwei Hao, Hanqing Wang, Xing Gao
code/data
Abstract

Text- and goal-conditioned human-object interaction (HOI) provides an intuitive interface for specifying high-level intents in physics-based character animation. Mapping such intents to physically plausible, contact-rich behaviors naturally calls for a generative planner and a physics-based controller, yet existing planner-controller systems either execute plans open-loop or close the loop while keeping the planner fixed, so execution feedback never reshapes what is planned. We present InfCoiL, a closed-loop framework for physics-based HOI from coordinated planner-controller learning. InfCoiL couples a morphology-aware flow-matching planner with a unified multi-morphology controller through a shared morphology interface in an autoregressive plan-and-execute loop, and alternately fine-tunes both components with reinforcement learning from the same physical rollouts. To optimize the flow-matching planner online, we introduce Semantic Flow Policy Optimization (SFPO), which stochasticizes selected denoising transitions with channel-wise semantic covariance over heterogeneous HOI features and assigns group-relative, sequence-level rewards from complete closed-loop executions. On held-out OMOMO sequences, InfCoiL outperforms open-loop hierarchical baselines in goal-reaching task success and physical tracking accuracy. Project page: https://infcoil.github.io/

0.22

Map2SLAM: From Streaming Map Prediction to Real-Time SLAM

cs.CV ·2026-11-08 ·Jian Gao, Lin-Zhuo Chen, Yihang Chen, Jiagang Chen, Jiacheng Dong +4 more
code/data
Abstract · 17 pages, 10 figures. Project page: https://robbyant-research.github.io/Map2SLAM/

Streaming 3D foundation models (3DFMs) estimate camera poses and dense geometry online, but they fall short of SLAM: their predictions are fixed once emitted, so loop closures cannot correct accumulated drift. With such models already providing strong geometric capability, a central question for SLAM becomes how to schedule their inference and how to keep its results as revisable state. We present Map2SLAM, which keeps LingBot-Map frozen, trains only a lightweight retrieval head, and builds SLAM on the model's inference interface, turning its predictions into observations of an external persistent SLAM state. The front-end tracks every incoming frame with the model's causal streaming inference and decides which frames are written to the persistent state as keyframes. The back-end asynchronously decides when and which past frames to re-observe, re-running the model's joint inference over them with fresh context; verified loop observations correct the trajectory online, and segment observations refine dense geometry after the sequence. All invocations share a single frame encoder pass per image, whose cached features also drive loop retrieval, so frequent re-invocation incurs no additional encoding cost. Map2SLAM achieves state-of-the-art performance on KITTI, VBR, and Oxford Spires, reducing mean ATE by over 40% compared with the best uncalibrated baselines, while maintaining real-time tracking at over 28 FPS at a resolution of $518\times154$ on an NVIDIA RTX 5090D GPU. Please visit our project page at https://robbyant-research.github.io/Map2SLAM/.

0.22

Auto-GlaKG: Automated Biomarker Reasoning for Explainable Glaucoma Detection

cs.CV cs.AI ·2026-11-08 ·Cheng Huang, Zeyu Han, Yian Hu, Yaopeng Ding, Jingxi Qiu +1 more
Abstract

Interpretable glaucoma detection requires a transparent connection between fundus images, clinically meaningful biomarkers, and diagnostic predictions. Knowledge graph (KG) reasoning provides this structure, but performance measured using annotated biomarkers does not establish performance when only images are available. We present Auto-GlaKG, an automated framework that combines image-based biomarker prediction with fixed clinical-prior rules. A ResNet50 network predicts the cup-to-disc ratio (CDR) and six morphological signs using biomarker supervision, and a KG reasoner aggregates their signed rule contributions into a glaucoma score. On 689 fundus images evaluated using stratified, group-aware five-fold cross-validation, Auto-GlaKG achieves a pooled area under the receiver operating characteristic curve (AUC) of 0.951 (95\% confidence interval: 0.931--0.969), with sensitivity of 0.970 and specificity of 0.779. The matched image classifier also achieves an AUC of 0.951, while the rule stage adds 0.2~ms per image. Replacing predicted CDR with its annotation increases fold-mean AUC from 0.953 to 0.995. The framework provides an explicit rule-level account of image-based predictions and identifies CDR estimation as a priority for validation using independently measured biomarkers.

0.22

Flex-DDCM: Zero-Shot Task-Oriented Image Compression via Diffusion Trajectory Guidance

cs.CV cs.LG ·2026-11-08 ·Yixin Gao, Runsen Feng, Xin Li, Zhibo Chen
Abstract

Recent zero-shot image compression methods exploit the powerful priors of pretrained diffusion models to produce perceptually pleasing reconstructions without compression-specific training. However, image compression serves not only human viewing but also machine analysis, while existing zero-shot codecs have limited flexibility in meeting the diverse reconstruction requirements of these applications. In this paper, we propose Flex-DDCM, a framework for zero-shot task-oriented image compression via diffusion trajectory guidance. Building on Turbo-DDCM, we adapt the coded diffusion trajectory through task-guided noise selection, embedding the effects of task guidance into the compressed representation. Guidance terms can be applied individually or combined, with timestep-dependent strengths controlling their influence throughout denoising. This design enables task adaptation without introducing additional decoding complexity. Experiments demonstrate flexible trade-offs among pixel fidelity, perceptual quality, and downstream task performance, with competitive performance against trained task-oriented codecs and faster decoding than Diff-ICMH.

0.22

Continuous modeling of manipulative participation for stance analysis: Online discourse on the 2024 Street Animal Law in Türkiye

cs.CL cs.AIcs.CYcs.SI ·2026-11-08 ·Ulya Bayram, Ebru Belkıs Güzeloğlu, Emel Kuşku Özdemir, Elif Üstündağlı Erten, Öyküm Esra Yiğit
Abstract · Submitted to Journal of Computational Social Science

Studies of online political discourse commonly infer public opinion through stance analysis while implicitly assuming that observed user participation is authentic. Meanwhile, studies focusing on detecting manipulative participation are typically modeling it as a binary phenomenon, overlooking the varying degrees to which users may engage in coordinated or inauthentic behavior. This study proposes a continuous modeling approach for manipulative participation and examines its integration with stance analysis. Using online discourse on the 2024 Street Animal Law in Türkiye as a case study, we collected a dataset comprising 7,320,574 tweets from 537,994 users. From this corpus, 301 users were selected through a high-information sampling strategy and manually annotated by domain experts using a 10-criterion rubric that yields user-level continuous manipulative participation scores. Evaluation under a unified user-level cross-validation protocol showed that the proposed framework accurately estimates both manipulative participation and user stance, with gradient boosting methods providing the best manipulative participation estimates and BERTurk achieving the highest stance classification performance. External validation on an independently annotated user set demonstrated good generalization performance. The findings support modeling manipulative participation as a continuous construct rather than a binary label, providing a more informative basis for computational analyses of the relationship between manipulative participation and stance in online political discourse.

0.21

Small Data Generative AI for Culturally Specific Artistic Creation: A Case Study on the Sonic Space of sarir

cs.SD cs.CY ·2026-11-08 ·Arshia Sobhan, Paul Triana, Philippe Pasquier
dataset
Abstract · 31st International Symposium on Electronic/Emerging Art (ISEA2026)

Large-scale generative AI (GenAI) models have become increasingly integrated into contemporary creative practice, yet they often reproduce cultural biases and fail to meaningfully engage with non-Western artistic traditions. In response, this paper explores a small-data approach that foregrounds artist-curated and culturally informed dataset creation and model training. Through a case study on sarir (the distinctive sound produced by the reed pen in Persian calligraphy), we demonstrate how small-data GenAI can operate within a highly situated cultural context that lies beyond the representational reach of large-scale GenAI models. We construct a bespoke sarir audio dataset and train two models, RAVE and StyleGAN2, adapting each to the sonic domain through audio synthesis and spectrogram generation. The resulting models learn the fine-grained timbral and gestural nuances of sarir, enabling both real-time interaction and extended generative output. By comparing these results with state-of-the-art text-to-audio models, we clearly demonstrate the disparity in their performance and highlight the expressive possibilities unlocked through the small data approach. This focused case study demonstrates the feasibility of small-data GenAI for culturally specific artistic creation and underscores the value of further investigations that extend this approach to other cultural materials and artistic domains.

0.20

LocateAnyMesh: Grounding 3D Mesh Queries in 2D Images with Vision-Language Models

cs.CV cs.AI ·2026-11-08 ·Chris Choy, Zhiding Yu, Jan Kautz
Abstract

Textured 3D meshes specify object geometry and appearance, making them a natural query for finding a particular asset in an image. We introduce LocateAnyMesh, a vision-language model that uses the mesh itself as part of its prompt. A frozen 3D encoder maps colored surface points to a compact token sequence, which a learned connector places alongside image and instruction tokens. We evaluate two grounding models with PPAT mesh tokens: LAM-Select selects image-region proposals, while LAM-Direct predicts coordinates directly. Neither requires rendered query views or view-template matching. New objects require a single encoder pass and no object-specific fine-tuning. LAM-Select improves localization on the reported real-image suites and the official BOP YCB-V evaluation. Secondary experiments compare mesh encoders, and a LAM-Direct study isolates surface color. A single-seed LAM-Select continuation supplies pointing and mesh-only description evidence without replacing the reported localization model. The results support mesh tokens as a query representation, with remaining limitations in presence decisions, identity binding, proposal coverage, and descriptive fidelity.

0.20

Task-adaptive anonymisation enables privacy-preserving sharing of clinical speech data

eess.AS cs.CL ·2026-11-08 ·Ziyun Cui, Haoyu Niu, Wen Wu, Hao Sun, Sihan Zhang +10 more
Abstract

Clinical speech is increasingly used for disease screening and clinical assessment, but directly distributing speech recordings poses privacy risks because they contain both speaker-identifying voiceprints and sensitive spoken content. Existing speech anonymisation approaches are primarily developed for general speech applications or task-specific clinical speech and overlook that different clinical tasks rely on distinct speech characteristics, necessitating task-adaptive anonymisation strategies to protect patient privacy while preserving clinical utility. Here we introduce CASA, a clinically adaptive speech anonymisation agent that transforms anonymisation from a fixed processing procedure into a task-guided optimisation process, dynamically tailoring voiceprint and content anonymisation to task-specific privacy and clinical utility requirements. Through recursive self-improvement (RSI), CASA iteratively evaluates and refines anonymisation strategies against task-specific privacy and clinical utility requirements, while learning from these outcomes to guide more efficient planning in subsequent clinical tasks. CASA is validated across 14 clinical tasks encompassing five languages, diverse elicitation paradigms, and mental health, cognitive and voice disorders. CASA preserved disease-related classification performance while meeting speaker-anonymisation requirements across all clinical tasks, versus only 8/14 for the strongest fixed method. Assessments of resistance to speaker re-identification by external attackers and identification and redaction of sensitive personal information in clinical speech further demonstrated CASA's dual protection of speaker identity and sensitive spoken content. Clinical expert assessments further supported the preservation of clinically relevant information and the acceptability of anonymised speech for controlled research sharing.

0.20

Out of Parameters, Saved by Syntax: UD-PrMPT for Parser Augmented Low Resource Machine Translation

cs.CL ·2026-11-08 ·Emma Thronson, Abhishek Purushothama, Amir Zeldes
Abstract

Recent work advancing machine translation quality using LLMs for very low resource languages has relied on the injection of lexicon-based prompt components to expose vocabulary to models. However, these approaches run into a ceiling when models must guess sentence meaning purely from plausible combinations of possible word senses mapped to strings. In this paper we test UD-PrMPT, an approach using syntactic analyses in the Universal Dependencies formalism as prompt components to improve MT in five diverse low resource languages: Beja, Chintang, Chukchi, Coptic, and Xibe. Our results show additive improvements which stack on top of the lexicon based approach which are stronger for higher quality manual analyses but consistent for predicted parses as well.

0.20

WarBias: A Bilingual LLM Benchmark of Stereotypes About Internally Displaced People and Veterans

cs.CL ·2026-11-08 ·Viktoriia Makovska, Tetiana Zakharchenko, Oleksandra Ostafiichuk
benchmark
Abstract

This paper contains examples of stereotypes about veterans and displaced people that may be offensive or distressing. Social bias benchmarks for language models commonly examine categories such as gender, race, and religion, often drawn from lists of legally protected characteristics. Veterans and internally displaced people (IDPs) fall outside these lists, and stereotypes about them, which can undermine access to employment, housing, and social participation, remain largely absent from language model evaluation. We introduce WarBias, a Ukrainian-English benchmark of stereotypes about these two groups. Built from a taxonomy of 120 stereotype claims grounded in research and monitoring reports, WarBias contains 1,007 stereotype-counter-stereotype-unrelated sentence triplets and 240 multiple-choice questions per language. A Ukrainian language model selected in a blinded human pilot generated the Ukrainian triplets. At least two independent annotators reviewed each triplet in the annotation cohort. The authors revised items in response to the reviews, and the final release was machine-translated into English. Each claim carries a human severity rating. In a baseline evaluation of six models, four prefer stereotypical Ukrainian sentences about at least one group, most strongly about IDPs, while English scores are mostly balanced. WarBias supports complementary evaluations of stereotypical preferences and context-sensitive judgments, and provides paired forget/retain material for unlearning-based bias mitigation. We document the benchmark's construction, intended uses, and limitations, and offer its construction method as a template for analogous benchmarks in other conflict-affected societies.

0.20

CallCleaner: Lightweight Bioacoustic Denoising from an Audio Foundation Model

eess.AS cs.LGcs.SD ·2026-11-08 ·Michael Xie, Gašper Beguš, Morris Alper
code/data
Abstract · Project page: https://callcleaner.github.io

Many studies of animal vocalizations require clean recordings, but animals are usually recorded in the field, with background noise from weather, vehicles, and other animals. Animal vocalization denoising is challenged by the scarcity of clean training data and the extreme diversity of animal vocalizations. Classical spectral gates leave residual noise, especially without voice activity detection (VAD), while detector-free neural denoisers designed for bioacoustics remove less noise. We show that a general-purpose source separation foundation model, SAM-Audio, removes more noise than either of these types of methods, when used without training and prompted with time spans from a simple animal VAD. To turn this heavy method into a practical tool, we distill it into CallCleaner, a 19M-parameter denoiser that needs no detector and runs ~90x faster than real time on a laptop CPU. We find that SI-SDR has limited utility in our setting as it penalizes regenerated output and conflates noise removal with call fidelity, a trade-off our separate metrics reveal. CallCleaner leaves 11 dB less noise around the call than Biodenoising, the leading neural bioacoustic denoiser with public weights, and among fast, CPU-capable systems it pushes the Pareto frontier of noise removal vs. call fidelity, without requiring a detector. We release code and weights for CallCleaner, which enables fast, private, local denoising of single-caller recordings without specialized hardware.

0.20

A Layer-wise Study of How Style Rewriting Breaks the Calibration of Zero-Shot LLM Fake-News Detectors

cs.CL cs.AI ·2026-11-08 ·Prabhjot Singh, Praney Goyal, Raksha Sharma
Abstract

Large language models (LLMs) are increasingly asked to judge whether a news article is true, and their stated confidence is often taken at face value. We test what happens to that confidence when an article's writing style, but not its facts, is manipulated, as in recent \emph{style attacks} on fake-news detectors. Tracing how the prediction and its confidence form layer by layer inside the network, we study three model families (LLaMA-3-8B, Qwen2.5-7B, Mistral-7B) on three fake-news benchmarks, and find the same three patterns in every model--dataset pairing. First, on the original articles the models are competent and largely honest judges: they settle on an answer, briefly become overconfident, and then correct their confidence in the final layers. Second, style-rewritten articles break this self-correction: the models become far less accurate, under the strongest attack, worse than random guessing, yet remain just as confident, so they are \emph{confidently wrong} precisely when a warning is most needed. Third, nearly all of the damage comes from rewriting into sensational or emotionally charged styles, while sober, neutral rewrites leave the models' honesty largely intact. The practical implication is that an LLM's confidence about veracity cannot be trusted as a safety signal when an article's style may have been manipulated.

0.19

Viscot-steer: Disentangling and Transferring Visual Chain-of-Thought via Activation Steering

cs.CV cs.AI ·2026-11-08 ·Ningning Wang, Beier Zhu, Xingyu Zhu, Kaiyun Yang, Xun Yang
code/data
Abstract · Project page: https://nn-wang.github.io/viscot-steer/

Visual chain-of-thought (visual-CoT), also known as "thinking with images", equips LVLMs with the ability to locate and use key visual evidence for reasoning. However, acquiring such visual-CoT capabilities requires costly post-training and tightly couples them to a specific backbone, making transfer to other LVLMs difficult. Existing cross-model transfer methods typically treat visual-CoT as a monolithic capability, using a unified representation that ignores the functional distinction between grounding and reasoning and introduces stage-irrelevant interventions. To address this problem, we propose Viscot-steer, a training-free activation steering framework for transferring visual-CoT capabilities across heterogeneous LVLMs. Specifically, Viscot-steer first disentangles visual-CoT into grounding and reasoning vectors and extracts them from the source-model layers associated with the corresponding capabilities. It then aligns the vectors with the activation space of the target model and steers the grounding and reasoning layers during bounding-box prediction and reasoning-trace generation, respectively. Across benchmarks covering high-resolution visual understanding, general visual perception and reasoning, and mathematical visual reasoning, Viscot-steer transfers capabilities from different visual-CoT models to architecturally diverse LVLMs and outperforms the strongest capability-transfer baseline by an average of 1.86 percentage points on the primary evaluation metrics.

0.19

Query-Only Prompt-Agnostic Attacks on Visual In-Context Learning

cs.CV ·2026-11-08 ·Xiao Hu, Muhammad Ibrahim, Hao Wen, Pin-Yu Chen, Jianliang Wu +1 more
Abstract

Visual In-Context Learning (VICL) enables promptable segmentation by conditioning RGB predictions on visual demonstrations, making it attractive for auto-labeling, but its robustness under query-only access and unknown deployment prompts remains underexplored. We study query-only, prompt-agnostic targeted attacks where the attacker observes only returned RGB outputs, without access to model internals or deployment prompts. Treating prompt-conditioned RGB rendering as a distinct attack surface, we introduce two appearance-space attacks: Color Hijack (CH) drives foreground colors toward the background to suppress visibility, while Task Hijack (TH) collapses instance-distinguishing colors toward a shared color assignment. To transfer across unseen prompts, we propose a Self-Prompt Surrogate (SPS) constructed from the clean query and its output. In a paired, matched-budget SegGPT comparison, SPS outperforms the Random-Prompt Surrogate (RPS), reducing CH instance foreground retention from 0.918 to 0.155 and TH palette-aware tolerance distance from 0.345 to 0.225. Across four VICL models on COCO and HierText, CH suppresses foregrounds across semantic-, instance-, and text-style outputs, while TH collapses instance colors under held-out deployment prompts. After palette-based decoding, TH also reduces the decoded instance count and mask retention while increasing instance merges. Finally, at 30% corruption, CH-corrupted annotations from a publicly accessible VICL-based auto-labeling pipeline reduce downstream YOLO-seg mAP50 from 0.761 to 0.281.

0.19

EASE-KV: Diagnosing the Gap Between Memory Availability and Effective Use

cs.CL cs.AI ·2026-11-08 ·Di Lu, Zechen Li, Xuanyu Yue, Jiabo Fan, Hao Xue +1 more
Abstract · Preprint

Reusable key-value (KV) memory avoids repeatedly prefilling retrieved text, but stored and injected memories do not necessarily become usable evidence. We present EASE-KV, a controlled framework for understanding this gap under fixed candidates and a frozen language model. EASE-KV separates three interfaces: whether required relations are preserved during encoding, whether preserved memories receive appropriate influence during allocation, and whether outcome supervision can identify useful allocations. Matched experiments show that causal encoding improves compositional accuracy by 25.0 percentage points over block-diagonal encoding, while the tested inference-time bridging methods fail to recover the lost contextual information. With unchanged stored states, diagnostic evidence-guided allocation improves LongMemEval accuracy from 31.2% to 38.4%, demonstrating that memory influence can be controlled without re-encoding. However, controllability does not guarantee learnability: binary correctness provides informative reward contrast for 53.0% of sampled query groups, compared with 73.6% under the evaluated shaped-reward setting. These findings show that effective KV memory requires more than retrieval and injection: the required relations must be encoded, preserved memories must receive appropriate influence, and supervision must distinguish better allocations from worse ones.

0.19

TID System for Speech-COSER Challenge 2026: ASR Track

cs.CL cs.AI ·2026-11-08 ·Fernando López, Jordi Luque
Abstract · TID system for Speech-COSER Challenge 2026

We describe the system submitted by Telefónica Innovación Digital to the Speech-COSER Challenge 2026, which features spontaneous speech from elderly speakers recorded in rural Spain. For ASR, we perform CTC-based pseudo-forced alignment scores audio-text match quality to guide a two-stage LoRA curriculum, followed by post-hoc magnitude pruning of the LoRA matrices to enhance generalization. An ensemble of seven systems combined via ROVER achieves 36.25\% unnormalized and 22.17\% normalized WER.

0.18

SurgWorld: Learning Surgical Representations Beyond Observation via Latent Prediction

cs.CV ·2026-11-08 ·Jingkun Yue, Pengxiang Wang, Siqi Zhang, Zinan Jia, Yiyao Zhang +5 more
Abstract · This work has been submitted to the IEEE for possible publication

Surgical procedures are inherently partially observable, requiring surgeons to continuously infer evolving anatomical dynamics and future surgical states beyond the current observation. However, existing self-supervised learning methods primarily learn from observed surgical videos through masked reconstruction or predicting representations within the camera's current field of view, lacking the capability to explicitly model the out-of-sight surgical dynamics. Inspired by latent predictive world models, we introduce SurgWorld, a video-native, self-supervised framework for surgical representation learning, which treats a specified spatial or temporal shift as a pseudo-action and learns to predict the target latent representation from a partial observation. Specifically, we propose Out-of-Sight Dynamics Modeling (OSDM) as the pretraining objective for predicting representations conditioned on relative spatio-temporal shifts. We further regularize the learned latent dynamics from OSDM with a Geometric Visual Similarity (GVS) objective by aligning patch-level relational structures between predicted and target representations, preserving spatial geometry and local visual coherence for dense perception. These two objectives are integrated through an agglomerative training framework that consolidates complementary supervisory signals from multiple teachers to learn a unified latent representation of surgical dynamics and visual geometry. Extensive transfer learning experiments on 17 surgical benchmarks show that SurgWorld consistently outperforms existing vision foundation models across surgical workflow recognition, fine-grained action understanding, and dense visual perception. We further conduct tailored qualitative and quantitative analyses, demonstrating that SurgWorld effectively captures surgical dynamics beyond the directly observed context through latent prediction.

0.17

Search-by-Answer Bias in Retrieval Models

cs.CL cs.IR ·2026-11-08 ·Adiel Avigad, Shauli Ravfogel, Yoav Goldberg
Abstract · 17 pages, 9 figures

Dense retrievers are meant to select passages that match a query's semantic specification. We show they suffer from search-by-answer bias: they favor passages that mention answer objects the model knows (scores as likely), over passages that support valid alternative answers. We test this with controlled factual contrasts from LAMA/TREx. For each retriever, we measure how strongly the model scores the canonical object over an alternative object using a token-prediction head, and ask whether this score predicts passage rankings. Across eight retrievers, including DPR, Spider variants, Contriever, E5, and Qwen3-Embedding, larger token-prediction scores consistently predict higher retrieval scores for passages containing the corresponding object. This holds when passages differ only in the answer object, and even when the passage that mentions the high-scoring object expresses a wrong relation. Dense retrieval thus exhibits search-by-answer bias during passage selection, shaping which evidence is retrieved.

0.17

Does Pose Encoding Matter? Analyzing Geometric Representations for Discrete Motion Vocabulary Learning

cs.CV cs.AIcs.DM ·2026-11-08 ·Amir Kimiyaie, Nahal Mafi, Hamed Tabkhi
Abstract

Compressing continuous human motion into a finite vocabulary of discrete tokens is now the first stage of many motion understanding and generation pipelines. Vector quantization partitions a latent space into Voronoi cells around a finite codebook, but the geometry of that latent space is inherited from the input pose encoding, and it is unclear which encoding admits an efficient discretization. We study four encodings that place skeletal motion in different geometric spaces: Gaussian joint heatmaps on the pixel lattice, which we show to be a morphological dilation of the joint point set by a quadratic structuring function; root-relative joint positions in Euclidean space; local joint rotations on $\SO$ through a continuous 6D embedding; and a redundant full kinematic encoding. For each, we train VQ-VAE tokenizers whose codebooks lie on the unit hypersphere, varying codebook size, temporal stride, and code dimension. Beyond reconstruction, we measure codebook utilization and the effective vocabulary size, i.e., the number of equiprobable spherical Voronoi cells with the same entropy as the observed token distribution. Rotation and position encodings keep converting additional codewords into effective vocabulary, whereas the full kinematic encoding saturates at a few hundred effective cells and heatmaps at fewer than sixty. The smallest code dimension gives the highest utilization for three of the four encodings, and the effect of temporal stride depends on the encoding. Encodings of the same pose family can thus yield effective vocabularies differing by an order of magnitude, a difference that reconstruction loss alone does not reveal, which points to input geometry as a first-order design choice in motion tokenization.

0.17

Differentiable Virtual-Slot Partitioning for Adaptive Tokenization in Vision Transformers

cs.CV ·2026-11-08 ·Gaoxiang He, Bin Zhu, Bo Xie, Xiaotong Yu, Yi Chen +1 more
Abstract

Adaptive tokenization aims to reduce the compute spent on uninformative image regions while preserving accuracy. Existing work relies on (i) fixed or frozen scoring rules, which cannot be trained toward the task loss, or (ii) discrete masks trained with straight-through estimators, which yield noisy gradients, while fully differentiable alternatives instead leave the pre-trained regular grid. We revisit adaptive tokenization from a regular-grid perspective and ask: can a two-scale quadtree be learned end-to-end without deleting a single token during training? We show that one coarse token and its four fine children can be made the two endpoints of a single continuous interpolation. We introduce Differentiable Virtual-Slot Partitioning (\dvsp{}), which keeps four \emph{virtual slots} per $2{\times}2$ region and blends their token features and attention mass with a single weight $α$, exactly reproducing the discrete grid at $α\in \{0,1\}$. On COCO detection with EVA-02-B, \dvsp{} retains $97.9\%$ of dense Box AP while removing $64.2\%$ of visual tokens, giving a $1.61\times$ end-to-end speedup, and it outperforms entropy-based APT, DART, and uniform lower-resolution dense adaptation at matched budgets; these gains depend on the conserved attention mass, whose removal costs $1.59$ Box AP. On a frozen Segmenter it also improves Cityscapes mIoU, and on frozen DeiT backbones it exceeds entropy selection at identical per-image token counts.

0.16

MoTo: Mixture of Tokenizers Towards Fair Multilingual Language Modeling

cs.CL cs.LG ·2026-11-08 ·Gül Sena Altıntaş, Colin Raffel
Abstract · Neurips 2026, 40 pages, 21 figures

Large language models rely on a single tokenizer that is chosen at training time and fixed thereafter. Most tokenizers are trained on English-dominant corpora, making them under-serve morphologically rich and non-Latin-script languages, where they can produce fragmented representations, longer sequence lengths, and reduced information density during training. We propose Mixture of Tokenizers MoTo, a modular tokenization framework that trains dedicated per-language BPE tokenizers and composes them into a unified superset vocabulary, where tokens shared across languages are deduplicated into a common ID space. Each tokenizer maintains its own normalization policy, pre-tokenization rules, and vocabulary budget. Despite allocating significantly fewer entries to English, MoTo retains English performance while delivering consistent gains across 18 out of 21 typologically diverse languages spanning six scripts, with improvements most pronounced on morphologically complex and non-Latin-script languages that are systematically underserved by English-dominant tokenizer design. Our results suggest that modular subword tokenization is a practical and extensible alternative to monolithic tokenization for multilingual parity.

0.16

TransArt3D: From Two States to Articulated 3D Reconstruction with Video Transition Priors

cs.CV ·2026-11-08 ·Chenhao Guan, Liying Yang, Xinxing Yu, Hao Mo, Tianshun Han +2 more
Abstract

Reconstructing interactable articulated assets from a few images is key to robotics and simulation, yet highly underconstrained. Existing optimization-based methods fit an articulated asset under a static 3D prior that scores each state independently, so per-state reconstruction errors leak into the estimated moving part and joint as false motion. Video diffusion models capture how objects move between states, but only implicitly in 2D. To this end, we present TransArt3D, an optimization-based framework that reconstructs an articulated 3D asset from two endpoint images by translating this video transition prior into explicit part structure and joint kinematics. It first lifts frames generated between the endpoints with a transition-coupled 3D consensus that applies the static 3D prior jointly across frames, stabilizing unchanging regions and initializing the parts and joint. A rectified-flow kinematic refinement then lets the video prior supervise a 3D motion field over the whole transition, while part assignment and a shared joint constrain the field throughout optimization, keeping per-state errors out of the articulation. Finally, a motion-aware canonical appearance fusion uses the recovered articulation to map both endpoint images onto one canonical surface, texturing regions that only one state reveals. On PartNet-Mobility, TransArt3D outperforms methods given the same two endpoints in geometry and joint accuracy, and surpasses methods given six observed states in articulated-state geometry, joint localization and appearance.

0.14

Source-Anchored Visual Autoregression for Extreme Image Compression

cs.CV ·2026-11-08 ·Bingxin Wang, Yuan Lan, Yang Xiang, Yiqing Huang
Abstract

In the extreme-low bitrate range below $0.01$ bpp, image compression becomes a constrained generation problem: the reconstruction depends on inferring plausible missing details while staying close to the input image. Autoregressive token models provide an efficient coarse-to-fine prior for this setting, but the transmitted early-scale tokens often leave object identity and texture underdetermined, which can lead to source-inconsistent reconstructions. We introduce Source-Anchored Visual Autoregression (SAVAR), a generative codec built on Visual Autoregressive (VAR) modeling. SAVAR transmits early-scale tokens and a compact caption. The tokens provide image-specific layout information, while the caption supplies global semantics. SAVAR reconstructs a partial VQ-VAE latent from the retained scales, maps it to the start-of-sequence initialization, and uses the same latent to condition the caption embedding before adaptive layer normalization, separating two roles that are typically driven by a shared condition in VAR-style generation. On the Kodak and CLIC2020 datasets in the extreme-low range, SAVAR reduces perceptual distortion relative to matched ARPC retained-scale operating points, lowering LPIPS by about $8$--$9\%$ and DISTS by up to $22\%$ at a lower bpp. It reconstructs a $512 \times 512$ image in about $218$ ms on one NVIDIA L20X GPU, retaining fast autoregressive decoding in the evaluated setting.

0.14

Perturbing Inputs Improves Pretraining Distillation

cs.CL cs.AI ·2026-11-08 ·Ziyang Cai, Yihe Dong, Dimitris Papailiopoulos, Shivam Garg
Abstract · 14 pages, 7 figures, 4 tables

Knowledge distillation trains a student language model to match a teacher's predictions on the training text. We ask whether learning the teacher's behavior beyond the training text can further improve the student. We investigate this using three input perturbations: additive token noise, blending two documents, and blending neighboring tokens. Specifically, for an early part of pretraining we perturb the inputs and train the student to match the teacher on them, and we spend the rest of the same token budget on ordinary distillation. When pretraining from scratch, the best method improves on standard distillation at both scales we tried, by 0.025 nats at 135M parameters and by 0.020 nats at 0.5B (mean of three seeds). At 135M this is 1.45 times the gain that distillation itself provides over next-token training. We also observe promising downstream transfer: at 0.5B, the best variant improves average zero-shot accuracy by about one percentage point across seven tasks. It also retains its validation-loss advantage after low-bit weight quantization. We also test a data-limited setting in which training data must be reused, and find that the gains persist. Finally, ablations suggest that, for additive noise, perturbing the teacher's inputs contributes most of the gain, while perturbing only the student's inputs yields much less.

0.14

Understanding When and Where Self-Supervised Speech Models Learn Phonemes and Prosodic Cues through Layer-Wise and Time-Resolved Probing

cs.CL ·2026-11-08 ·Masaru Tanibata, Shun Takahashi, Hiroki Ouchi, Sakriani Sakti
Abstract

Self-supervised speech models (S3Ms) encode a rich hierarchy of acoustic, prosodic, and phonetic information, yet when and where it emerges during pretraining remains unclear. We train frozen-encoder probes on every Transformer layer across pretraining checkpoints of two matched models: HuBERT-Base, whose stage 2 is a separately initialized model trained on new pseudo-label targets, and single-objective wav2vec 2.0-Base. Our five tasks span an acoustic-to-abstract gradient: absolute pitch (F0), speaker-relative pitch, lexical stress, phoneme boundaries, and 38-way phoneme identity. Acoustic-prosodic cues and phone boundaries are decodable from the earliest checkpoints and barely shift across layers, whereas phoneme identity is constructed during pretraining, re-localizing to upper-middle layers under HuBERT's stage-2 targets at a matched update count. Acquisition order therefore tracks abstraction from the acoustic surface, not the segmental/suprasegmental distinction, extending to prosody an account established for phonetic, lexical and syntactic structure; the no-switch wav2vec 2.0 shows no comparable re-localization.

0.13

DIY Customized Speech Recognition for a Person with Speech Disabilities

cs.SD cs.CY ·2026-11-08 ·Paulo A. Condado, Fernando G. Lobo
position
Abstract

General-purpose state-of-the-art automatic speech recognition (ASR) systems fail systematically for speakers with moderate to severe speech disabilities. This paper describes the design, implementation, and evaluation of a customized ASR system built for the first author, a researcher with cerebral palsy and dysarthria. We fine-tuned OpenAI's Whisper medium and large-v3 models using Low-Rank Adaptation (LoRA) on audio samples recorded from the first author's own speech on consumer-grade hardware. On a held-out test set of 100 previously unseen utterances, the corresponding transcriptions resulted in a Word Error Rate drop from 49% to 17% for medium and from 30% to 12% for large-v3, while the Character Error Rate fell from 26% to 7% for medium and from 16% to 5% for large-v3, indicating that the residual errors are minor phonemic substitutions. Based on these models, the first author developed AVoz, a mobile application that acts as a real-time interpreter, converting his speech into synthesized voice for the listener. All processing is performed locally, ensuring that no audio data leaves the user's machine. We argue that the methodology is replicable. Moreover, we publish the full LoRA configuration, the implementation, and a step-by-step recording and training protocol, so that other people with speech disabilities, or their family members and caregivers, can build their own systems with a few hours of labeled audio data and a couple of days of GPU time.

0.12

Reaching Is Not Finding: Map-Anchored Visual Prompting for the Last Mile of Aerial Vision-Language Navigation

cs.CV ·2026-11-08 ·Ningbo Li, Kebin Liu, Jingcan Chen, Weichen Zhang, Rong Ding +2 more
benchmark
Abstract

Aerial vision-language navigation (VLN) asks a drone to find an object described in natural language. The description usually locates the object relative to named landmarks, such as streets and buildings. We observe that current navigators often bring the drone close to the target and then stop at the wrong place: they reach the target without finding it. This gap is shared by navigators of very different designs, from imitation learners to a fine-tuned 7B vision-language model. A simple test reveals the cause. When we replace the target object in the instruction with another one next to the same landmarks, no navigator goes to it, and most stop where they did before. Navigators learn to reach the landmarks, not to identify the described object. We therefore propose MAP-VP, a plug-and-play module for the last step. After any navigator stops, MAP-VP draws the landmarks named in the instruction onto the drone's current view, using the landmark outlines that the benchmark provides on a map. A fine-tuned 4B vision-language model then locates the target relative to these landmarks. The training data are generated automatically from the benchmark annotations. Neither drawing the landmarks nor fine-tuning helps much alone, but together they are highly effective. Without retraining any navigator, MAP-VP significantly improves all six navigators we test on CityNav. It roughly doubles the success rate of the two strongest trained navigators, raising HETT from 25.9% to 54.6% and FlightGPT from 24.2% to 48.9%, whereas CityNav's official detector-based last step adds less than two points to HETT.

0.12

NanoNet: Ultra-Lightweight Network for MRI Brain Tumor Classification toward Edge AI

cs.CV ·2026-11-08 ·Tran Nguyen-Le, Huu-Thuan Huynh, Minh-Triet Tran, Trung-Nghia Le
benchmark
Abstract

Automated brain tumor classification from magnetic resonance imaging (MRI) can support accessible medical image analysis, but conventional deep learning models often exceed the resource budgets of highly constrained edge-AI platforms. We present NanoNet, an ultra-lightweight convolutional neural network designed for resourceconstrained MRI classification and motivated by future edge-AI applications. Inspired by MobileNetV2, NanoNet substantially reduces network depth, channel capacity, expansion ratios, and final feature dimensionality, resulting in only five inverted residual blocks and 8,132 parameters. We further investigate dependency-aware structured pruning, knowledge distillation, and INT8 quantization-aware training for additional compression. Experiments on the BRISC 2025 benchmark show that NanoNet achieves 92.80% accuracy and 92.97% macro F1 with 8,132 parameters and 10.498 MFLOPs. Across three compression seeds, it exceeds the mean accuracy of 70%-pruned MobileNetV2 by 5.20 percentage points while using 60.2% fewer parameters and 9.8% less computation. Converted INT8 NanoNet achieves 89.70% accuracy and 90.22% macro F1 across three QAT seeds. Earlier TensorFlow Lite measurements indicate 27.7 KB of baseline model storage, motivating future edge-AI deployment.

0.12

Semantically Aligned KV Cache Compression for Efficient Agentic Autoregressive Video Generation

cs.CV cs.LG ·2026-11-08 ·Weizhong Huang, Huaijia Kou, Tianyu Xie, Xiawu Zheng, Liujuan Cao
theory
Abstract · 32 pages, 12 figures

Agentic autoregressive video generation combines semantic planning, sequential generation, and visual verification for controllable synthesis. Growing visual histories enlarge the key-value (KV) cache, increasing memory use and attention costs across candidate rollouts. Existing compression methods primarily optimize individual generators without the planner's semantic constraints, risking low priority for states supporting requested attributes or actions. We introduce SemanticKV, a training-free method for semantically aligned KV cache compression. To our knowledge, SemanticKV is the first KV cache compression method for agentic autoregressive video generation that explicitly uses planner-extracted semantic constraints to guide retention. Text cross-attention identifies constraint-related video positions, whose self-attention queries are collected into separate groups. Attention weights are averaged within each group and maximized across groups, giving each cached entry a high score when strongly used by any represented constraint. A single top-k selection retains high-scoring entries under the memory budget, leaving their keys, values, and positions unchanged. For fixed query groups and equal entry costs, this selection minimizes an additive upper bound on worst-group discarded attention. The method operates independently within each candidate without retraining or additional video generations. We evaluate four AR video backbones on three benchmarks for 5-, 30-, and 60-second generation. On Causal Forcing at a compression ratio of 0.7, SemanticKV obtains a VBench Total Score of 84.10, compared with 84.05 for the strongest compressed baseline and 83.99 for Full KV. It achieves a 1.22x speedup over Full KV and reduces KV-cache memory from 6.04 GB to 1.81 GB, a 3.35x reduction.

0.12

Beyond the Transcript: Measuring and Improving the Use of Acoustic Information in Audio Language Models

cs.CL ·2026-11-08 ·Linkai Peng, Baorian Nuchged, Yuyang Yao
benchmark
Abstract

Audio language model benchmarks provide comprehensive evaluation across diverse tasks. However, some tasks can be answered from the transcript or even the question alone, and benchmark scores can therefore confound acoustic understanding with language modeling ability. We distinguish lexical tasks, for which the transcript usually suffices, from global-attribute and within-recording-structure tasks, which require the model to analyse the recording beyond its words, either as a whole, for properties such as the speaker and emotion, or within it, for patterns such as stress and intonation. We evaluated 19 open models and Gemini 3.1 Pro on 92 tasks from 6 benchmarks and estimated the recording gain, the accuracy advantage of the original recording over references that retain its words, by comparing the recording with synthetic readings of its transcript and by adding it to the transcript. On attribute and structure tasks, the open models fall behind Gemini by 28.6 and 41.9 percentage points, and about half or more of each deficit lies in the recording gain. Nevertheless, linear probes in Gemma-4 12B and Qwen2.5-Omni-7B show that the acoustic information needed for answering is already readable in early language-model layers, while the readout of many cues declines in the layers that follow. We therefore constructed training data and trained LoRA adapters in these layers, with a content anchor, which raises attribute and structure accuracy by 12.8 to 44.4 points in the two models and keeps mean lexical accuracy within 1 point of the base model. Most of the improvement appears as larger recording gains. These results suggest that training improves the use of acoustic information that the models already encode.

0.10

PhysMoE: Fast Spatial Grounding with Single-Token Coordinate Representation in Modality-Decoupled VLMs

cs.CV cs.AIcs.RO ·2026-11-08 ·Ruimeng Liu, Denan Liang, Ruofei Bai, Fuming Zhang, Kuan Xu +2 more
Abstract

Embodied Vision-Language Models (VLMs) are backbones for robotic planning and manipulation. However, for coordinate prediction in grounding tasks, existing architectures suffer from two distinct bottlenecks: discrete tokenization disrupts spatial continuity with high decoding latency, while standard point regression fails on multi-contact affordances and irregularly shaped targets. To resolve these challenges, we propose PhysMoE, introducing three core contributions: (1) sharing lower layers for multimodal context while routing upper layers to specialized spatial (pixel) and linguistic (text) experts by physical modality, eliminating gradient interference; (2) encoding each coordinate as a single token by reusing the visual encoder's learned positional embeddings, injecting 2D spatial priors to natively align coordinates with visual features and ease training; and (3) differentiable Signed Distance Field (SDF) supervision on pointing, accommodating affordances and irregular geometries without costly mask decoders. Compared to the base model, PhysMoE achieves a 15.7% average gain and an 8.31x throughput speedup across 9 coordinate-based grounding benchmarks (trajectory, affordance, and localization), while preserving multimodal understanding and reasoning across 11 text-based embodied benchmarks.

0.09

ForeAct: Geometry-Guided Action Learning from World Representations

cs.CV cs.RO ·2026-11-08 ·Wei Lu, Jiaxi Zhang, Yunheng Wang, Taowen Wang, Yuetong Fang +6 more
Abstract

Robotic manipulation requires representations that capture both scene geometry and its evolution during interaction. Existing approaches have largely treated these two aspects separately. World action models capture such dynamics by coupling action prediction with future video generation, but their reliance on iterative diffusion or rectified-flow inference leads to high computational cost, while their representations lack explicit spatial structure. Explicitly modeling scene geometry typically requires auxiliary modules that increase both parameter count and computational cost. Conversely, geometry-centric policies primarily recover observed spatial structure and often lack explicit supervision for anticipating future geometric changes during interaction. To address these limitations, we introduce ForeAct, which repurposes a pretrained video world model as a single-pass representation backbone for low-latency control. ForeAct directly processes the clean latent of the current observation in one forward pass, eliminating future-frame synthesis and iterative denoising while retaining the backbone's pretrained spatiotemporal priors. Rather than introducing a separate geometry backbone, ForeAct uses learnable tokens supervised by next-frame relative depth to shape the shared video representation through bidirectional self-attention. This compact interface turns the spatial knowledge already encoded in the backbone into a future-aware representation for action generation, with little parameter overhead. The lightweight geometry readout is used only during training and removed at deployment, enabling low-latency, geometry-aware control without depth input or an external geometry backbone. Without additional large-scale pretraining, ForeAct achieves 96.6% average success on LIBERO, 63.5% on SimplerEnv-Bridge, and a 74.4% success rate across 90 real-world rollouts.

0.09

GenRM-Flow: Generators are Process-aware Reward Models in Flow Matching

cs.CV ·2026-11-08 ·Siming Fu, Ruizhe He, Zheming Fu, Zeyue Xue, Fengkai Liu +5 more
position
Abstract

Aligning video generation models with human preference depends critically on the quality of the reward model that scores generated videos. Current video reward models, however, source their reward signal from a model's representational rather than generative capacity: they either repurpose vision-language models as pixel-space scorers, or train auxiliary scoring heads on top of frozen generator features. The generator's actual ability to predict velocities, denoise latents, and produce videos plays no role in scoring its own generations. We argue that this overlooks a structural property of offline preference learning. Any preference optimization objective that approximates an offline preference dataset mathematically targets a strictly defined KL-constrained optimal policy. By isolating the reward term from this analytical solution and mapping the exact generation log-likelihood to its flow-matching surrogate, the empirical reward emerges deterministically as the relative difference in continuous-time velocity-prediction errors. We formalize this equivalence into GenRM-Flow, a framework where a preference-tuned video generator inherently operates as a process-aware reward model directly in the noisy latent space. We further show theoretically that distinct fine-tuning objectives, whether contrastive like DPO and IPO or non-contrastive like RWR, are alternative optimization pathways toward the same optimal policy, each inducing a reward readout from the generator's velocity-prediction errors. Furthermore, the extracted process reward seamlessly substitutes external VLM scorers in FlowAWR. Operating purely on native latent outputs without auxiliary reward networks, this generative reward drives measurable improvements in downstream video generation.

0.08

From Game Generation to Game Playing: Fully Automated Ability Training for Vision-Language-Action Models

cs.CV ·2026-11-08 ·Changran Liu, Qi Wu, Wenchao Ma, Sharon X. Huang, Haomiao Jiang
Abstract

Training vision-language-action (VLA) models to follow gameplay instructions across diverse games is limited by the cost of collecting demonstrations. We introduce a framework that automatically generates game environments and expert demonstrations. An LLM-powered agent uses game-building tools to construct executable environments and tasks targeting specified skills. Programmatic experts then use privileged simulator state to generate aligned instructions, rendered observations, and keyboard and mouse actions. We train our game policy, Script2Play, by post-training a 4B vision-language model with behavior cloning on 150.3 hours of synthetic Minecraft and Roblox gameplay, without human gameplay demonstrations or reinforcement learning. Evaluation on multiple public game benchmarks demonstrates that Script2Play follows gameplay instructions effectively and generalizes across game engines, highlighting the effectiveness of targeted synthetic supervision.

0.08

PCG-RAG: Progressive Compositional Grounding for Retrieval-Augmented Generation in Complex Web Question Answering

cs.IR ·2026-11-08 ·Linyao Zheng, Xuhang Shi, Zhifang Mao, Chongqi Qu, Xiuquan Hou
Abstract

Complex Web question answering requires retrieval-augmented generation (RAG) systems to identify target entities from multi-clue descriptions and connect evidence scattered across long Web pages. Individual pages often support only some of the required semantic constraints, while clues discovered on one page guide subsequent searches on others. However, premature reliance on partially supported entity hypotheses can divert retrieval from the original information need. RAG systems must therefore use partial evidence to guide further searches while seeking the evidence needed to establish the intended entities. We propose PCG-RAG, a novel RAG framework based on Progressive Compositional Grounding (PCG). Progressive Evidence Acquisition directs subsequent searches toward missing evidence for candidate attributes and relations. Compositional Reasoning separates candidate-guided retrieval from evidence-based answering along a directed acyclic graph (DAG) of subquestions. Provisional names make dependent searches specific, while answering tasks use accepted results and unresolved identifying conditions, assessed against the original question. New passages connect the identifying clues to the requested relations, supporting evidence-based downstream answers. Evidence Grounding combines identifying and relational source passages with supported intermediate answers, connecting evidence across Web pages to the original conditions during final generation. On BrowseComp-Plus and WebWalkerQA, PCG-RAG improves accuracy over the leading compared baselines by 5.43 and 5.15 percentage points, respectively.

0.06

When Should Diffusion Language Models Reconsider Their Commitments?

cs.CL cs.AIcs.LG ·2026-11-08 ·Haoyang Zheng, Yidong Ouyang, Blake Jianhang Chen, Tian Tan, Ziqian Lin +3 more
Abstract · This paper is being withdrawn because agreement has not yet been reached regarding Google's authorization for its public release

Masked Diffusion Models (MDMs) are well known for fast text generation via parallel decoding. However, independent parallel sampling can commit to inconsistent predictions, and less is known about how to selectively revise these early token commitments. In this work, we introduce $\underline{\mathrm{P}}$lanned $\underline{\mathrm{U}}$nmasking-and-$\underline{\mathrm{R}}$evision Joint $\underline{\mathrm{L}}$earning (PURL) for MDMs, a training-based framework that seamlessly unifies parallel generation with selective token revision. We derive a normalized joint transition whose likelihood factorizes into edit selection, unmask selection, and token prediction. These components are trained jointly on partially masked sequences constructed according to the model's own unmasking policy. During inference, all three decisions are supported by a single shared forward pass, enabling the model to decode masked tokens and revise noisy tokens. Across logic puzzles, mathematical reasoning, and code generation, PURL improves accuracy at matched unmask budgets, with relative gains up to 47.4% accuracy improvement and up to an $8\times$ reduction in scheduled decoding rounds. Beyond these gains, our analysis further reveals general principles of revisable decoding: token edits become more effective as context accumulates; early edits can steer unfinished trajectories, whereas later edits tend to remain localized; and effective edits therefore depend on both calibrating when to intervene and learning where and how to revise. Together, these suggest that effective parallel generation depends not on revising more tokens, but on making the right commitments revisable at the right time.

0.06

Trajectory-Regularized Distribution Distillation for One-Step Generation

cs.CV cs.AI ·2026-11-08 ·An Zhao, Yixiang Zhou, Zejian Li, Yi Guo, Pei Chen +1 more
Abstract

Distilling flow matching models for one-step generation generally follows two paradigms. Trajectory alignment supervises the student with the teacher's probability flow ODE, but its performance degrades under very small sampling budgets. Distribution matching directly optimizes the generated distribution, but estimating its reverse Kullback-Leibler gradient typically requires a separate network to track the evolving student score. We propose trajectory-regularized distribution distillation (TRD), which combines distribution-level and trajectory-level supervision through a momentum-mixture policy predicted in a single student forward pass. This policy serves simultaneously as a one-step generator, an estimator of the student's marginal velocity that removes the need for a second trainable network in distribution matching, and a closed-form time-dependent velocity field supervised along the teacher ODE trajectory. Experiments on SANA-1.6B and Stable Diffusion 3.5 show that TRD outperforms representative consistency and distribution matching baselines in generation quality and training efficiency for both one-step and few-step generation, with a trade-off in sample diversity.

0.05

Security & Privacy 19

EVAX: Towards a Practical, Pro-active & Adaptive Architecture for High Performance & Security

cs.CR ·2026-11-08 ·Samira Mirbagher Ajorpaz, Daniel Moghimi, Jeffrey Neal Collins, Gilles Pokam, Nael Abu-Ghazaleh +1 more
top venuehighly cited
Abstract · Author manuscript of 2022 55th IEEE/ACM International Symposium on Microarchitecture (MICRO), pp. 1218-1236 (2022). 19 pages

This paper provides an end-to-end solution to defend against known microarchitectural attacks such as speculative execution attacks, fault-injection attacks, covert and side channel attacks, and unknown or evasive versions of these attacks. Current defenses are attack specific and can have unacceptably high performance overhead. We propose an approach that reduces the overhead of state-of-art defenses by over 95%, by applying defenses only when attacks are detected. Many current proposed mitigations are not practical for deployment; for example, InvisiSpec has 27% overhead and Fencing has 74% overhead while protecting against only Spectre attacks. Other mitigations carry similar performance penalties. We reduce the overhead for InvisiSpec to 1.26% and for Fencing to 3.45% offering performance and security for not only spectre attacks but other known transient attacks as well, including the dangerous class of LVI and Rowhammer attacks, as well as covering a large set of future evasive and zero-day attacks. Critical to our approach is an accurate detector that is not fooled by evasive attacks and that can generalize to novel zero-day attacks. We use a novel Generative framework, Evasion Vaccination (EVAX) for training ML models and engineering new security-centric performance counters. EVAX significantly increases sensitivity to detect and classify attacks in time for mitigation to be deployed with low false positives (4 FPs in every 1M instructions in our experiments). Such performance enables efficient and timely mitigations, enabling the processor to automatically switch between performance and security as needed.

0.71

Using the Planted Clique Conjecture for Cryptography: Public-Key Encryption from Planted Clique and Noisy $k$-$\mathsf{LIN}$ Over Expanders

cs.CR ·2026-11-08 ·Riddhi Ghosal, Isaac M Hair, Aayush Jain, Amit Sahai
top venue
Abstract

We give a public key encryption scheme that is provably secure against poly-size adversaries, assuming $n^{\log^αn}$ hardness of the standard planted clique conjecture, for any $α\in (0,1)$, and a relatively mild hardness conjecture about noisy $k\mbox{-}\mathsf{LIN}$ over expanders that is not known to imply public-key encryption on its own. Both of our conjectures correspond to natural average-case variants of NP-complete problems and have been studied for multiple decades, with unconditional lower bounds supporting them in a variety of restricted models of computation. Our encryption scheme answers an open question in a seminal work by Applebaum, Barak, and Wigderson [STOC'10].

0.71

Self-Healing by design: Immune-system inspired autonomous cyber resilience

cs.CR ·2026-11-08 ·Shari Finner, Swarna Kumarswamy-Das, Bart Gijsen
published
Abstract · The content of this paper was presented at MAST Australia 2024, 19-21 November in Adelaide. The presentation was accepted after review of th…

Digital innovations have strongly improved ship capabilities, including command, control & communication, navigation and combat management. In the wake of digitisation and geopolitical stress, resilience against cyber-attacks has become a key challenge. Autonomous Cyber Resilience (ACR) methods, inspired by features of the human immune system, provide a promising perspective. Proofs of concept of autonomous self-healing and self-protecting software have been realised in cloud environments using containerisation, monitoring and deployment technology. However, developing a suitable "cyber immune system" for more complex digital systems is still an intricate challenge. In this paper, we share insights on how ACR methods can be combined with resilience mechanisms present in critical naval and aeronautical manoeuvring systems. Using TNO's "OpPic" application, we show how TNO's Self-Healing for Cyber Security software can provide autonomous mission- and threat-adaptive resilience. On this basis, six guidelines for the design of autonomous cyber resilient systems are derived. The ultimate goal is to enable operators to oversee and control their system's "cyber immune response", to be able to continuously adapt to evolving threats and mission priorities.

0.63

SkillCODER: Towards Auditing and Attribution of Copyright Infringement in LLM Agent Skills

cs.CR ·2026-11-08 ·Enhao Huang, Chunshu Xia, Yiming Li, Yuchen Yang, Bingrun Yang +3 more
top venuecode/data
Abstract · 29 pages, 6 figures. Accepted at ACM CCS 2026. Code: https://github.com/EonHao/SkillCODER

Agent skills package reusable workflows and task-specific procedures into deployable natural-language artifacts for LLM agents. Their scale and readability make them valuable but vulnerable to unauthorized redistribution and repackaging. Existing prompt copyright auditing methods struggle with long skills and introduce token-level perturbations that compromise readability. We propose SkillCODER for copyright auditing and buyer attribution in a black-box, content-only setting. During embedding, SkillCODER parses a skill into a section-aware intermediate representation, SkillIR, to identify structural anchors. It distributes conditional instructions across these anchors to form an Artifact-Gated Capsule, designed to activate under verification queries and remain dormant otherwise. Controlled-Vocabulary Error-Correcting Codes encode buyer-specific fingerprints through controlled lexical choices within the capsule. During verification, the defender issues paired positive and negative probes, extracts capsule evidence, and recovers ownership and buyer identity through error-correcting decoding. Across three domains, six models, and three agent frameworks, SkillCODER achieves 99.2% owner verification and 99.3% buyer Top-1 attribution accuracy while preserving task fidelity, substantially outperforming prompt-level baselines. Embedded signals remain robust under paraphrase, compression, clause deletion, and reorganization. Under five adaptive attacks, including a method-aware LLM sanitizer and multi-buyer collusion, ownership evidence remains reliably detectable while buyer attribution degrades gracefully.

0.58

Tracing MCP Security Vulnerabilities: Constructing and Analyzing an MCP Security Vulnerability Dataset

cs.CR ·2026-11-08 ·Xu He, Sagar Patel, Shu Wang
top venuedatasetcode/data
Abstract · 9 pages. Published at Data4SoftSec / IEEE SPW 2026. Post-publication dataset link: https://huggingface.co/datasets/XuViewer/mcp-vulnerabilit…

As a core protocol for integrating large language models (LLMs) with external data sources and tools, the Model Context Protocol (MCP) has quickly become a key interface in emerging AI agent systems. As the MCP ecosystem expands, reported vulnerabilities have also increased, creating a growing software-security attack surface. However, MCP vulnerability information remains scattered, noisy, and difficult to study systematically. This paper addresses that gap by constructing a curated MCP vulnerability dataset from CVE disclosures. Starting from 119 MCP-related records retrieved through February 2026, we filter unrelated cases out and obtain a validated snapshot of 24 MCP vulnerabilities. Our filtering methodology explicitly addresses acronym ambiguity and achieves 96.7% inter-rater agreement on a sampled exclusion audit while retaining all known high-severity MCP cases identified in public advisories. Using this initial dataset snapshot, we find that 87.5% of validated vulnerabilities are Critical or High severity, and that injection and path traversal vulnerabilities account for 66.7% of cases. We further identify recurring issues involving insufficient input validation, inadequate access control, and protocol-level security gaps. Through six representative case studies, we illustrate common exploitation patterns and map them to practical defenses. We release the dataset, reproducibility tooling, and a sustainable maintenance workflow as a foundation for future MCP security measurement, benchmarking, and defense research.

0.56

Delegating Computation with Approximate Arithmetic: Interactive Proofs for Wizards

cs.CR ·2026-11-08 ·Aritra Das, Debayan Gupta, Arup Mondal, Mihir More
Abstract

We study interactive proofs for arithmetic circuits over the complex numbers $\mathbb{C}$ in which both the prover and the verifier compute approximately (unlike muggles, wizards can use magic, i.e., floating-point numbers). The goal is to prove that a claimed output $y$ is close to the exact output $C(x)$: for thresholds $0<Δ_0<Δ$, the honest prover is accepted whenever $|y-C(x)|\leΔ_0$, and every prover is rejected with high probability whenever $|y-C(x)|>Δ$. Existing proof systems for real-valued data verify rounded finite-field computations. Bitan et al.~[EUROCRYPT~2026] gave an approximate sum-check protocol. We extend the protocol of Goldwasser, Kalai, and Rothblum (GKR) [STOC~2008] to public-coin interactive proofs for logspace-uniform layered circuits over $\mathbb{C}$ of size $S$, depth $L$, and $n$ inputs, under public bounds on gate values and wiring coefficients. At constant soundness error, the verifier runs in time $(n+1)\mathrm{poly}(L,\log S,b)$, the communication is $\mathrm{poly}(L,\log S,b)$, and the prover runs in time $\mathrm{poly}(S,b)$, where $b$ is the bit precision. To the best of our knowledge, these are the first doubly efficient interactive proofs in which both parties compute approximately. GKR reduces two claims about a layer to one on a random line, which leaves the unit circle, where our error bounds hold. We instead label gates by complex roots of unity and combine the two claims with a random root of unity as coefficient. We choose the error thresholds backward from the input layer, so that a large output error survives the verifier's rounding at every step. The honest prover needs only $O(S \log S)$ complex operations. We also obtain zero-knowledge proofs from one-way functions, one-round arguments from private information retrieval, and interactive PCPs whose proof length is polynomial in the witness length.

0.36

Treating the Model as an Untrusted Component: Deterministic Policy Enforcement, Human Approval and Taint Tracking Against Prompt Injection in Tool-Using Agents, with Validation on Open-Weights Models

cs.CR cs.AI ·2026-11-08 ·Mustafa S. Aljumaily, Nawar S. Alseelawi, Raghad Alshabandar
code/data
Abstract · 23 pages, 8 figures, 18 tables. Original research contribution. Code, data and the two-rater oracle audit: https://github.com/nawaralseelawi…

Tool-using LLM agents act on instructions embedded in the data they read, so prompt injection is a structural rather than a prompt-wording problem. We compare in-prompt defences (hardened prompts, spotlighting, LLM detectors) with deterministic enforcement outside the model (four reference monitors of increasing strength), human approval and taint tracking, judged by one effect-based oracle independent of every defence in an executable environment of 13 tasks and 12 tools, with argument-fuzzing attackers and severity-weighted harm. In an evaluation with real open-weights models (Qwen2.5-1.5B and Qwen2.5-7B; 156 attack episodes per defence and model), the models proposed the attacker's call in 64% and 84% of attack episodes behind the task-bound monitor, which admitted 1.3% and 7.7%; guardrail effects depended on the model (a prompt guardrail moved obedience from 67% to 65% and from 86% to 55%, spotlighting plus a detector to 12% and 0%) against static attacks. Simulation with an assumed model covers what no real run did: under the assumed erosion law a persistent adaptive attacker or a compromised model drives every guardrail stack to 100.0% attack success, whereas the task-bound monitor holds severe actions at 0.0%; its residual (harm 0.69/10) falls by 62% under taint-triggered human review, and a human layer without a deterministic monitor fails against approval spamming unless each episode has a deterministic approval budget. Implementation quality matters: a naive monitor admits 18 of 33 effective attack variants, a hardened one 6 and a capability-bound one 1. The real-model evidence is limited to two open-weights models and static attacks, the human-approval and taint results are simulation-only, and the larger model's detector stack contradicts the assumed erosion law. Code, data, a two-rater oracle audit and a notebook for repeating the real-model evaluation are provided.

0.36

Beyond AUROC: Deployment-Aware Evaluation of Model-Internal Jailbreak Detectors

cs.CR cs.CL ·2026-11-08 ·Mohammad Tehabsim, Basil Al-Housani, Nour Alhussien, Phung Lai
published
Abstract · Accepted to the 13th IEEE Special Session on Privacy and Security of Big Data (PSBD 2026), IEEE International Conference on Big Data (IEEE B…

Large language models (LLMs) are increasingly deployed in user-facing systems, where jailbreak attacks can bypass safety controls and induce harmful and policy-violating outputs. Recent model-internal detectors use hidden representations, gradients, or internal signals to identify such prompts, but their practical security value depends critically on behavior at low false-positive operating points rather than on area under the receiver operating characteristic curve (AUROC) alone. We study two data dependencies that shape this behavior: calibration data, which determine the detection threshold, and reference data, which can influence detector scores. Across four model-internal detectors, four LLMs, and twelve jailbreak families, thresholds calibrated on curated benign prompts frequently fail to preserve their false-positive rate (FPR) budgets on real-user traffic. Recalibration improves false-positive control but reveals substantial family-level coverage gaps, including cases with high AUROC but low true-positive rate (TPR). Causal reference-set interventions show that changing reference composition can substantially alter family-level detection and that improving coverage of one attack family can reduce coverage of others. These findings show that calibration and reference data are part of a detector's security behavior. We therefore recommend evaluating model-internal jailbreak defenses using realized FPR on deployment-like traffic and per-family TPR at explicit low-FPR operating points, with AUROC used as a complementary ranking metric rather than the primary security measure.

0.33

Attacking Slide Puzzle CAPTCHAs Based on Time and Space Generative Adversarial Networks

cs.CR ·2026-11-08 ·Rui Guo, Xuelian Xie, Botao Xu, Kesheng Liu, Wenxian Wang +1 more
dataset
Abstract

CAPTCHAs can protect websites from malicious attacks. As a typical representative of behavior CAPTCHAs, slide puzzle CAPTCHAs have become increasingly popular in recent years due to their user-friendliness and high security. Previous attack methods on slide puzzle CAPTCHA mainly face challenges in the following four aspects. (1) The defense mechanisms of previous slide puzzle CAPTCHAs were relatively simple. The current slide puzzle CAPTCHAs have added complex interferences, detected slide trajectory features in the two-dimensional space, and added other defense mechanisms. (2) The previous notch detection methods were difficult to accurately locate the notch positions. (3) Previous attack trajectories were limited to the horizontal dimension, making them impossible to pass the existing detection. (4) The previous attack trajectories were generated by fitting functions, which were relatively simple and could not well simulate the slide characteristics of human users. In response to the above challenges, firstly, we systematically analyze the defense strategies of six most popular slide puzzle CAPTCHAs: Capy, Dingxiang, Geetest, Kyger, Shumei and Tencent. Secondly, for each type of CAPTCHA, we construct a notch detection dataset and a slide trajectory dataset separately. Thirdly, we use advanced YOLOv11 to detect the notch positions and train generative adversarial networks (GAN) to generate attack trajectories. As far as we know, we are the first to generate attack trajectories by directly learning human slide trajectory characteristic. According to the experimental results, this method cracked six most popular slide puzzle CAPTCHAs with success rates of over 90\%, and the attack time is only about 5 seconds, indicating that the security of existing slide puzzle CAPTCHAs is facing challenge. At the same time, we propose various methods to enhance the security of slide puzzle CAPTCHAs.

0.27

Agentic Attribution of Cloud Resource Ownership

cs.CR cs.AI ·2026-11-08 ·Tiltan Gilat, Gal Engelberg, Ariel Zadok, Leon Goldberg
Abstract · 14 pages, 2 figures, 5 tables

Security tools raise findings across cloud environments, and AI agents now investigate and prioritize them. A finding is still resolved by people: it has to reach the people responsible for fixing it and accountable for the resource it concerns. Ownership is the organizational context that decides who that is. Where recorded, it is declared by hand in configuration management databases or owner tags and goes stale unless kept current; accounts are shared, and automation creates much of the infrastructure. The knowledge sits in work records, code history, audit logs and identity directories, encoded in each organization's own conventions. Recovering it poses four challenges: the signal is scattered across sources, none of them authoritative; the conventions that carry it differ between organizations; reaching an owner takes multi-hop links; and those related to a resource differ in the strength of their relation. We define ownership as a set of entities, people or teams, each related to the resource by one of five patterns ordered by strength. We present a reference architecture for agentic ownership attribution that derives ownership from these records instead of declaring it: it profiles an organization offline into tested signal procedures and an ownership playbook, then attributes each resource with a coordinator agent and parallel investigators, with confidence computed by rules. On 100 resources of 50 Amazon Web Services (AWS) resource types in one organization, we pool the architecture's candidates with those of an unprimed general-purpose agent and verify every evidence piece against the records. The architecture reaches 92.1% weighted recall and 86.6% weighted precision, against 54.8% and 77.4%, at 0.55 times the agent's cost per unit of coverage. The difference lies in people who build and change resources without an explicit owner, found through work records and code history.

0.26

ErosionBench: Probing Alignment Decay in Enterprise Multi-Agent Systems

cs.CR cs.AIcs.MA ·2026-11-08 ·Riya Bharathwaj, Akshara Prabhakar, Roshan Ram, Avni Kothari, Sarah Tan +2 more
benchmark
Abstract · 10 pages plus references and appendices

As enterprises delegate consequential work to multi-agent LLM systems, benign-looking requests can escalate into unsafe actions as they propagate across sessions through poisoned shared state, long-context documents, and trusted peer messages: no single message is malicious, yet the assembled tool action is. Prior benchmarks isolate a single agent under prompt injection or score multi-agent attacks by their final output, leaving cross-session erosion of shared state out of scope. We introduce ErosionBench, an environment-grounded benchmark that executes such attacks and verifies alignment decay at the process level, scoring the realized tool call against authorized intent and provenance with deterministic contracts, not output text. Each scenario is authored as a deterministic state graph before any prose is rendered, so every state change traces to a known cause. It spans four enterprise domains and four decay mechanisms across 1,728 scenarios with matched benign controls, labeled against the OWASP Top-10 for LLM Applications, plus 198 image-based attacks, each paired with a text equivalent, so a defense must survive the same attack in either channel. Perimeter defenses suppress attacks only by over-refusing 54-92% of legitimate work, so a low attack-success rate can conceal severe over-refusal. As a reference defense, we propose a calibrated Meta-Auditor pairing synchronous state checks with asynchronous trajectory monitoring; on a family-disjoint 906-scenario held-out split it reaches 98.5% recall at a 2.9% false-positive rate. Since it monitors trajectory, not surface content, its detection is channel-invariant: the strongest auditor flags image-delivered attacks about as reliably as their text equivalents (recall gap $\leq$2 points per model). Across GPT, Gemini, and Claude auditors, calibration keeps false positives low while recall tracks model capability, strongest for Claude.

0.22

Do Benign-Trained Hidden-State Probes Detect Tool Errors Under Adversarial Conditions?

cs.CR cs.AIcs.CL ·2026-11-08 ·Bibek Bhandari, Kshitij Lingthep
Abstract

Although hidden-state probes efficiently monitor model operation, it remains unclear whether probes trained on benign tool use remain useful for detecting incorrect actions under adversarial conditions. In this research, we assess this problem from the perspective of operational tool-action correct- ness. we use four AgentDojo suites, four conditions, and three models: Qwen2.5-7B-Instruct, Qwen2.5-14B-Instruct, and GLM- 4.6V-Flash (9B). We perform a five-fold, task-grouped out-of-fold (OOF) evaluation for each model. We train step-specific logistic- regression probes only on benign and non-defended trajectories and evaluate them on held-out trajectories for benign, prompt- injection, system-override, and tool-poisoning conditions. We compare our hidden-state probe to mean tool-call-token entropy, negative mean tool-call-token log-probability, and self-consistency uncertainty. The paired tests highlight six significant differences after Holm correction. All of the tests involve Qwen2.5-14B- Instruct under the system-override condition, with all three scalar uncertainty metrics obtaining higher area under the receiver operating characteristic curve (AUROC) and average precision (AP) than the hidden-state probe. These results suggest that benign-trained hidden-state probes should not be assumed to reliably detect operational tool-action errors under adversarial conditions.

0.19

PRISM: Causal Counterfactual Minimization of Agent Tool Privileges via Information Bottleneck Role Induction

cs.CR cs.AIcs.LG ·2026-11-08 ·Yang Luo, Qingni Shen, Tairui Wang, Jiajun Li, Zhonghai Wu
Abstract · 13 pages, 3 figures

This paper shifts the object of role engineering from human users to large language model (LLM) agents, and proposes PRISM to address the over-authorization that arises when an agent obtains full tool privileges through the Model Context Protocol (MCP). PRISM formulates whether a tool is necessary for a task as an intervenable causal proposition, and applies structured interventions in a sandbox to learn a counterfactual necessity model. It then compresses trajectories into discrete latent variables through an information bottleneck, and determines the number of roles from the knee of the rate-distortion curve. Finally, it formalizes pruning as a bilevel Stackelberg game, in which a generative flow network samples diverse near-optimal solutions. On three public benchmarks, AgentDojo, $τ$-bench and ToolEmu, PRISM compresses the number of authorized tools per role by 78.3%, attains a task success rate of 78.2%, reaches a counterfactual necessity accuracy of 88.7%, and lowers the blast radius to 15.9%.

0.18

On the Relative Security of Prompt Injection: Mechanised Definitions, Defences and Verdicts for LLM Agents

cs.CR cs.AI ·2026-11-08 ·Brijesh Dongol
theory
Abstract

An LLM agent that reads untrusted content, such as an email or a web page, may follow instructions hidden in it. Noninterference is too strong for such agents, since a useful agent lets untrusted content influence what it does and may disclose the user's data by design, and declassification needs a release policy written for each task. We instead state security relative to a reference agent, which runs the same plan under the same runtime monitor with a model that never obeys untrusted content. An agent is relatively secure if every leak it has is a leak the reference already has. Our framework, PlanTERS, instantiates relative security, a notion from speculative execution, with a small agent language in which the attacker chooses how the model follows instructions, a composition with runtime monitors, and attacker models that differ in what they observe. A fence theorem shows that revoking injectable tools at the first untrusted read, and never granting them back, makes every plan relatively secure at the event level on finite runs, for every model behaviour in which each injection is a single call and the model obeys on a schedule fixed in advance; with a fairness condition, this extends to possibly infinite runs. We prove or refute relative security for agents with fences, budgets, approvals, loops and dual LLMs, in Isabelle/HOL.

0.16

Least-Privilege Auditing of Agent Tool Calls with Intent Graphs and Reasoning-Guided Escalation

cs.CR cs.AI ·2026-11-08 ·Doron Shavit
Abstract

LLM agents that call tools are exposed to indirect prompt injection: instructions hidden in tool outputs can steer an agent into calls the user never asked for. We present a runtime audit that derives an executable intent graph from the user's objective and the deployment's tool catalog with one LLM call per task, and checks every tool call against it with a deterministic loop that makes no LLM call (median 5.7 microseconds per call). Two further mechanisms recover utility. Lookups of the user's own state bypass the audit, because the harm of an attack lies in its write or outward step. The calls that remain flagged are escalated to a reasoning guardian, in the spirit of tool-integrated reasoning models such as START: a reasoning model that sees only the objective, the plan, the call history without tool outputs, and the flagged call, and can verify facts with a restricted Python-expression tool. A current agent showed no baseline attack success, so end-to-end validation remains open.

0.16

Robust and Private Swarm Learning via Zero-Knowledge Validator Admission and Consensus Aggregation

cs.CR cs.AIcs.LG ·2026-11-08 ·Marco Arazzi, Mert Cihangiroglu, Antonino Nocera
Abstract

Swarm Learning improves resilience and removes the need for a single trusted coordinator by distributing model aggregation and validation across participating nodes. However, it also delegates the validation of model updates to peers whose private data may be biased, unrepresentative, or adversarial, and existing Swarm Learning protocols generally do not attest that data, leaving decentralized validation as an attack surface. We present a robust, privacy-preserving Swarm Learning framework that makes verifiable data quality the condition for admission to the learning task. A validator joins only after a zero-knowledge attestation proves, without revealing any raw sample, that a reference model ran correctly on its private data. The attested outputs also yield a label-free class profile, so the network assembles a collectively IID validator pool from individually skewed validators as accurately as an oracle with the true labels (total-variation distance to uniform 0.025, against 0.021 for the oracle and 0.21 for random selection). Admitted validators score client updates and, through a Tendermint-style Byzantine Fault Tolerant consensus, agree on a gradient mask and a quality-derived aggregation weight for each submission, replacing dataset-size weighting with a quality-aware signal. Adapting the cut-layer idea of Split Learning, a challenge-based audit lets peers re-verify that a mask was computed on the attested data without an expensive end-to-end zero-knowledge proof. Across four benchmarks under non-IID partitions the framework matches FedAvg on clean data; under a gradient-scaling attack that collapses FedAvg to random chance, it retains 94% of clean accuracy on CIFAR-100 and over 99% on FashionMNIST, where FedAvg achieves only 2.1% and 11%. The result is a private, robust, and auditable foundation for decentralized learning.

0.15

Attack Injection to Vehicle Reactions: Extensible Closed-Loop Adversarial Evaluation Framework

cs.CR cs.NI ·2026-11-08 ·Mohamed Gaber, Tahsin Wahid, Mohammad Mamun, Salimur Choudhury
Abstract · 6 pages, 3 figures, 5 tables

Autonomous vehicles can use information exchanged with other vehicles and roadside infrastructure to support driving decisions. However, applications that rely on this information can make unsafe decisions when it is falsified. Studying the resulting traffic effects requires following received information through application decisions, vehicle reactions, and subsequent interactions. We present an extensible closed-loop adversarial evaluation framework built on VaN3Twin, a platform for traffic and wireless network simulation, with a diverse attack library and a labelled dataset covering benign and attack conditions. A key advantage of the framework is that it can be extended with additional attacks, message paths, receiving applications, and logging fields. The added library contains 76 configurable attack profiles adapted to message formats defined by the European Telecommunications Standards Institute. These profiles are organized into nine groups of related attacks, referred to as suites. The framework connects live attack injection and simulated network delivery to receiving applications, whose selected commands update vehicle movement. The resulting state feeds subsequent message generation, closing the loop. The dataset links physical vehicle states and message exchanges to attack labels, application decisions, selected commands, and traffic outcomes. Evaluation with individual and combined attack suites confirmed execution of all profiles, and every suite produced requests linked to attacked messages. A recorded sequence also showed a vehicle reacting to an attacked message and subsequently influencing another vehicle through an unmodified message describing its updated movement. These findings demonstrate how live injection and vehicle feedback support the study of direct and indirect effects as traffic evolves.

0.11

Coding Agents Propagate Vulnerabilities in Code Bases

cs.CR ·2026-11-08 ·Chenhao Sun, Niels Mündler-Sasahara, Mark Vero, Martin Vechev
Abstract

LLM-based coding agents are widely used to develop critical production software. Even though coding agents are known to occasionally write vulnerable code, there has not been any investigation into how code base vulnerabilities evolve over time under coding agent contributions. In this work, we close this gap by investigating whether and how vulnerabilities in existing code lead coding agents to introduce similar vulnerabilities in the code they write. We first evaluate coding agents on 96 tasks that require implementing a new backend application in a repository where other applications already exist. Second, we investigate repository-level code generation on 20 pairs of tasks in real C/C++ projects. In each task, the agent is asked to complete code critical to memory safety. We find that in both settings, insecure existing code leads to agents writing vulnerable code more often. In backend generation, insecure neighbors reduce the share of correct and secure implementations by 3.4% on average, and in C/C++ repository-level coding, the average gap between secure and insecure contexts reaches 35.4%. Prompting to follow existing code-style or to be careful about security increases, respectively decreases, how strongly agents propagate vulnerabilities, but neither stops them. When asked to explain their implementation choices, coding agents attribute them to following another code site in 76.5% of cases, even though they recognize the introduced vulnerability in 69.9% of cases. These findings show that agents use existing vulnerabilities as templates for new ones indiscriminatorily, turning each contribution into an opportunity to introduce more vulnerabilities. As such, as agents write a growing share of production code under decreasing human review, code bases can increasingly degrade in security if security vulnerabilities are not thoroughly fixed continuously.

0.11

RoCoMark: Robust Code Watermarking Against LLM-Based Functionality-Preserving Rewriting Attacks

cs.CR ·2026-11-08 ·Hongjie Zhang, Xingcheng Li, Hanzhou Wu, Zhihua Xia, Xinpeng Zhang
Abstract · 17 pages (minor update)

Large Language Models (LLMs) present significant challenges to source tracing of generated code, as LLM rewriting can substantially alter the syntactic and structural characteristics of watermarked code. However, existing watermarking methods often suffer severe performance degradation under LLM-based functionality-preserving rewriting attacks. We attribute this vulnerability not merely to the intensity of rewriting, but to a fundamental distribution mismatch between the perturbations encountered during watermark training and the global, non-deterministic rewriting behaviors induced by LLMs. In order to deal with this challenge, we propose RoCoMark, a Robust Code waterMarking framework that explicitly learns and adapts to LLM rewriting behaviors. RoCoMark is a multi-round minimax game between the watermark embedder and LLM rewriter. In each round, the embedder first learns threat profiling from LLM-rewritten code by optimizing the watermark encoder through contrastive learning and knowledge distillation while mitigating catastrophic representation drift. Then the embedder improves strategies by optimizing the watermark encoder and extractor simultaneously. Finally, an adaptive attack-evolution mechanism based on Reflexion and short- and long-term memory pools is applied, enabling continuous interaction between the watermark embedder and evolving rewriting strategies. This iterative attacker-defender process progressively improves the robustness of the watermark against diverse and previously unseen LLM rewriting behaviors. Extensive experiments on four typical benchmarks, under attacks generated by five LLMs, show the effectiveness of RoCoMark. RoCoMark achieves an average detection AUROC of 74.2% and a Rewrite BitACC of 75.3%, while maintaining a code functionality pass rate above 99.7%.

0.10

Systems & Networking 23

Samyama-SDG: A Declarative GraphQL @count and Regex/Ontology Synthetic Data Compiler with $O(1)$-Memory Referential Integrity and Causal Ground Truth

cs.DB ·2026-11-08 ·Madhulatha Mandarapu, Sandeep Kunkunuru
code/datahot direction
Abstract · 8 pages, 3 figures, 4 tables. v2 adds an end-to-end differentially private fit-generate-audit example on a public table (Section 4) and posi…

Synthetic data generators face a divide between declarative relational mockets and domain causal simulators. Declarative generators specify entities, regexes, and foreign keys via schemas, but buffer parent primary-key pools in RAM ($O(N)$ memory) and lack ontology bindings or causal ground truth. Conversely, domain simulators emit realistic streams and ground-truth labels, but require thousands of lines of procedural code per domain. We present Samyama-SDG (sdg-rs), a Rust synthetic data compiler unifying declarative schema compilation with ontology-bound causal simulation. At Layer 1, Samyama-SDG compiles Cardinality-Extended GraphQL schemas (@count, @pk, @ontology, @regex, @dist, @fk, @after) and an Attribute Data Dictionary into a topologically sorted DAG. Combining stateless counter-based seekable PRNGs (SplitMix64 over $(\text{seed}, \text{entity}, \text{row}, \text{field})$) with index-bijective primary keys $f_{\text{PK}}(s, E, i)$ lets child tables resolve uniform or Zipfian foreign keys and multi-hop @after constraints in $O(1)$ memory without buffering parent tables. At Layer 2, causal processes layer fault propagation and machine-scorable ground truth (ground_truth.jsonl) over the same seekable PRNG and samyama-ontologies binder. Across three schemas, Samyama-SDG generates $1{,}000{,}000$ records in $128.79\text{ ms}$--$172.43\text{ ms}$ ($5.80\text{M}$--$7.76\text{M}$ rows/s) and $5{,}000{,}000$ records in $676.12\text{ ms}$--$808.56\text{ ms}$ at flat $31.4\text{ MiB}$--$35.1\text{ MiB}$ memory, with $0$ conformance errors across $1{,}150{,}000$ foreign keys and $900{,}000$ temporal constraints.

0.77

Edge Computing for Computer Vision in IoT: Feasibility and Directions

cs.NI cs.CVcs.DC ·2026-11-08 ·Panagiotis Savvidis, George A. Papakostas
top venuehighly cited
Abstract

The convergence of decentralized architectures integrating Machine Learning, Computer Vision and Low Power Wide Area Networks is increasingly becoming an integral part of our daily existence. Internet of Things serves as a real-time data conduit enhancing decision making via embedded technology and continuous data exchange. This paper explores the feasibility of Edge Computing as a foundational pillar in this evolving landscape. We experiment under real world, dynamic conditions, evaluate the technological aspects, strategies, process flows and key observations under the broad Edge Computing domain. Research pathways include Multi-access Edge topologies in future 6G networks, model quantization, and satellite-enhanced communication platforms. Additionally, a discussion is added supporting the advanced AI functionalities, including zero-shot learning, multi modal perception, and decentralized generative AI, thereby broadening the scope of intelligent applications across various domains. The significance and research objective of this study are threefold: (1) evaluation of LoRaWAN and satellite IoT communication strategies, (2) analysis of CV workloads on edge hardware and (3) future research directions where Edge Computing can support low-latency, energy-efficient and socially impactful IoT applications. By explicitly addressing these aspects, we aim to establish a clear link between the technological feasibility, ultimately with a practical and socioeconomic relevance.

0.54

Multi-Granularity Communication-Computing Resource Mapping for User Association in MEO-Assisted LEO Satellite Networks

cs.NI ·2026-11-08 ·Yi Jing, Chunxiao Jiang, Jiawei Wang, Jiachen Sun, Yafeng Zhan
Abstract

Low-Earth-orbit (LEO) satellite networks are evolving into spaceborne communication-computing infrastructures. Their user association must jointly account for the radio link and the CPU, GPU, NPU, memory, cache, storage, energy, queue, and thermal states that determine executable service capacity. This paper studies a MEO-assisted LEO architecture in which medium-Earth-orbit (MEO) regional state aggregators collect and relay time-stamped telemetry, the network operation control center (NOCC) maintains the global logical resource view and makes the final association decisions, and LEO satellites provide access and onboard execution. We develop a multigranularity representation and service-primitive-driven mapping that converts a service-level request into a target-LEO-dependent quantized resource vector. The integer formulation, exact mixed-integer linear programming (MILP) reference, and a capacity-aware primal-dual algorithm share the same mapped objective. Experiments use real-world service bandwidth traces, a 1584-satellite LEO shell, multiple traffic seeds, controlled load and mapping error, delayed telemetry, and temporal sessions. Relative to static mapping, the proposed method improves realized utility by 10.64% and lowers end-to-end outage by 5.70 percentage points. Target quantization reduces signaling by 19.84% with negligible utility loss. Additional load, delay, temporal, runtime, and exact-MILP audits delimit the operating region and remaining mapping uncertainty.

0.46

ICLayout-Bench: Benchmarking LLM Agents for Custom Circuit Layout Design

cs.AR cs.AIcs.SE ·2026-11-08 ·Zhangmai Li, Tianrui Ma, Zheng Weng, Mo Zou, Zidong Du +6 more
benchmarkcode/data
Abstract · 12 pages, 7 figures, 3 tables, including appendices. Project page: https://lizhangmai.github.io/ICLayout-Bench Code: https://github.com/lizh…

Large language model (LLM) agents are being explored for circuit design, but their ability to produce high-quality custom layouts has not been extensively evaluated. ICLayout-Bench is a benchmark in which agents use electronic design automation (EDA) tools to construct and optimize GDSII layouts from transistor-level circuit netlists. These layouts are evaluated using physical verification, parasitic extraction, and post-layout simulation. Layouts that pass the physical and functional checks are scored on electrical performance and area. The benchmark includes 165 tasks across 13 process design kits (PDKs), covering technology nodes from 3 nm to 180 nm. The tasks span analog, mixed-signal, and digital circuits in open-source and commercial EDA flows. The experiments evaluate 22 configurations spanning seven models on 21 public and 19 private tasks. Mean layout scores range from 4.22 to 80.26 on public tasks and from 1.43 to 67.01 on private tasks, on a 100-point scale. Higher reasoning effort does not consistently improve mean scores, and configurations with comparable scores can incur substantially different model costs.

0.43

Verifiable, Privacy-Preserving and Collaborative Multi-Objective Optimization with NSGA-II on Blockchain

cs.DC ·2026-11-08 ·Goshgar Can Ismayilov
Abstract

Collaborative multi-objective optimization allows independent problem solvers to contribute their solutions to a shared problem. However, their validity cannot be reliably verified in open settings if the parties are reluctant to disclose their solutions. Therefore, this paper proposes a novel zk-NSGA-II protocol for verifiable, privacy-preserving and collaborative multi-objective optimization on blockchain. To the best of our knowledge, this is the first work that executes Non-Dominated Sorting Genetic Algorithm II (NSGA-II) on the RISC Zero Zero-Knowledge Virtual Machine (zkVM) where evolutionary generations are certified by succinct proofs that are verified on-chain. Problem solvers maintain their own independent populations as the set of solutions in the decision space and their corresponding objective vectors in the objective space. They contribute to a global on-chain Pareto front by aggregating their local off-chain Pareto fronts according to the rule of non-domination which admits a contribution if it is non-dominated with respect to the existing contributions in the global front. To prevent fabricated contributions, population states across generations are linked through hash chains. The correctness of the zk-NSGA-II protocol is first evaluated on twelve problems from the ZDT and DTLZ benchmarks. Second, its feasibility is shown in a real-world portfolio optimization case study with the Hang Seng dataset. Performance on these problems is measured with nine different metrics across three complementary perspectives for optimization, zero-knowledge proof and blockchain. The empirical findings indicate that trustless collaborative multi-objective optimization is feasible with the existing zero-knowledge and blockchain technologies.

0.36

OOM: Offload-Optimized Full-Parameter MoE Training via Memory-Execution Co-Design

cs.DC cs.LG ·2026-11-08 ·Ruijia Yang, Shiyuan Lin, Chengxi Liao, Zeyi Wen
code/data
Abstract · 19 pages, 9 figures. Code: https://github.com/RegiaYoung/OOM-MoE

Offloading model state to host memory and streaming layer weights let a single multi-GPU node train large mixture-of-experts (MoE) models. GPU memory then becomes a performance budget, shared by resident expert state, retained activations, and execution worksets. The best allocation depends on the expert execution path, batch size, and sequence length. We present OOM (Offload-Optimized MoE), a full-parameter MoE training system that co-designs expert execution and memory allocation. For expert-parallel execution, OOM updates resident experts on their owning GPUs and jointly plans state residency and MoE activation retention according to pipeline dependencies. For local execution, OOM bounds GPU worksets through dependency-aware partitioning; in hybrid models, it also retains attention results on the host to avoid costly recomputation and selectively regenerates layer boundaries through cheaper layers to make room. Calibrated cost models guide joint configuration within GPU and host budgets to minimize step time; OOM selects among the configured paths through short executions. On four H100s, OOM trains with 1M-token contexts, whereas FSDP2, ZeRO-Offload, and MegaTrain exhaust GPU memory at 512K. It also trains a 122B-parameter MoE model, on which all three exhaust host memory. On four 24 GB RTX 4090s, it trains 512K-token contexts and 1,024 1K-token sequences per step without gradient accumulation. Across matched-global-batch comparisons, OOM delivers 1.58x and 4.08x geometric-mean (up to 2.14x and 7.77x) throughput over CPU-offloaded FSDP2 on the two platforms while using the least host memory among the evaluated systems -- as little as 1/3.0 of FSDP2's.

0.34

A Per-Step Load-Balancing Engine for Expert-Parallel MoE Serving

cs.AR cs.DC ·2026-11-08 ·Jingyuan Li, Maolin Wang, Hayden So
code/data
Abstract · An anonymous version of this manuscript was submitted to HPCA 2027 on August 1, 2026. The original submission PDF and its SHA-256 checksum a…

In expert-parallel Mixture-of-Experts serving, a layer finishes only when the most loaded GPU of its group does. Token routing shifts at every decode step and concentrates on a few experts, so most of the group's reserved GPU time waits on stragglers. Recovering that time means re-deciding, at every layer of every decode step, how each replicated expert's tokens divide between its copies. The exact linear-program solvers that make this decision optimally take longer than many of the layers they serve, and their working set outgrows GPU shared memory at deployment scale. We introduce Relief, a bidirectional pairwise relaxation that makes the linear program unnecessary, coming within a fraction of a point of the per-step optimum in a few passes of integer halving over the token counts. We build and measure three deployments, CUDA kernels on the serving GPU, the same verified engine on an FPGA over PCIe, and a fixed-function block beside the accelerator. On decode routing from three open MoE families, Relief holds the balance near the per-step optimum at every group size and layer time measured. Scored as effective utilization relative to that optimum, the kernel alone matches or exceeds every deployable scheme at every group size, from the hundred-microsecond layer on. The on-die engine widens that lead at every point, recovering up to 37.3 percentage points over the deployed static placement, while adding 0.02% of a current accelerator's transistors.

0.33

Practical Deep Learning-Based Block Demapping for mmWave MIMO Systems

cs.AR ·2026-11-08 ·Syed Asrar ul haq, Jesus O. Lacruz, Joerg Widmer, Sumit J. Darak
published
Abstract · Submitted for publication to IEEE Transactions on Wireless Communications

Deep learning is expected to play a central role in 6G systems, by enabling data-driven physical layer designs that can adapt to practical wireless impairments. This is particularly important for millimetre-wave (mmWave) multiple-input multiple-output (MIMO) systems, which are strongly affected by residual channel distortions, interference, and hardware non-idealities that are difficult to capture with analytical models. In this paper, we propose a deep learning-based soft demapper that performs block-wise processing of received symbols instead of conventional symbol-wise detection. The demapper employs a one-dimensional convolutional neural network to learn temporal structure induced by practical propagation and hardware effects, while producing soft bit outputs compatible with soft-decision decoding stages. The proposed approach is trained and evaluated for a 4$\times$4 wideband mmWave MIMO system at 60~GHz, using an over-the-air dataset collected across multiple days and settings. Experimental results demonstrate consistent improvements over a conventional soft-decision demapper, including up to a 2.6$\times$ reduction in bit error rate for 16QAM and up to a 17.6$\times$ reduction for 64QAM under real over-the-air mmWave MIMO conditions. The proposed demapper is further implemented on a Zynq RFSoC platform using a reconfigurable convolution engine and fixed-point arithmetic, achieving an approximately 34.3$\times$ speedup compared to a processor-based implementation, and demonstrating that learning-based demapping can be deployed efficiently on resource-constrained edge hardware.

0.32

Priority-Aware Deficit Round Robin for AXI4 Bandwidth Fairness in ASAP-7nm Process

cs.AR ·2026-11-08 ·Tasmin Khan, Tishna Sabrina, Ch. Md. Rakin Haider
benchmark
Abstract · 6 pages, 5 figures

Modern systems-on-chip place real-time clients, processor cores and bulk data engines on the same AXI4 interconnect, where they contend for shared slaves. AXI4 fixes the ordering rules and carries a 4-bit quality of service (QoS) field, but leaves the service policy to the implementer, and that policy determines how long each master waits and how much bandwidth it receives once a slave saturates. We propose a QoS-aware deficit round robin (DRR) architecture that charges each grant in beats, and benchmark it against two architectures built behind the same port list: round robin, and QoS priority with anti-starvation aging. Charging in beats recovers bandwidth fairness while leaving every burst whole, so the address phase is issued once and locality for memory access is preserved. All three are driven by the same deterministic traffic generator and taken from RTL to GDSII on the ASAP7 7-nm predictive PDK. Under a 1:16 burst-length mismatch, the round robin and QoS architectures both hold the beat ratio between the two competing masters at 1:16, while our design brings it to 1:1 at full load. Mean read latency for the priority master falls from 95 cycles under round robin to 67 with QoS and 37 with QoS-aware DRR. At 500 MHz this serves the priority master in 74 ns against 190 ns. The measurements give a per-policy price and a traffic-based guide for choosing among the three.

0.29

A Simple Phase-King Protocol with Predictions

cs.DC ·2026-11-08 ·Antonia Giannoulea, George Papadoulis, Nikos Leonardos, Aris Pagourtzis
Abstract

Inspired by recent work (Ben-David, Dzulfikar, Ellen, and Gilbert, $\textit{Byzantine Agreement with Predictions}$, PODC 2025) we study synchronous binary Byzantine Agreement in the learning-augmented setting, where each process receives predictions about which processes are honest or Byzantine. We present a deterministic Phase King protocol for the anauthenticated setting, combining early stopping with a specifically designed prediction-aware king selection procedure. For at most $t<n/3$ Byzantine processes, the protocol handles arbitrarily many prediction errors, improving upon the $O(n^{3/2})$ bound of Ben-David et al., and terminates within $O(P)$ communication rounds, where $P=1+\min\{\mathbf{B}/n,f\}$, $\mathbf{B}$ is the total number of incorrect prediction bits held by honest processes, and $f$ is the actual number of Byzantine processes. Our protocol uses $O(n^2P)$ messages and $O(n^2P\log n)$ bits, avoiding the initial cubic bit cost of exchanging complete prediction vectors. In comparison to very recent independent work (Dufay, Dzulfikar, and Gilbert, $\textit{Predictions Can Only Help! Communication Efficient Byzantine Agreement with Predictions}$, DISC 2026), which achieves the same error tolerance and round complexity but better communication bounds, our construction provides a much simpler Phase King approach that completely avoids involved preprocessing constructions, rendering it more practical, especially for achieving agreement in time-evolving networks. We believe that the novel prediction-aware king selection technique of our scheme is of independent interest and can find further applications in learning-augmented distributed consensus.

0.29

A Decentralized Quantum-Resistant Notary Scheme for Blockchain Interoperability with a Novel Dynamic Trust Model

cs.DC ·2026-11-08 ·Koosha Esmaeilzadeh Khorasani, Saeed Moradi, Sara Rouhani
published
Abstract · This paper has been accepted at the IEEE Blockchain Conference 2026 (IEEE Cybermatics 2026)

Blockchain interoperability is essential for enabling secure and reliable interactions across heterogeneous networks. However, existing notary schemes face two key issues: quantum vulnerability and rigid trust management. To address the first issue, this paper proposes a decentralized, quantum-resistant framework integrating post-quantum primitives ML-KEM and ML-DSA alongside AES-256, optimizing performance by using asymmetric cryptography for authentication and non-repudiation, and delegating bulk encryption to symmetric schemes. Under parallel loads of up to 270 transactions per second, this hybrid design reduces the end-to-end performance gap between the post-quantum and elliptic-curve approaches to under 6% on average, converging to within 0.2% at peak load. The second issue is handled by a dynamic trust model that uses Popoviciu's inequality to derive an adaptive discount factor $α$ based on variance in a Verifier's behavior, removing the need for fixed thresholds or multi-stage configuration, while a tunable dynamicity factor $k$ enables control over system sensitivity. We evaluated the trust model over 200 epochs under adversarial conditions, where the proposed model detects and responds to misbehavior that the baseline model fails to penalize adequately. These results demonstrate that quantum-resistant blockchain interoperability is practically achievable, with the proposed trust mechanism offering greater adaptability and resilience than existing approaches.

0.28

Confidence-driven adaptive semantic transmission for task-oriented communication

cs.NI cs.CV ·2026-11-08 ·Suman Das, Prajnyajit Mohanty, Siddhartha S. Borkotoky
Abstract

Image-based remote monitoring over wireless networks faces a trade-off between communication cost and classification reliability. Transmitting complete images is communication-intensive, whereas relying solely on lightweight edge classifiers may compromise accuracy. To address this challenge, we propose Confidence-Driven Adaptation (CDA) for image transmission and classification. An edge device (ED) captures images and applies a lightweight classifier before transmission to a remote server. Based on its classification confidence, the ED selects among three transmission modes with varying levels of information: a short text string reporting its predicted class when confidence is high, a compressed Region of Interest (ROI) at intermediate confidence levels, or an uncompressed ROI when confidence is low. The server accepts the predicted class directly in the first case and performs classification using a more powerful classifier in the other two cases. Numerical evaluations under bandwidth-limited, lossy links and latency-constrained settings demonstrate that CDA achieves high task accuracy at the receiver compared with non-adaptive baselines that transmit only the edge-classifier outcome or the uncompressed ROI.

0.24

Federated learning for remote IoT over LPWAN: Preliminary insights from structural health monitoring

cs.NI ·2026-11-08 ·Lakshmiprasanna Bandaru, Anshika Singh, Tarutal Ghosh Mondal, T. Jothi Saravanan, Siddhartha S. Borkotoky
Abstract

Realizing federated learning (FL) in remote regions lacking reliable cellular infrastructure is challenging, as it requires dependable communication among participating devices. Low-power wide-area networks (LPWANs) offer an attractive communication solution for such scenarios. However, their limited data rates necessitate careful investigation of the communication-learning trade-offs and mechanisms to reduce communication overhead. This paper presents an investigation of the Long Range (LoRa) LPWAN technology for FL-based structural health monitoring (SHM), a representative remote IoT application. We consider a SCAFFOLD-based FL framework for multi-class anomaly detection and analyze the latency-accuracy trade-off. We also examine strategies for allocating the LoRa spreading factor (SF) during model-update exchange. Numerical results show that, with appropriate SF allocation, LoRa-supported FL-SHM can achieve competitive learning accuracy. In particular, jointly accounting for link distances and clients' dataset sizes when selecting the SF provides a favorable balance between convergence performance and communication latency. These preliminary findings offer useful insights into the design of FL systems over LPWANs for remote IoT deployments.

0.24

PRISM: Preference-based Reward and Uncertainty-Integrated Synthesis Minimization for RTL PPA Optimization

cs.AR cs.SE ·2026-11-08 ·Armin Abdollahi, Negin Ashrafi, Mehdi Kamal, Massoud Pedram
Abstract

This paper introduces a budget-aware, multi-objective framework (called PRISM) for RTL optimization that improves power, performance, and area (PPA) while sharply limiting expensive commercial synthesis tool calls. Starting from a synthesizable baseline and functional testbench, the method iterates through LLM-driven rewrite generation, fast compile/simulation equivalence filtering, low-cost proxy feature extraction, and selective commercial synthesis tool (Synopsys Design Compiler (DC)) evaluation. To prioritize which candidates deserve high-quality synthesis, we train a preference-based reward model using pairwise ranking supervision derived from commercial synthesis tool outcomes, and combine ensemble uncertainty with a diversity-aware acquisition rule to avoid early bias and redundant evaluations. A key advantage is that PRISM maintains a best-tradeoff set G of non-dominated designs throughout the run: G preserves the best solutions found so far, is fed back into generation and refinement, and enables consistent, cumulative improvements rather than repeatedly restarting from the baseline. Across RTLLM v2.0 modules, the approach consistently yields post-synthesis gains under fixed commercial synthesis tool call budgets, and ablations confirm that reward-guided selection and candidate dropping are central to sample-efficient progress. PRISM efficacy is evaluated across multiple technology nodes (7 nm, 15 nm, and 45 nm) using identical pipelines without retraining.

0.22

When Is High-Bandwidth Flash Cost-Effective for LLM Inference?

cs.AR ·2026-11-08 ·Zhiyu An, Yiping Li, Wan Du
Abstract

Can high-bandwidth flash (HBF) make LLM inference cheaper than adding GPUs? By storing weights closer to the GPU than SSD, HBF may reduce the number of GPUs needed for model capacity. However, this integration is only justified if it meets service targets and its cost is recovered. We simulate this integration using public component data across nine dense and mixture-of-experts (MoE) models, five workloads, two GPU platforms, and 1--100 GPUs. At fixed demand, HBF reduces the minimum GPU count to meet a service target in 98 of 468 comparisons; majority of them are paused and short sessions, 63 remain cost-effective at half a GPU price per HBF stack. For single short-chat users, HBF enables large MoEs, e.g. DeepSeek V3 and Kimi K2, to meet a relaxed service target on one RTX~4090 GPU, where conventional serving fails. At fixed GPU count, HBF improves throughput in 48\% of matched comparisons below 16 GPUs and 13\% at larger counts. Even at 100 RTX~4090-scale GPUs, cascaded HBF raises offline throughput by 16.4--26.5$\times$ for three large MoEs. We found savings and throughput gains arise in different scenarios.

0.22

CU-3D: A Scalable Legality-Aware RTL-to-GDS Flow for 3D-IC Physical Design

cs.AR ·2026-11-08 ·Mingjun Wang, Guanqi Li, Yihan Wen, Lancheng Zou, Mingjun Li +8 more
Abstract

Face-to-face (F2F) hybrid bonds shorten interdie connections, but data, clock, and power-grid (PG) connections must be realized together. CU-3D is a RTL-to-GDS physical design flow based on commercial EDA tools. It derives role-compatible hybrid-bond (HB) sites from a 17-metal ASAP7-based technology model, complete HB footprints, and constructed PG geometry. To coordinate partitioning with placement flexibility, Scaled refines an inherited 2D placement, whereas Fresh places the receiving die first and uses its pin locations and clock loads to guide placement of the other die. Data and clock HB assignments are jointly optimized while retaining alternative data sites and reserving capacity for subsequent PG assignment. The selected HB sites and stacks are preserved during 3D clock-tree synthesis, supply-aware cell insertion, and routing. Across eleven designs, including a 133-macro processor, the flow achieves zero backend DRC violations and passes signal-connectivity checks. Nine paired designs admit lower-footprint, lower-power implementations; four improve setup WNS. On JPEG, early joint planning gains 87 ps and mobility-aware partitioning gains 346 ps. Path analysis explains why cell-delay savings can outweigh net-delay increases.

0.21

GARDEN: Efficient GPU-Accelerated Billion-Scale GNN Training via Caching-Sampling-Batching Co-Design

cs.DC ·2026-11-08 ·Qingrui Zhu, Lei Liu, Yushu Cai, Kai Sheng, Zhiwen Chen +1 more
Abstract

Graph Neural Networks (GNNs) have shown impressive performance in diverse tasks such as social network analysis and recommendation systems. However, scaling their training to billion-edge graphs is hindered by the memory wall: GPU memory capacity falls short of the massive storage requirements for node features and topology. Sampling-based GNN training frameworks mitigate this by streaming subgraphs from CPU to GPU, but they incur prohibitive PCIe data transfers and non-coalesced memory accesses, resulting in bandwidth underutilization and low compute efficiency. We propose GARDEN, a GPU-accelerated sampling-based system that co-designs caching, sampling, and batching to eliminate these bottlenecks. GARDEN introduces three key innovations: (1) an Adaptive High-Fidelity Cache Allocator (AHFCA) that combines offline reuse profiling and online hotness tracking to dynamically cache features and adjacency lists within a fixed GPU memory budget, avoiding static allocation waste; (2) a Grouping-based Parallel Sampler that reorders stochastic neighborhood accesses(GPS) to maximize warp occupancy and hide latency on power-law graphs; and (3) a Super-batch Degree Classifier(SDC) that fuses mini-batches into degree-homogeneous groups, enabling cache-friendly coalesced transfers and high GPU utilization. Evaluations on real-world billion-scale graphs demonstrate that GARDEN achieves 4.54x, 1.27x, and 2.97x end-to-end training speedups over DGL-UVA, DUCATI, and FastGL, respectively, while maintaining model convergence and accuracy.

0.20

Ostinato: Fast CPU-GPU MoE Inference for Local Agents via Resident Experts in Every Forward Pass

cs.DC cs.LG ·2026-11-08 ·Junning Zhu
Abstract

Language-model agents increasingly run on workstations with a few consumer GPUs, where large mixture-of-experts (MoE) models keep part of their experts in host memory. With existing CPU-GPU engines, these agents get fast prefill or fast decode, not both: llama.cpp's expert cache skips forward passes of more than 32 tokens to keep the experts of decode cached, so an agent waits 2.1-2.8 s after each tool output, whereas its default layer placement shortens this wait but is 1.5-1.6$\times$ slower in decode. Our key insight is that a short prefill touches too many distinct experts to upload, yet sends 84% of its expert activations (routes) to a stable set of experts that fits in GPU memory. We present Ostinato, a CPU-GPU MoE inference engine that computes these experts on the GPUs in every forward pass without changing what the model computes. Short prefills split their routes, leaving fewer than one of a token's six routed experts per layer to the CPU; long prefills upload each distinct cold expert once; and each decode step runs as one fixed-shape CUDA graph per GPU. On real coding and customer-service agent sessions with DeepSeek-V4-Flash on four RTX 4090D GPUs, Ostinato is the fastest engine in both phases and finishes every session sooner than all six baselines. Against llama.cpp and KTransformers, it cuts the median wait after a tool output by 22-55% and finishes sessions 1.4-2.5$\times$ sooner.

0.19

NCCL-BOT: Closing the Loop on NCCL's Cost Model with Online Bandits

cs.DC cs.LG ·2026-11-08 ·Hanan Shteingart, Amedeo Sapio, Chen Avin, Devendar Bureddy, Manjunath Gorentla Venkata +5 more
Abstract · 46 pages, 13 figures, 38 tables

NCCL picks an algorithm-protocol pair for every collective from an analytic cost model (AUTO) that never sees a measured latency. The model is a strong prior but a brittle final decision: priced on forced-arm runs, its choice is exact on most contexts and badly wrong in a few. We built NCCL-BOT, a tuner plugin that closes the loop: a per-context bandit, warm-started from the model, learns from measured latencies. We replay a forced-arm H100 sweep, tuned at 4-8 nodes and scored at 16-32, and price the model's choice on the same forced arms; that sweep has no stock-NCCL run. The bandit cuts the contexts where the model's choice loses more than 10% from 19 to 1 (p99 103% to 9.9%, mean 6.1% to 1.6%; p=0.047 over nine scenarios after Holm correction). The prior and continued learning both matter: cold-started bandits, and committing after one probe of each arm, lose more on the mean in every scenario. Supervised selectors fit once on the same data and tuned with the same budget match the mean, and their heavier tail is not significant (7 against 1 contexts above 10%, p=0.06); at seen scales they match or beat online selection. With hyperparameters frozen from H100, UCB1 cuts AUTO's p99 from 60% to 23% and from 56% to 15% on two GB200 fabrics, but its contexts above 10% go from 16 to 8 on one and from 10 to 13 on the other, and scenario-level tests do not separate the two. Live, on two latency-emulated node sets of one H100 system with one run per policy, the learners repair most of a gross AUTO error on one (selection loss 31% to 6-8% for deterministic Greedy and UCB1; average all-reduce bandwidth 1.18-1.24x stock) but average 0.94-0.99x stock on the other, and on both lose to stock below 1 MiB (on the first through plugin overhead); on H100 the cost model is exact there offline, so small sizes can be left to it. We do not show a net end-to-end win.

0.18

PLDP-IDS: Personalized Local Differential Privacy for Infinite Data Stream

cs.DB ·2026-11-08 ·Leilei Du, Yikun Hu, Xu Zhou, Yang Cao, Peng Cheng +2 more
Abstract

Existing $w$-event local differential privacy (LDP) mechanisms for streaming data impose a uniform privacy window and budget, forcing all users to adopt the most stringent privacy requirement. We study personalized continual frequency estimation under LDP, where each user specifies her own window size and privacy budget, and identify two challenges. First, a single sampling size cannot balance sampling and perturbation errors for privacy groups with heterogeneous requirements. Second, borrowing estimates from groups with larger budgets reduces variance but introduces bias when privacy preferences correlate with the data. We first propose Optimal Population Selection (OPS) and two OPS-guided mechanisms, PLPD and PLPA, which use one privacy-compatible sampling size for all users. Since one size cannot fit all groups and pooling their estimates is biased under privacy--data correlation, we further develop PLPD$^+$ and PLPA$^+$, which combine group-aware population scheduling with bias-aware cross-group fusion while preserving personalized $w$-event LDP. Experiments on real-world and synthetic streams show that, compared with the direct personalized extension of LDP-IDS, PLPA$^+$ reduces the average mean squared error (AMSE) by 52.3\% and 40.8\% on average, respectively, and remains 4.3--5.1$\times$ more accurate than PLPA under strong privacy--data correlation.

0.17

Agents Propose, Code Verifies, Humans Decide: Verified Agentic Ingestion of Heterogeneous Scientific Tables into Relational Databases

cs.DB cs.AIq-bio.QM ·2026-11-08 ·Bharatha Rankothge, Suzanne Hodgkins, Virginia Rich, Michael Kirby
Abstract · 7 pages, 4 figures

Public scientific datasets rarely arrive analysis-ready. They arrive as messy folders of wide matrices, long tables, spreadsheets, and per-cohort files whose keys and encodings differ, and every reanalysis must reconnect them first. A coding agent can automate this, but its mistakes can look plausible, and without an audit trail, a mistake that corrupts data can go uncaught. We present an AI-assisted ingestion pipeline built on one rule: LLM agents propose, deterministic code verifies, and humans decide. Narrow agents see compact, citable evidence about the files, never the data, and return typed proposals. A verifier with 39 check types tests each proposal against that evidence in two loops, one for the schema's structure and one for its meaning, and a human reviewer approves the schema and the load. Code compiles PostgreSQL DDL, reshapes the data, quarantines rows it cannot place, records provenance, and reconciles every staged row against the database. On four public biomedical datasets, one held out from development, 9 of 20 first proposals failed a check and every failure was caught: the loops repaired eight, and a one-line reviewer edit fixed the ninth. All 45.1 million values loaded with zero reconciliation gap. Discovering a dataset's structure took 4-7 LLM calls and preparing its data at most one more, costing under $0.50 per dataset. In a separate test, each of nine planted data errors was quarantined with a reason or recorded in an audit log.

0.15

Efficient VLA Inference via Retrieval-Based Speculation and Hardware Acceleration

cs.AR cs.RO ·2026-11-08 ·Zhengyang Cheng, He Zhu, Zeyu Guo, Zihan Liu, Anbang Wu +4 more
benchmark
Abstract

Vision-language-action (VLA) models enable robots to understand and execute complex instructions, but their autoregressive (AR) action generation introduces significant latency on resource-constrained robotic platforms. Speculative decoding can accelerate AR-based inference by parallelizing action generation and verification; however, existing approaches rely on costly draft models, which are inefficient on edge platforms. In this paper, we propose Zora, a hardware-software co-design framework for accelerating AR-based VLA inference. Specifically, Zora introduces a lightweight retrieval-augmented generation (RAG)-based draft model that exploits robotic dynamics without additional neural draft models. We further reduce batch verification overhead through a kinematics-guided action-acceptance strategy and a KV-cache-sharing attention mechanism. To address the retrieval overhead introduced by RAG-based drafting, we co-design dedicated hardware support, Zora core, to address the memory irregularity in RAG-based drafting. We evaluate Zora on four VLA models with the LIBERO benchmark. Zora retains comparable accuracy while achieving up to 3.4$\times$ speedup with software-only optimization and up to 7.6$\times$ speedup with hardware support over GPUs.

0.13

Multi-Node Transformers for Object Detection and Captioning Across Distributed Edge Networks

cs.DC cs.CVcs.LG ·2026-11-08 ·Saurabh Kumar, Dewant Katare, Prakriti Jain
Abstract

This paper presents a unified transformer-based framework for object detection and dense region captioning across distributed edge networks. The architecture uses a shared visual encoder followed by detection, region extraction, and autoregressive language-decoding stages, allowing localization and region description to reuse the same image features. This design removes the second image-encoding path required by decoupled detection and captioning pipelines. To support execution on resource-limited devices, a workload-allocation formulation assigns model stages and input streams across connected edge nodes while accounting for execution time, memory capacity, processor utilization, power limits, and intermediate tensor transfer cost. The framework is evaluated on a six-device edge testbed comprising four NVIDIA Jetson Orin Nano and Orin NX inference nodes and two Raspberry Pi 5 auxiliary CPU nodes, with TensorRT FP16 and INT8 execution on the NVIDIA devices. On a Jetson Orin Nano, the unified INT8 model achieves 38.1 FPS with 3.5 GB peak memory use and 11.8 W measured power. Under four concurrent streams, the proposed allocation achieves 35.2 FPS and 28.4 ms mean end-to-end latency with an internode transfer rate of 6.8 MB/s.

0.08

Software & Programming 18

Specializing Meta Multi-Objectivization as Baseline for Software Engineering Optimization

cs.SE cs.AI ·2026-11-08 ·Chengjie Chuan, Pengzhou Chen, Tao Chen
top venue
Abstract · Accepted by IEEE Transactions on Software Engineering (TSE)

Given the discrete nature of many software engineering (SE) optimization problems, trapping at local optima remains a crucial issue in the optimizer designs. Meta Multi-Objectivization (MMO) is an optimization model for overcoming local optima by introducing an extra, auxiliary objective. Yet, MMO is originally designed only for configuration tuning, leaving its suitability for other SE problems in doubt, particularly regarding the choice of the auxiliary objective. In this paper, we study the possibility and resolution for specializing MMO as a general baseline across a range of SE optimization problems, making it a fundamental counterpart for future SE optimizers that seek to mitigate the issue of local optima. We do so by providing both theoretical remit and empirical evidence: we analytically show how the choice of the auxiliary objective is crucial while demonstrating the effectiveness of MMO under nine SE optimization problems comprising 212 cases when the right one can be chosen, i.e., MMO is able to outperform the other state-of-the-art and general-purpose optimizers in 79% and 94% of the cases, respectively. Drawing on the theoretical understanding and empirical findings, we then propose L2RAC, a tool that automatically recommends the best auxiliary objective choice for a given SE problem/case, hence helping to specialize MMO therein. Through extensive experiments, we reveal that, in general, L2RAC can more accurately rank the choice of auxiliary objective for MMO under a case than other guesses by up to 60%, saving as much as 99% of the cost that would otherwise be required to conduct trial-and-error when using MMO.

0.47

Two barriers for minimal time firing squads, with applications to four and five states

cs.FL nlin.CG ·2026-11-08 ·Kia Keng Giam
publishedtheory
Abstract · 34 pages, 18 figures, accepted (awaiting print) at the Journal of Cellular Automata

A minimal-time solution of the firing squad synchronization problem synchronizes every line of n cells at time 2n minus 2. Six states suffice; whether five do has been open since 1987. Every line agrees with the half line until its right ends echo returns. For minimal time solutions with any number of states, we prove that cells at distance j behind the half lines front become time-periodic only after time (1.5 minus epsilon)j, for every epsilon greater than 0 and infinitely many j, and that the half line is never periodic in time and space on a wedge at its left border, hence not eventually regular in our sense. With certified satisfiability checks, we exclude four states using the lines of lengths 2 to 5 and parts of the next five, and show that every five state minimal time solution uses at least 15 non quiescent neighbourhoods on its half lines first 78 anti-diagonals. We give five state rules synchronizing exactly the lengths 2 to 14.

0.47

Spectral Stability, Fatou Basin Contraction, and Machine-Checked Multipliers in Holomorphic Extensions of Generalized Syracuse Dynamics

cs.LO cs.SCmath.CVmath.DS ·2026-11-08 ·Volkan Dağlı, Zerrin Dağlı, Dağhan Dağlı, Emrah Yurtseven, Murat Emre Özkaya
theoryunusual
Abstract · 3 pages, 3 tables; includes ancillary Lean 4 formalization (anc/CollatzAttractor.lean, 20 machine-checked theorems, 0 sorry) and Python veri…

We embed the Syracuse operator $T(n) = n/2$ (even $n$), $T(n) = (3n+1)/2$ (odd $n$) and the Collatz map $C(n)$ into $\mathbb{C}$ via entire functions of exponential type $π$: $f(z) = \frac{1}{4}(1 + 4z - (1 + 2z)\cos(πz))$ and $F(z) = \frac{1}{4}(2 + 7z - (2 + 5z)\cos(πz))$. We prove five results: (1) every integer periodic cycle $C \subset \mathbb{N}^+$ of $T$ is a strictly attracting Fatou cycle of $f$ with multiplier $μ_f(C) = 3^{N_{\mathrm{odd}}}/2^N = \prod_{n \in C\text{ odd}} \frac{3n}{3n+1} \in (0,1)$, and $μ_f(O_{1,2}) = μ_F(O_{1,4,2}) = 3/4 < 1$; (2) on the Fatou basin $B(O_{1,2})$, $g = f^{\circ 2}$ conjugates to $Φ_j(g(z)) = \frac{3}{4}Φ_j(z)$ with Koenigs Taylor jets $c_2(1) = -\frac{13}{2}π^2$ and $d_2(2) = -\frac{9}{2}π^2$; (3) the Schwarzian derivative is strictly negative on $\mathbb{Z}$ ($S(f)(n) \le 3π^2 - \frac{3}{8}π^4 \approx -6.9196 < 0$), and a repelling Julia 2-cycle $O_{\mathrm{rep}} \subset J(f)$ ($μ\approx 1.2148 > 1$) isolates $B(O_{1,2})$; (4) on $(\mathbb{Z}_2, μ_{\mathrm{Haar}})$, the 2-adic Lyapunov exponent $λ_{\mathrm{Haar}} = \frac{1}{2}\ln(3/4) \approx -0.143841 < 0$ equals $\frac{1}{2}\ln|μ_f(O_{1,2})|$; and (5) we certify 20 theorems in Lean 4 (anc/CollatzAttractor.lean, 0 sorry). These results establish local Fatou invariants and 2-adic ergodic averages for $f(z)$, leaving open whether every $n_0 \in \mathbb{N}^+$ lies in $\bigcup_{k \ge 0} f^{-k}(B(O_{1,2}))$.

0.37

Application-Defined Abstractions to Separate Numerical Intent from Computation Orchestration

cs.SE cs.MS ·2026-11-08 ·Elena Aschmann, Anshu Dubey, Mohamed Wahib, Jens Domke
Abstract

In High-performance Computing the holy grail has been to translate numerical algorithms expressed in mathematical notation into code that performs well across a range of platforms. A variety of portability frameworks have abstracted data layout, parallelization and orchestration from application's view, however, the resulting format is still code that conforms to the idioms of the framework. Additionally, substantial refactoring is typically needed to achieve performance goals. Increasing sophistication of Large Language Models (LLMs) naturally raises the question whether they can be used to make application codes and/or solvers both portable and performant. The obstacle is the lack of correctness guarantee in LLM produced code. In this paper we pose the question differently -- can we start with expressing numerical intent of the code into an immutable core arithmetic component isolated from the surrounding code such that the surrounding code can be safely generated by the LLMs? We answer this question by using Spark, a mature compressible hydrodynamics solver in Flash-X, a multiphysics software system. We use LLM assistance in devising the abstractions and embedding them in the solver. Once properly verified, the resulting solver is lowered to concrete realization deterministically with tools also built with LLM assistance. The driver of the numerics and the code for interfacing with the application are generated by the LLMs. The resulting solver is exercised within Flash-X and with a separate environment we built for this experiment. Our results show that this is a viable approach for producing correct and performance portable code.

0.35

MicroSim: Contrastive Graph Neural Networks for Binary Code Similarity Detection

cs.SE cs.LG ·2026-11-08 ·Moussa Moussaoui, Tarik Houichime, Abdelghani Souhar, Abdelalim Sadiq
benchmarktheory
Abstract

We present MicroSim, a system for binary code similarity detection that represents each function as a graph built from IDA Pro Hex-Rays microcode and compares such graphs with a contrastive Siamese graph neural network (GNN). We evaluate on the TREX cross-optimization benchmark, a corpus of 849,072 function-level microcode graphs spanning optimization levels -O0 through -O3, under a strict class-disjoint split in which no function key is shared between train and test. Under this protocol, MicroSim reaches 0.9972 area under the receiver operating characteristic curve (ROC-AUC) and 0.9762 Matthews correlation coefficient (MCC), outperforming Gemini (0.9332 ROC-AUC), SAFE (0.9752), and jTrans (0.9740) retrained on the same split, as well as a structure-free Bag-of-Words (BoW) baseline (0.9484). We analyze the gap between pair classification and ranked retrieval in graph transformers, showing that Cross-domain Similarity Local Scaling (CSLS) under unified tie-resolved evaluation raises GraphGPS mean reciprocal rank (MRR) from 0.2542 to 0.2559 (R@1 = 0.1639, R@10 = 0.4561), strictly satisfying the theoretical MRR upper bound. Finally, over 5 seeds with equal compute, the pairwise cosine-embedding loss attains a slightly higher ROC-AUC (Δ = -0.0256, 95% confidence interval [-0.3282, +0.2770]), whereas the supervised-contrastive information noise-contrastive estimation (InfoNCE) loss is significantly better on thresholded classification and ranked retrieval.

0.34

AWS Serverless Infrastructure from WebAssembly Components

cs.SE cs.DC ·2026-11-08 ·Philipp Vollmer
Abstract

The infrastructure definitions and the code for serverless applications are often only loosely connected. This yields a large error class: the referenced infrastructure components might not exist. It is also hard to keep the application and its permissions in sync. To address this problem we present an infrastructure from code solution where we provision cloud infrastructure from Wasm Interface Type (WIT) - the language defining the interface of WebAssembly components. We limit ourselves to the basic example of an AWS Lambda function accessing S3 buckets. We build two interpreters for the WIT language: one that derives the definition of S3 buckets as infrastructure as code and least-privilege IAM permissions, and a generic runtime that runs the component as a Lambda function and provides access to the S3 buckets as capabilities. We formulate consistency properties that ensure the existence of the S3 buckets and correct permissions of the Lambda, and provide evidence that our interpreters satisfy them. We show how selected failures for deployment of such cloud components can be prevented using our technique and how a baseline setup fails at this task. Together with the strict sandboxing of the WebAssembly runtime we achieve an instance of capability-based security for access to AWS infrastructure as a further result.

0.33

Addressing Bias Early in the AI Lifecycle: Social Determinants of Health in Functional and Cognitive Assessment Labels of Nursing Home Residents

cs.LO stat.OT ·2026-11-08 ·Isabel Amaya-Rodríguez, Cristina Martin, Nekane Larburu, Teresa Garcia-Navarro, Roberto González Novas +8 more
Abstract · 14 pages, 24 tables

Fairness in clinical artificial intelligence (AI) is usually evaluated after model training, yet bias can enter earlier in the AI lifecycle, during data collection and labelling. When routine clinical assessments serve as target variables, systematic differences in these labels across sociodemographic groups are carried into model development, validation and updating. This study addresses this early lifecycle stage by auditing labels from three standardised instruments used in residential care: the Barthel Index (BI), the Mini-Mental State Examination (MMSE) and the Global Deterioration Scale (GDS). Using electronic health records of 2,407 residents from 12 nursing homes in the Basque Country, Spain (2011-2021), we assessed four social determinants of health (SDoH) - sex, educational attainment, lifetime work activity and marital status - with two group fairness metrics, Disparate Impact (DI) and Statistical Parity Difference (SPD), computed for each outcome category with bootstrap confidence intervals. Disparities were found for all four attributes and across the three instruments: men, residents with intermediate education and never-partnered residents were more frequently classified in categories indicating preserved function or cognition. These disparities may reflect genuine differences in health status, measurement properties of the instruments, assessment practices or selective admission to residential care, and cannot be attributed to a single cause with the present design. Regardless of their origin, any model trained on these labels would inherit them, and they could persist through later lifecycle stages unless explicitly examined. These findings support pre-training label auditing as the first step of a continuous fairness assessment spanning model development, validation, deployment and updating.

0.33

SWHForest: A Structural Curation and Search Layer for Large-Scale Code Corpora

cs.SE cs.LG ·2026-11-08 ·Federico Pennino, Andrea Gurioli, Maurizio Gabbrielli, Paolo Ferragina
benchmarkcode/data
Abstract · 5 pages, 3 figures, 2 tables. Submitted to the MSR 2027 Data and Tool Showcase Track. Code: https://github.com/halykoss/swhforest ; Data: ht…

Pre-training code language models relies on massive code corpora usually streamed over the internet. However, researchers rarely need a multi-terabyte archive in its entirety; instead, their workflow centers on curated subsets: selecting well-documented files, filtering out repetitive boilerplate, auditing for benchmark contamination, or retrieving specific code patterns. Yet, because modern corpora are typically published as file pointers and basic attributes rather than source text, no general structural layer exists to select which files to download: curating a subset requires downloading and parsing millions of files first just to discard most of them. We present SWHForest, an open-source framework and public dataset that provides this missing structural layer for any pointer-based corpus without redistributing source code. A single pass over the archive parses each file into an AST and immediately discards the source text, keeping only its structural summary. Queried locally or exposed by archive providers as a server-side filtering API, SWHForest selects matching files first (via SQL or index intersection) and downloads only surviving candidates. Across the ten largest languages of Stack-Edu (52.1M unique files, 341M definitions), SWHForest answers curation queries in milliseconds while cutting downloads by 46-72%, reducing live structural searches from hours of per-language streaming to seconds or minutes.

0.31

FESOM2.8 in single precision: climate fidelity over six decades and performance on CPUs and GPUs

cs.MS physics.ao-phcs.PF ·2026-11-08 ·Suvarchal Kumar Cheedela, Nikolay V. Koldunov, Sergey Danilov, Jan Streffing, Dmitry Sidorenko +1 more
Abstract

Halving the floating-point precision of an ocean model can make it substantially faster, but whether the rounding errors stay negligible over climate time scales has rarely been tested. We implement single precision in the global ocean-sea-ice model FESOM2 and compare 60-year hindcasts with double-precision controls. A first implementation lost salt steadily, accumulated a heat-content deficit of 129 ZJ, and shifted major ocean fronts. Controlled experiments trace the drift to the tracer update: small increments added to a large salinity value are rounded away, and the increments themselves are biased. Accumulating tracer tendencies and layer thicknesses in 64 bits, and carrying the rounding error of each tracer update into the next, removes the drift for at most 8% additional runtime. Single- and double-precision simulations then differ in heat content and mean state about as much as two double-precision runs with slightly perturbed initial conditions, sea ice differs by at most 0.2%, and he salinity difference is four orders of magnitude smaller than before. In the coupled model AWI-ESM3, single precision adds no detectable heat source or sink over 60 years. Single precision speeds up the 60-year hindcasts by a factor of 1.35, and the model by up to 1.66 on meshes of up to 5 km resolution; on GPUs, a C++/Kokkos version runs up to 1.27 times faster and needs 43% less memory. Whether reduced precision is safe for climate integrations is decided by long-term budgets, not by short tests.

0.31

Parametric Proof-Carrying Compilation for Quantum Error-Correction

cs.PL quant-ph ·2026-11-08 ·John Zhuoyang Ye, Jens Palsberg
theory
Abstract

We present PC-QEC, a proof-carrying compilation framework for quantum error correction parameterized by both code distance and syndrome-extraction scheme. It covers all four schemes---Standard, Flag, Shor and Knill---with circuit-level distance guarantees for every legal code distance in each proved family under the declared error and readout contracts. The circuit-level model permits single-wire Pauli errors at every operation's qubit operands, capturing data errors, readout errors and ancilla errors that propagate into multi-qubit hook errors. For each family and scheme, supplied code-algebra and invariant proofs combine with proved scheme-specific propagation lemmas into a finite symbolic certificate whose size is independent of code distance. We demonstrate family-wide guarantees for five code families across all four schemes. For rotated Surface code, one Standard Scheme certificate proves a distance lower bound $d$ for every odd $d\ge3$ and checks in 214\,s on average. Lean 4 mechanizes the metatheory and exact circuit-level distances for canonical rotated Surface, unrotated Surface and Toric code families.

0.30

Declarative DNS - Resolving Recursive Resolution

cs.LO cs.NI ·2026-11-08 ·Felix Linker, Yue Huang, David Basin, Huayi Duan
theory
Abstract

DNS zone configuration errors are a frequent source of outages and frustration, particularly in combination with DNSSEC. Research has attempted to address the problem of zone configuration errors using formal methods, but existing proposals (i) only verify the absence of specific, known configuration errors, and (ii) do not consider DNSSEC. We present a new formal model of the DNS that covers most of DNSSEC, making it one of the most complete DNS models to date. In addition, we present a new tool, DecDNS, that performs a resolvability change-impact analysis of zone configuration updates using our model. We show that such a change-impact analysis can identify configuration errors without explicitly searching for them. We evaluate DecDNS both formally and empirically. We prove that DecDNS is sound and complete and show that its results agree with widely used DNS software. Finally, we explain how our approach can be integrated into operational practice, helping guide zone management in the real world.

0.29

Advancements in Cloud-Native Systems: A Systematic Mapping Study of Reliability and Scalability across Cloud-Native, Edge-Native and AI-Enabled Systems

cs.SE cs.AIcs.MAcs.NI+1 ·2026-11-08 ·Shrinivass Arunachalam Balasubramanian
Abstract

Cloud-native systems now extend to the network edge and host artificial-intelligence (AI) workloads, including AI agents, but evidence on their reliability and scalability is scattered and uneven. We report a PRISMA 2020 systematic mapping study with an evidence-graded synthesis. Five logged OpenAlex searches on 11 October 2026 covered cloud-native systems, edge AI, agentic AI for operations, reliability of agentic applications and physical AI; 189 records from an earlier exploratory search were added separately. After 301 duplicate removals and an automated peer-review venue rule, 4,611 records were screened against written criteria and a five-level relevance scale. The review includes 1,794 primary studies from 2017-2026 and 95 secondary studies, with every decision released in Dataset S1. Each primary study was assigned to a 13-category taxonomy and graded as simulation, prototype or testbed, production data, production deployment, or not determinable from the abstract. Of 1,003 studies with a determinable level, 66.2% used prototypes or testbeds, 18.0% simulations and 15.8% production data or deployments. A 93-study synthesis core, appraised from indexed abstracts, shows production deployments of cascading-failure prevention, trace- and graph-based root-cause analysis and predictive autoscaling, and production-trace evaluations of cold-start mitigation. Edge reliability and physical AI rely mainly on simulations and testbeds. AI-agent reliability uses tool-grounded diagnosis, explicit safety properties, trace-driven repair and workload-aware scheduling, while empirical studies show that agent frameworks remain fragile. Conclusions are limited to the documented corpus. The principal gaps are production evidence for edge-native systems, shared resilience benchmarks and dependable agentic operations.

0.28

Symbolic Synthesis with Token-Level LLM Guidance

cs.PL ·2026-11-08 ·Derek Egolf, Stavros Tripakis
Abstract

We propose a synergistic combination of classical program synthesis techniques and contemporary large language models (LLMs) to solve the problem of recursive program synthesis from first-order logic specifications. Our approach leverages the strengths of both paradigms: we use LLMs as a source of statistical bias to guide the search for a correct program. At the same time we use techniques from classical synthesis to short-circuit on semantically incorrect prefixes and to avoid sampling semantically equivalent responses. The statistical bias is derived from the next-token distribution of an LLM, which is highly context-dependent. We operate with the understanding that (1) there is some class of synthesis instances that can be discharged by small models and symbolic tools, and (2) every other instance can be decomposed into these instances. Therefore, our intention is to expand the reach of small, open-weight LLMs and symbolic synthesis techniques, not to compete with proprietary frontier models on every possible instance. These are different tools for different jobs. In our experimental evaluation, we find that our method solves more benchmarks than either small LLMs or classical synthesis techniques alone. Our method solves 30/32 benchmarks using only a Qwen2.5 model with 32 billion parameters. Qwen2.5 solves 0 benchmarks on its own, and only solves a few additional benchmarks when combined with traditional constrained decoding techniques.

0.25

NumericalDiff: Numerical Program Differencing via Value-Based Trace Alignment and Intervention

cs.SE ·2026-11-08 ·Xiang Chen, Huanhuan Xia
Abstract · The arXiv v1 version corresponds to Project version 0.0.1

Numerical programs implementing the same computational task can exhibit different numerical behavior because of their execution stacks, program representations, or program logic. Source text and final outputs alone provide limited diagnostic information: numerical differences can propagate through corresponding computations, while distinct executions can produce equal outputs. We present NumericalDiff, an execution-based method that records assignment-level numerical events, aligns traces using a tolerance-aware longest common subsequence independent of variable names and source locations, and optionally reexecutes after replacing one intermediate value with its counterpart. Among the original and all single-intervention executions, it lexicographically maximizes matched events and minimizes unmatched events. NumericalDiff supports code comparison across language and framework front ends, currently including NumPy, PyTorch, JAX, C, C++, Fortran, and Octave. We evaluate NumericalDiff on 36 controlled cases comprising 675 pair/input instances: 36 execution-stack, 588 program-representation, and 51 program-logic instances. All 24 operator-injection instances expose trace differences, including three with tolerance-equal final outputs, while complete trace alignment is achieved for all 441 canonical cross-representation instances. Within the 51 program-logic instances, intervention improves 24 alignments and raises macro-average matching coverage from 0.4210 to 0.6513. The results show that value-based trace alignment and reexecution can delimit propagated numerical differences across heterogeneous implementations.

0.23

Decidability of Elementary Modal Logics is Undecidable

cs.LO ·2026-11-08 ·Jakub Michaliszyn
theory
Abstract

We study the decidability of modal satisfiability over classes of Kripke frames defined by universal first-order sentences. Although some such classes yield undecidable satisfiability problems, decidability is guaranteed for wide families of frame conditions, including universal Horn sentences and universal sentences implying transitivity. A more than decade-old open problem asks a natural question: can decidable and undecidable cases be distinguished algorithmically? We prove that this is impossible, already for equality-free universal frame conditions over a single binary relation. More precisely, there is no algorithm that, given such a condition, decides whether its associated local modal satisfiability problem is decidable. We first establish the result for conditions with equality and then eliminate equality by passing to quotients of worlds with identical incoming and outgoing edges.

0.16

Keel: A Verified In-Kernel Proof Checker for Proof Carrying BPF

cs.PL ·2026-11-08 ·Yalu Cai, Tung-Che Chang, Santosh Nagarakatte
theory
Abstract · 66 pages

Recent systems such as the eBPF Certificate Framework (BCF) and Proof-Carrying BPF address the imprecision in the Linux eBPF verifier with proof-carrying code. When the verifier's abstract domains cannot prove a safety condition, an in-kernel proof checker checks an untrusted user-space refutation proof. A single bug in this checker can compromise the security of the entire kernel. While formalizing BCF's in-kernel checker, we discovered 13 bugs such as unsound rewrite rules, silently unchecked proof steps, and memory safety errors. We demonstrate that forged certificates can trick the kernel to accept arbitrary unsafe eBPF programs. To address this problem, we propose Keel, a formally verified in-kernel proof checker for the Linux eBPF verifier. First, we specify the checker as a pure functional program in Lean and prove its soundness against bit-vector semantics. We decouple the steps of the proof certificate checker from the mathematical rewrite rules, which enables us to extend the rewrite rules without the need to change the checker. Second, we implement the checker in EmbC, a deeply embedded C fragment, and prove that our C implementation refines the functional specification without undefined behavior or memory leaks. Third, using an amortized potential argument, we prove linear time and memory bounds for the C checker. Keel is correct-by-construction while also being faster than the BCF checker inside the kernel, 1.3X in total and 2.0X on the slowest 1.4% of certificates.

0.14

Replay Fidelity in LLM Agent Evaluation: Randomness, Hidden State, and Test Oracles in AppWorld

cs.SE cs.AI ·2026-11-08 ·You Liu, Quanchao Lu
Abstract · Empirical study; 11 pages, 3 figures, 5 tables; historical AppWorld census with a bounded BFCL consistency check

Reproducibility in large language model (LLM) agent evaluation depends on replay fidelity: which parts of an execution trace are reconstructed, controlled, and compared. This matters for agent debugging, failure analysis, and counterfactual evaluation of stateful tool use. We audit all 71 unsuccessful executions in a fixed historical AppWorld corpus using 284 isolated fixed-code replay attempts. Reconstructed declared settings admit 42 cases under printed-output, pre-oracle database-delta, and native-verdict/test-count checks; resetting the random-number generator before each interaction admits 68. The 26-case gap contracts to four under database/oracle checks, with one endpoint unknown. Mechanism studies identify four materialized-state discrepancies despite matching outputs and verdicts. In one admitted endpoint pair, an identical API call yields matching responses but different row effects; common future Python RNG removes that deterministic difference in two paired repetitions. A separate JWT credential-equality program and alignment controls expose authentication/write differences, with missing repetitions retained. For all 10 affected post-oracle export cases, native export from retained output trackers restores the reference comparison without changing original cached contents. A purposeful 12-case BFCL pilot shows no discrepancy under its tested public-state, return, and verdict checks. We preserve resource censoring and infrastructure deviations. The contribution is a bounded empirical characterization of replay eligibility, hidden state, and test-oracle integration, without estimating debugger accuracy, repair gains, or method-ranking distortion.

0.12

Beyond Code Translation: Agentic Android-to-HarmonyOS Migration

cs.SE ·2026-11-08 ·Yinghang Ma, Wuyang Dai, Song Wang, Filipe Cogo, Zhen Ming +1 more
Abstract

Migrating Android applications to HarmonyOS is an important yet challenging software engineering task. Existing large language model (LLM)-based migration approaches primarily translate code at the method or file level, overlooking project-wide dependencies and workflow coordination. Moreover, ArkTS, the primary language for HarmonyOS, is relatively low-resource, and migrated applications often lack sufficient tests for reliable verification, making end-to-end migration error-prone. We present A2HTrans, an agentic framework for end-to-end Android-to-HarmonyOS application migration. Rather than performing isolated code translation, A2HTrans decomposes applications into functionality-aware migration tasks, prioritizes them through dependency analysis, and orchestrates translation, validation, and iterative repair within a unified execution workflow. To improve migration reliability, A2HTrans employs a tool-augmented verification strategy that combines compiler-based syntax checking with LLM-based semantic validation and automated repair. We evaluate A2HTrans on 14 Android applications spanning diverse sizes and domains using Claude-Sonnet-4.6 as the underlying migration model. Compared with Claude Code, A2HTrans preserves 8.3% more functional completeness, improves behavioral correctness by 8.0%, and reduces migration time by 28.8%. These results demonstrate that structured agentic execution substantially improves the reliability and efficiency of project-level cross-platform software migration.

0.08

Theory & Algorithms 30

Fast Approximate $\ell_p$ Chamfer Distance via Lopsided Embeddings and Structured JL

cs.DS cs.LG ·2026-11-08 ·Ying Feng, David P. Woodruff
top venue
Abstract · Accepted at NeurIPS 2026

For two $d$-dimensional point sets $A, B$ of size $n$, the Chamfer distance from $A$ to $B$ is defined as $\text{CH}(A,B)=\sum_{a \in A} \min_{b \in B} \mathsf{dist}(a, b)$, where $\mathsf{dist}$ is some underlying distance measure. We present efficient algorithms for approximating the Chamfer distance under the $\ell_p$ norm for $1 < p \le 2$. For the general case $1 < p < 2$, we utilize lopsided $\ell_p$ to $\ell_1$ embeddings with weak guarantees. We show that they are sufficient for preserving the Chamfer distance. In the high-dimensional regime, we use Fast Matrix Multiplication techniques to speed up the lopsided embeddings. These lead to $O(nd\cdot \mathrm{poly}(\log \log n, \log(1/\varepsilon))/\varepsilon^2)$ total time. For the specific Euclidean case ($p=2$), we leverage the Fast Johnson-Lindenstrauss Transform based on Toeplitz matrices and re-analyze the previous $\ell_1$-specific Chamfer algorithm in $\ell_2$. These achieve a runtime matching the recent state-of-the-art $\ell_1$ result of $O(nd(\log \log n + \log (1/\varepsilon))/\varepsilon^2)$, improving upon the previous $O(nd \log n/\varepsilon^2)$ bound for $\ell_2$. We also give additional results for the angular distance and upper and lower bounds in the streaming setting.

0.45

Homological Barriers to Stable Learning of First-Order Nash Equilibria

cs.GT cs.LG ·2026-11-08 ·Ashkan Soleymani, Gabriele Farina, Patrick Jaillet, Georgios Piliouras
published
Abstract · The Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS) 2026

We identify a homological barrier to stable learning of local Nash equilibria, also known as first-order Nash equilibria (FONE), in constrained non-concave differentiable games. We consider continuous deterministic learning dynamics that fix every FONE point and aim to attract all initial conditions to the full FONE set. We show that this convergence guarantee is impossible when the target set has nontrivial positive-degree cohomology, such as that arising from a loop, and the state space is compact and contractible. We exhibit this barrier in a simple non-concave quadratic zero-sum game on a box whose exact FONE set consists of a diagonal six-edge loop and an isolated point. Consequently, no such dynamics can make the full FONE set a stable global attractor. We further show that degree two is the minimal polynomial degree at which this barrier can arise and extend the construction to obtain nontrivial homology in arbitrarily high degrees. Finally, we establish that the barrier persists for sufficiently small approximate-FONE targets under small perturbations, yielding an open, positive-volume family of quadratic zero-sum games with the same homological barrier.

0.41

Lower bound on the size of splitting systems

cs.DM math.CO ·2026-11-08 ·Fan Chung, Namiko Matsumoto, Arya Mazumdar
theory
Abstract · 13 pages

For even $t$ and an $n$-set $X$, an $(n,t)$-splitting system is a family ${\mathbb{B}} \subseteq 2^X$ with the property that for each subset $Y \in {X \choose t}$, there exists $B \in \mathbb{B}$ such that $|B \cap Y| = t/2$. The problem of interest is to determine the minimum size of such splitting systems. We show that any $(n,t)$-splitting system has size at least of order $t$, improving the previous lower bound of $\sqrt t$ while the best known upper bound is of order $t^{3/2}$.

0.30

Online Correlated Exponential Clocks

cs.DS ·2026-11-08 ·Zhiyi Huang, Kalen Patton, Peilin Yang, Jiahao Zhao
Abstract

In online resource allocation, we seek to allocate an incoming sequence of requests $j \in J$ among a set of offline resources $i \in I$, where decisions must be made immediately without knowledge of the future. State-of-the-art results typically follow from relax-and-round methods. For many problems, recent advances have come from improved methods of online rounding with negative correlations. However, such rounding schemes have largely been implemented on a case-by-case basis, making it difficult to transfer improvements and new ideas between problem settings. This leads us to ask the question: Can we develop unified methods of online rounding with negative correlations which can obtain state-of-the-art guarantees across different resource allocation settings? We answer this question affirmatively by introducing the framework of Online Correlated Exponential Clocks (OCEC). In this framework, fractional allocations $x_{ij}$ are assigned exponential clocks $Y_{ij} \sim \mathrm{Exp}(x_{ij})$, and each request $j$ is allocated to the resource $i$ with smallest $Y_{ij}$. The key idea is that the sequence $(Y_{ij})_j$ for each fixed $i$ is negatively correlated via an OCEC generator, resulting in better-than-baseline guarantees. Our contributions are two-fold: First, we introduce OCEC rounding as a problem-independent interface which any online algorithm can use. This isolates the challenge of online rounding into the self-contained problem of designing an effective OCEC generator. Second, we demonstrate the flexibility of this method by giving a single OCEC generator which yields improved competitive bounds in three distinct problems: 0.556 for Display Ads, 0.530 for AdWords, and 4.69 for squared-load balancing. Previous best results were obtained by Blanc and Charikar (FOCS 2021), Chen, Huang, and Sun (FOCS 2024), and Borst and Kashaev (SODA 2027), respectively.

0.30

A note on exact external information complexity

cs.CC ·2026-11-08 ·Henning Wunderlich
Abstract

We study the prior-free zero-error private-coin external information complexity $B(f)$ of finite two-party functions, with correctness on every input. An elementary rectangle/entropy argument gives $B(\mathsf{AND})=\log_2 3$. A complete classification of Boolean $3\times3$ matrices leaves, up to input permutations and output complementation, only the two-overlapping-rectangles function $F$ without an exact value. An entropy potential and a finite rational protocol give $2.1212661667060486842\ldots\le B(F)<2.130998318$. We also determine its deterministic external information and zero-error communication costs. A separate product-prior certificate and internal-information additivity yield $R_0^{\mathrm{pub}}(F^{\times t})\ge2.021991187178562\ldots\,t$, strictly exceeding the fractional-cover rate $2t$.

0.30

Compression of Sources with Memory and Private Random Access

cs.IT ·2026-11-08 ·Venkat Chandar, Aslan Tchamkerten, Shashank Vatedka
Abstract · 22 pages, 2 figures

Private local decoding imposes a stringent constraint on compression: the codeword bits used to recover a source symbol must reveal nothing about the remaining symbols beyond what the recovered symbol itself reveals. For memoryless sources, this constraint is known to be compatible with entropy-rate compression. We show that the same holds for any stationary ergodic source over a finite alphabet. Specifically, every rate above the entropy rate is achievable with exact privacy and error-free local decoding on a set of source sequences whose probability tends to one. In contrast, we exhibit non-ergodic sources for which private local decoding incurs a strictly positive rate penalty.

0.29

Õptimal Facial Distance Labeling in Planar Graph

cs.DS ·2026-11-08 ·Noy Rotbart, Oren Weimann
theoryhot direction
Abstract

We show how to assign labels of size $\tilde O(n^{1/3})$ to the vertices of a face $F$ in an $n$-vertex unweighted undirected planar graph $G$ s.t one can deduce the distance between any two vertices of $F$ given their labels alone. This matches (up to log factors) the $Ω(n^{1/3})$ lower bound of Gavoille, Peleg, Pérennes, Raz from 2004. Prior to our work, the smallest labels were of size $O(n^{1/2})$.

0.28

Linear Set Reconciliation with $O(d)$ Additive Redundancy

cs.DS cs.IT ·2026-11-08 ·Tianren Liu, Rui Xu, Tong Yang, Kangyang Zhou
Abstract

In set reconciliation, two parties holding sets $A, B \subseteq [n]$ with $|A \triangle B| \le d$ seek to recover their symmetric difference from short $\mathbb{F}_2$-linear sketches, which is equivalent to recovering a $d$-sparse vector $x = \mathbf{1}_{A \triangle B} \in \mathbb{F}_2^n$ from its syndrome $Hx \in \mathbb{F}_2^m$. While the information-theoretic counting bound requires $L(n,d) := \log \sum_{i=0}^d \binom{n}{i} = d\log(n/d) + Θ(d)$ bits, existing polynomial-time schemes either use at least $d\lceil\log(n+1)\rceil$ bits (algebraic sketches, leaving an $Ω(d\log d)$-bit gap) or incur a multiplicative $(1+δ)L(n,d)$ overhead (peeling-based sketches), raising the question of whether $L(n,d) + O(d)$ bits is achievable with fast updates and decoding. We construct a public-coin randomized $\mathbb{F}_2$-linear sketch of length $L(n,d) + O(d) + λ\log^{O(1)} n$ that recovers each fixed input of weight at most $d$ with probability at least $1 - 2^{-λ}$, with $\log^{O(1)} n$ update time and $d\exp(O((\log\log n)^2)) = dn^{o(1)}$ worst-case decoding time. In particular, for $n \le 2^{\operatorname{polylog}(d)}$ and $λ\le d^{1-Ω(1)}$, this achieves $L(n,d) + O(d)$ bits with $\operatorname{polylog}(d)$ update time and $d^{1+o(1)}$ decoding time. Our construction encodes $d$ permuted rows via randomly scaled Pinsketch syndromes and sketches the resulting bit array column by column. Decoding columns sequentially from prefix predictions cancels recovered rows and exposes undecided rows as decoder-side hints, reducing each column to a hinted sparse recovery problem that we solve via a recursive sketch, HintSketch.

0.28

Amortized Cell-Probe Lower Bounds for Dynamic Boolean Problems

cs.DS cs.CC ·2026-11-08 ·Young Kun Ko
theoryhot direction
Abstract · 20 pages

We prove the first unconditional amortized $Ω\big((\log N/\log\log N)^2\big)$ cell-probe lower bound for a Boolean dynamic problem. For insert-only multiphase Disjointness with $N$ query sets over $[N]$ and word size $w=Θ(\log N)$, a fixed distribution of interleaved insertions and queries forces this expected average cost per operation, excluding preprocessing. The bound allows sufficiently small constant error for each query of every fixed legal sequence and imposes no restriction on individual operation costs. It transfers to incremental directed reachability, incremental subgraph connectivity, and fully dynamic triangle detection. We also prove $t_q=Ω\big((\log N/\log(t_uw))^2\big)$ for amortized insertion time $1\le t_u=N^{o(1)}$ and expected query time $t_q$, when the number of insertions before the final query is not announced in advance. We reach the quadratic-logarithmic scale with bounded error by combining interleaving with the 2.5-round communication framework [Ko26a, Ko26b] and controlling errors among accepted simulations.

0.27

Fractional distance-edge-monitoring in lexicographic products

cs.DM ·2026-11-08 ·Xiumin Wang, Minglei Mao, Yaping Mao, Ralf Klasing
theory
Abstract

We give an exact formula for the fractional distance-edge-monitoring number of $G\circ H$, where $G$ is connected of order $m\ge2$ and $H$ has order $n\ge2$. By the previously established endpoint-monitoring property, the fractional monitoring program reduces to fractional vertex cover, whose linear-programming dual is fractional matching. Writing $d_c(H)$ for the critical difference of $H$, we obtain $\operatorname{dem}_{f}(G\circ H)=(mn-D(G,H))/2$, where $D(G,H)=\max_{A\in\mathcal{I}(G)}\bigl(d_c(H)|A|-n|N_G(A)|\bigr)$ and $\mathcal{I}(G)$ denotes the family of all independent sets of $G$, including the empty set, while $N_G(A)$ is the open neighborhood of $A$. This formula characterizes the products attaining $mn/2$ and gives an equality criterion for the fractional matching lower bound of Ma, Cheng, Mao and Wang. Under the stated hypotheses, the lower bound of Ma et al. is attained if and only if either factor has a fractional perfect matching or $H$ is edgeless. Two outer trees show that the factor orders and fractional matching numbers do not determine the product value.

0.27

Parametrize Integer Programming by Packing Imbalanced Circuits

cs.DS math.OC ·2026-11-08 ·Yuheng Shi
Abstract

We study the standard form Integer Programming problem (IP) parametrized by the imbalanced circuit packing number. This number is $0$ if and only if $A$ can be transformed to a totally unimodular matrix by row operations, in which case (IP) is solvable in polynomial time. We show that if the constraint matrix represents a graphic matroid, (IP) is fixed-parameter tractable when parametrized by the imbalanced circuit packing number of the matrix. The proof combines an Erdős-Pósa property for edge-colored non-monochromatic cycles with a reduction to mixed-integer programming using few integer variables. In contrast, by a reduction from the parity potential problem, we show NP-completeness of feasibility of (IP) when the constraint matrix represents a cographic matroid, even if kernel of the constraint matrix is $2$-regular and imbalanced circuits pairwise intersect.

0.27

Joint Design of Communication and Control in Center-Fed Pinching-Antenna Systems

cs.IT ·2026-11-08 ·Xu Gan, Yuanwei Liu, Hyundong Shin
Abstract

An analytical and optimization framework for the joint design of communication and control (JDCC) in a center-fed pinching-antenna system (C-PASS) is proposed. Within this framework, the mathematical coupling between data throughput and closed-loop control cost is characterized for JDCC. The analysis captures the mutual interference between concurrently transmitted data and control streams during uplink state uploading and downlink command delivery. Based on this coupling, an achievable communication-control performance region is established for C-PASS. Accordingly, a throughput maximization problem is formulated to jointly optimize digital and pinching beamforming subject to control-cost constraints. To address this problem, an alternating optimization algorithm is developed based on constraint reformulation and resource decomposition. The proposed framework reveals how control requirements shape achievable data throughput, guiding the joint beamforming design. Numerical results demonstrate that 1) the proposed C-PASS design achieves 110.2\% higher frame-average throughput than conventional PASS, and 2) the proposed concurrent data-control transmission protocol increases frame-average throughput by up to 66.7\% over separated transmission.

0.26

Proof of the Aaronson-Ambainis Conjecture: An Exposition

cs.CC quant-ph ·2026-11-08 ·Guy Blanc, Jordan Docter, Carmen Strassle, Li-Yang Tan
Abstract · 30 pages, 6 figures

We give an exposition of a recent proof of the Aaronson-Ambainis conjecture, which states that every degree-$d$ polynomial $p : \{0,1\}^n \to [0,1]$ has a variable of influence at least $\mathrm{poly}(\mathrm{Var}(p)/d)$. This implies that quantum query algorithms for decision problems can be efficiently simulated classically on most inputs. It also yields a near-equivalence between the status of the $\mathsf{BPP}$ vs. $\mathsf{BQP}$ problem in the random oracle and unrelativized worlds. We organize the proof around four main steps and discuss how each connects to the literature on the analysis of boolean functions and probabilistic combinatorics. To bring out the key ideas, we do not optimize parameters, and our presentation yields weaker exponents than the original. $\textbf{Authors' note.}$ The proof was discovered by David Turturean using a research system developed at Physical Superintelligence (PSI), in combination with a non-public Anthropic model. The proof was then shared with the authors, who worked to understand it and prepare this exposition. Anthropic's public models were used in this process. The authors did not receive financial compensation from Anthropic or PSI. This article is not intended for publication at a conference or in a journal.

0.26

Impropriety and resilience of the chromatic number

cs.DM math.COmath.PR ·2026-11-08 ·Ross J. Kang, Colin McDiarmid
Abstract · 22 pages

We show that, for $p$ bounded away from $1$ and $np\to\infty$, the chromatic number of the binomial random graph $G_{n,p}$ is resilient to removal of a subgraph $H$ with $o(n \log(np))$ edges. It follows that the same assertion holds with $H$ being a subgraph of maximum degree $o(\log(np))$; and both these results are best possible up to a constant factor. We also prove corresponding results for the random $r$-regular graph $G_{n,r}$, when $r\to\infty$ and $r/n$ is bounded away from $1$, where $H$ has $o(n\log r)$ edges or maximum degree $o(\log r)$. Our results rely on a natural connection between chromatic resilience to edge removal and improper graph colouring. This work provides counterpoint to the work of Sudakov and Vu (2008) and Alon and Sudakov (2010), which treated chromatic resilience for $G_{n,p}$ subject to the addition, rather than removal, of edges.

0.25

Linear Objective Problems: A Complexity Framework for Bilevel Optimization

cs.CC cs.DMmath.COmath.OC ·2026-11-08 ·Christoph Grüne, James B. Orlin, Lasse Wulf
theory
Abstract

Many bilevel and robust optimization problems are complete for the second level of the polynomial hierarchy, but such results are usually proven one problem at a time with problem-specific reductions. A recurring obstacle is that the lower level must select an optimal solution rather than merely a feasible one, whereas standard reductions preserve only whether a solution exists, not which solutions are optimal. We introduce linear objective problems (LOP), a framework for optimization problems whose solutions are explicit subsets of a ground set and whose objective is linear in these elements. We define tight solution-embedding reductions, which map the optimal solutions of one problem element by element onto the optimal solutions of another, without having to recognize optimal solutions, which is coNP-hard in general. This gives a notion of LOP-completeness, with Satisfiability as the canonical complete problem, even with unit weights. We show that many classical problems are LOP-complete, including Vertex Cover, Dominating Set, TSP, Knapsack, and Steiner Tree, and we introduce a duality that transfers LOP-completeness between minimization and maximization problems. As our main application, we prove a meta-theorem for min-max regret optimization under interval uncertainty: for every LOP-complete problem, the min-max regret version is $Σ_2^p$-complete. Instead of a separate proof for each nominal problem, it suffices to show that the nominal problem is LOP-complete. This recovers known results for the interval min-max regret versions of Knapsack and Maximum Benefit Set Cover and yields completeness for over 50 further problems.

0.24

Tight B-Trees for Adjacency Labeling

cs.DS cs.DM ·2026-11-08 ·Noy Rotbart
Abstract · 8 pages, 2 figures

We show that planar graphs admit adjacency labels of size $\log n+O(\log\log n)$: every vertex gets a label of this many bits, and the adjacency of two vertices is decided from their labels alone, in constant time. This improves the $\log n+O(\sqrt{\log n\log\log n})$ of Dujmović et al. (J. ACM 2021) and the $\log n+O(\sqrt{\log n})$ of Gawrychowski and Janczewski (SOSA 2022). Equivalently, $n$-vertex planar graphs have an induced-universal graph with $n(\log n)^{O(1)}$ vertices, down from $n2^{O(\sqrt{\log n})}$. The same holds for graphs of bounded genus, apex-minor-free graphs and $k$-planar graphs. For every proper minor-closed class we get $\log n+O(\sqrt{\log n\log\log n})$, down from the $\log n+O((\log n)^{3/4})$ of Dujmović et al. (2026). Both results come from tight B-trees: nodes are split into parts of nearly equal size, so every node but the topmost stays within a factor $1\pm O(1/\log n)$ of its target size, and a root-to-leaf path loses only $O(\log\log n)$ bits.

0.24

Sharp Rates for Optimal Recovery of Bandlimited Functions from Localized Oversampling

cs.IT math.CA ·2026-11-08 ·Wentao Huang, Haizhang Zhang
Abstract

It is well-known that one can recover a function $f$ bandlimited to $[-δ,δ]$ ( $0<δ<π$) from its finite sample data $f(j),-n\le j\le n$, with an approximation error exponentially decaying to zero as $n$ tends to infinity. Various regularized Shannon sampling formula have been proposed to achieve such exponential approximation. So far the best rate has been $$ O\biggl(\frac1{\sqrt{n}}\exp(-(π-δ)n)\biggr). $$ We establish that the optimal recovery algorithm in the worst-case scenario achieves an approximation error $ρ_{n,δ}$ satisfying $$ c_* \frac1{\sqrt{n}}(\tan\fracδ4)^{2n+1/2}\le ρ_{n,δ}\le C^* \frac1{\sqrt{n}}(\tan\fracδ4)^{2n+1/2}, \quad 0<δ\leqπ,\quad n\geq1, $$ where $c_*,C^*$ are positive constants independent of $n$ and $δ$. This result identifies the best possible exponential rate to reconstruct bandlimited functions from two-sided finite oversampling data. In other words, $$ ρ_{n,δ} \asymp \frac{\tan\fracδ4}{\sqrt{n}}\exp\bigl(-c_δ(π-δ)n\bigr), $$ where $c_δ$ is strictly decreasing on $(0,π]$ and $$ \lim_{δ\toπ^-}c_δ=1,\ \lim_{δ\to 0^+}c_δ=+\infty. $$ We also make the new finding that one can also reconstruct $f$ bandlimited to $[-δ,δ]$ with exponentially decaying error from one-sided sample data $f(j)$, $1\le j\le n$, and the sharp rate of approximation error $ρ_{n,δ}'$ of the optimal algorithm is $$ ρ_{n,δ}' \asymp \sqrt{δ/π}\sin^n(δ/2). $$

0.23

Closing the Complexity Gap for Wheeler Graph Recognition on $d$-NFAs

cs.CC cs.FL ·2026-11-08 ·Kengo Hashimoto
theory
Abstract

Wheeler graph recognition asks whether a directed graph whose arcs are labeled by symbols from a totally ordered alphabet admits a Wheeler order, a total order on its vertices compatible with the labels of arcs and the order of their tails. A $d$-NFA is a nondeterministic finite automaton in which at most $d$ transitions with any given label leave each state. For Wheeler graph recognition on $d$-NFAs, Alanko et al. gave a polynomial-time algorithm for $d \leq 2$, whereas Gibney and Thankachan proved NP-completeness for every fixed $d \geq 5$. The computational complexity for $d = 3$ and $d = 4$ remained open. We resolve both open cases by extending the reduction of Gibney and Thankachan with a new encoding of clauses of Not-All-Equal 3-SAT. Specifically, we prove that recognition is NP-complete even for directed acyclic graphs over an ordered binary alphabet with a unique source reaching every vertex and total out-degree at most three at every vertex. Consequently, Wheeler graph recognition on $d$-NFAs is NP-complete for every fixed $d \geq 3$.

0.22

QoS-Aware Energy-Efficient Sub-THz Near-Field Multiuser Hybrid MIMO Systems Employing Circular Arrays

cs.IT cs.ET ·2026-11-08 ·Priya Gupta, Prabhanshu Yadav, Suraj Srivastava
Abstract

Near-field communications have emerged as a promising technology for future sixth-generation (6G) networks. However, designing energy-efficient hybrid beamfocusing schemes for multiuser (MU) systems remains challenging due to the non-convex quality-of-service (QoS) constraints and hardware limitations of hybrid architectures. In this paper, we investigate a hybrid massive multiple-input multiple-output (MIMO) architecture for a near-field sub-THz MU downlink system employing a large uniform circular array (UCA) at the base station. A QoS-aware transmit power minimization problem is formulated for energy-efficient beamfocusing and addressed through a two-stage penalty-driven joint optimization (PDJO) framework, where auxiliary variables are introduced to reformulate the QoS constraints into second-order cone (SOC) constraints. Furthermore, a near-field polar-domain codebook is developed via joint angle- and distance-domain sampling. We establish that the fully digital (FD) beamfocuser lies within the column space of the proposed codebook, enabling hybrid precoder design to be formulated as a multiple measurement vector (MMV)-based sparse approximation problem. The formulated problem is then solved using a low-complexity simultaneous orthogonal matching pursuit (SOMP) algorithm under hardware and transmit power constraints. Simulation results demonstrate that the proposed framework achieves spectral- and energy-efficiency performance close to the ideal FD beamfocusing scheme while providing a favorable trade-off between hardware complexity and system performance.

0.22

Algorithms for Soft Tree Containment on Tree-Child Networks

cs.DS q-bio.PE ·2026-11-08 ·Erdem Ugurlu, Louxin Zhang
Abstract · 38 pages, 8 figures

Phylogenetic networks generalize evolutionary trees by allowing lineages to merge. The Tree Containment problem asks whether a network $N$ displays a tree $T$ on the same taxa. When the polytomies of $T$ reflect uncertainty rather than simultaneous divergence, the relevant notion is soft display, in which $N$ may refine the polytomies of $T$. We take it in its asymmetric reading. Soft Tree Containment is NP-hard on reticulation-visible networks even at reticulation indegree three, and it was open for a non-binary tree and a tree-child network with multifurcating tree nodes or reticulations of indegree above two. We solve this case. A cluster characterization reduces soft display to realizing all clusters of $T$ by one switching, and in a tree-child network the vertices that can realize a cluster lie on a single chain. The per-cluster requirements then form a constraint network of width one, which arc consistency decides without search, and a $2$-SAT instance when $N$ is binary. This yields, to our knowledge, the first polynomial-time algorithm for soft Tree Containment on single-labelled tree-child networks of arbitrary reticulation indegree, and a linear-time algorithm when $T$ is binary. The same method decides the symmetric variant of the problem, in which the polytomies of $N$ may be resolved as well.

0.21

Improving the Trifference Lower Bound via Non-Perfect Hashing

cs.IT ·2026-11-08 ·Chen Yuan, Ruiqi Zhu
theory
Abstract · 18 pages

We strictly improve the Körner-Marton lower bound $\frac14\log \frac{9}{5}$ on the asymptotic rate of trifferent codes. We obtain this improvement by constructing non-perfect $(9,3)$-hash codes with rate above $\log(9/5)$ and then applying the Tetra-code construction, which gives $R_3\ge\frac14R_{9,3}$. More generally, for every $q\ge4$, we prove $R_{q,3}>B_{q,3}+\frac1{250000}$. We also extend the method to $(p,2,3)$-list recovery. For every $q\ge4$ and $0\le p<\frac{q-2}{12(q-1)}$, we prove $R_q^{\mathrm{LR}}(p)>B_q^{\mathrm{LR}}(p)$. Here $B_{q,3}$ and $B_q^{\mathrm{LR}}(p)$ denote the respective probabilistic lower bounds. For every fixed $0\le p<1/18$, the rate improvement is bounded below by a positive constant independent of $q\ge4$. For every fixed $0\le p<1/12$, we also obtain a positive rate improvement independent of $q$ for all sufficiently large $q$.

0.21

On the Complexity of Extending Prescribed Linear Forests to Hamiltonian Cycles in Hypercubes

cs.CC math.CO ·2026-11-08 ·Huazhong Lü, Tingzeng Wu
theory
Abstract

Let $Q_n$ be the $n$-dimensional hypercube, and let $P\subseteq E(Q_n)$ be a set of prescribed edges. It is known that, for $n\ge 2$, every such set $P$ with $|P|\leq 2n-3$ whose edges form a linear forest is contained in a Hamiltonian cycle of $Q_n$. This bound is sharp for general linear forests. The computational complexity of deciding whether a prescribed edge set of a hypercube can be extended to a Hamiltonian cycle was posed as an open question by Dvořák [SIAM J. Discrete Math. 19 (2005) 135--144]. We prove that the problem is NP-complete even when the prescribed edges form a spanning linear forest with no isolated vertices. We give a polynomial-time reduction from a restricted 3-SAT problem, using local spanning path-forest gadgets for variables and clauses that are uniformly embedded in the hypercube. The same reduction therefore establishes NP-hardness of two versions of Hamiltonian-cycle extension problems for general prescribed edge sets.

0.19

The Query Complexity of Envy-Free Division: Cake, Chores, and Mixed Resources

cs.GT ·2026-11-08 ·Alireza Farhadi, Mohammadtaghi Hajiaghayi
Abstract

We study the query complexity of envy-free division for cake, chores, and resources with both positive and negative values. For arbitrary nonnegative nonatomic chore costs, we give a deterministic exact protocol using $2^{n\log_2 n+O(n)}$ queries and cuts. For both cake and chores, a Las Vegas protocol uses $O(n^{12}\log(2n))$ expected queries and $O(n^{11}\log(2n))$ expected cuts. Conversely, every exact protocol with fixed positive success probability requires $Ω(n^2)$ expected queries in general, even with arbitrarily many output cuts at no query cost. For mixed cake and chores, no uniform finite expected-query bound exists even when every Cut returns the leftmost root or every Cut returns the rightmost root. This holds for any fixed positive multiplicative approximation factor and success probability, already when all net totals are one. An additive impossibility holds on the same normalized domain.

0.19

Optimal Polynomial-Query Approximation for Consistent Submodular Maximization with Short Lookahead

cs.DS cs.LGmath.OC ·2026-11-08 ·Vaneet Aggarwal
Abstract

Elements arrive one at a time, and after each arrival an algorithm holds a set of at most $k$ arrived elements. Its goal is a held set whose value under a monotone submodular function is close to the best value that $k$ arrived elements can achieve. The recourse of an update is the number of elements inserted into or removed from the held set, and we require it to be bounded independently of $k$ and of the number $n$ of arrivals. With evaluations only on sets of arrived elements and polynomially many of them, the best ratio a randomized algorithm can achieve is $2-\sqrt{2}\approx0.5858$. We show that seeing future arrivals raises the best ratio for polynomially many evaluations to $1-1/e\approx0.6321$, for deterministic and for randomized algorithms; no algorithm with polynomially many evaluations exceeds $1-1/e$, even when all elements are available at once. For every $\varepsilon>0$, a deterministic algorithm that sees the next $\lceil\varepsilon k/3\rceil$ arrivals keeps a set of value at least $1-1/e-\varepsilon$ times the optimum at every time, changes $O(\varepsilon^{-3})$ elements per arrival and makes $(n+1)^{O(\varepsilon^{-2})}$ evaluations. With the whole arrival sequence known in advance, a deterministic algorithm attains the same ratio with $O(\varepsilon^{-2})$ changes per arrival. A randomized algorithm with the same short lookahead makes $O(n^2k)$ evaluations, changes $O(\varepsilon^{-2})$ elements per arrival, and attains the same ratio in expectation at each fixed time. The main tool is a greedy algorithm with an explicit guarantee for each of its partial solutions combined with each initial segment of the known future arrivals. A second tool moves between two such solutions one element at a time while keeping the value high, and the deterministic algorithm times each move using upper bounds on the optimum from failed searches.

0.16

The Complexity of Computing a Competitive Equilibrium in Fisher Markets with Chores

cs.GT ·2026-11-08 ·Christian Kroer, Ruta Mehta, Tianlong Nan
theory
Abstract

We study the complexity of computing competitive equilibria in Fisher markets with divisible chores and positive rational disutility coefficients and earning requirements. For linear disutilities, we prove that exact equilibrium computation is CLS-complete. In particular, the problem remains CLS-hard even when the earning requirements are binary in $1$ and $2^k$, where $k$ is an integer. We prove hardness by encoding the first-order conditions of a box-constrained quadratic program in equilibrium prices and earning flows, using a compact minimum-cost flow construction. For convex constant-elasticity-of-substitution (CES) disutilities, exact equilibrium prices may be irrational. We formulate the problem of finite-precision equilibrium price computation, in which we seek rational prices with a certified market clearing error at most a given rational tolerance encoded in binary. We prove that this problem is CLS-complete for every fixed rational exponent $ρ>1$. To prove this hardness, we encode a quartic polynomial through the fourth-order expansion of the CES price potential function corresponding to the Eisenberg-Gale (EG) dual program, and show that every sufficiently accurate approximate equilibrium of the constructed market lies near uniform prices. Additionally, we also identify a tractable special case of chores Fisher markets. In particular, we provide a strongly polynomial algorithm for exact equilibrium computation in linear markets whose disutility matrix has rank at most two and earning requirements are arbitrary positive rationals. For $n$ agents and $m$ chores, the algorithm uses $O(nm(n+m))$ rational arithmetic operations and comparisons, with polynomial intermediate encoding length.

0.14

Induced-Minor-Free Graphs Admit a Star-Based Separator of Size $\tilde{O}(\sqrt{n})$

cs.DS math.CO ·2026-11-08 ·Arnold Filtser, Nikolai Karol, Hung Le, Florian Lehner
theory
Abstract

Given an $H$-induced-minor-free graph $G$ with $n$ vertices, we construct a balanced separator of $G$ consisting of $\tilde{O}(|V(H)|\sqrt{n})$ star subgraphs. Our result is an induced-minor-free analog of the famous separator theorem for $H$-minor-free graphs by Alon, Seymour, and Thomas [J. Amer. Math. Soc., 1990]. A key idea in our construction is a new decomposition of $H$-induced-minor-free graphs, analogous to the cop decomposition for $H$-minor-free graphs by Andreae [J. Combin. Theory Ser. B, 1986]. We also discuss several algorithmic applications of our result.

0.12

Welfare in Envy-free Contracts: When Agents Have Sparse Interests

cs.GT ·2026-11-08 ·Junjie Chen, Cong Chen, Takuro Yamashita
Abstract

We study welfare maximization under envy-free contracts, where a principal assigns tasks and offers success-contingent payments to heterogeneous agents with hidden effort. For unit-demand agents, we establish a sharp complexity boundary: the problem is polynomial-time solvable when each agent is interested in at most two tasks, but NP-hard with at most three interests. Our algorithm analyzes tree-structured dependencies among forced assignments and reduces optimization to maximum-weight independent set in a perfect graph. When agents may receive multiple tasks, no polynomial-time constant-factor approximation exists unless $\mathrm{P}=\mathrm{NP}$. Finally, with homogeneous success probabilities and rewards, we establish a welfare- and revenue-preserving equivalence between unit-demand contract design and envy-free pricing. This yields polynomial-time welfare maximization without restricting interests, while revenue maximization remains NP-hard.

0.11

Internal Queries for Maximal Closed Substrings

cs.DS ·2026-11-08 ·Simone Faro, Francesco Pio Marino, Arianna Pavone
theory
Abstract

Searching for repetitions inside a selected part of a text raises a subtle boundary problem: removing neighbouring characters can create new maximal answers. We study this effect for maximal closed substrings (MCSs). A string is closed if it is a single letter or has a nonempty proper prefix occurring exactly twice, as its prefix and suffix; a closed occurrence is maximal if neither available one-character extension is closed. Previous work established $O(n\log n)$ bounds on the global MCS families of a length-$n$ text and developed enumeration algorithms. Internal queries instead ask for the MCSs of a chosen substring, with maximality evaluated inside that substring. Separating boundary occurrences from interior ones gives an $O(n\log n)$-word index that reports the $c$ answers in $O(\log n+c)$ time and supports counting and longest queries in $O(\log n)$ time. Our main result reduces reporting space to $O(n)$ words, with $O((c+1)\log^{1+\varepsilon} n)$ query time and $O(n\log^{1+\varepsilon} n)$ construction time for any fixed $\varepsilon>0$. Thus, after preprocessing, reporting requires no additive scan of the queried substring, which is particularly useful for long queries with few answers. Independently, we prove that at most $3(s-1)$ MCS occurrences of any length-$s$ substring cross a fixed gap between consecutive characters, with maximality evaluated inside that substring. The same bound holds for global MCSs contained in a length-$s$ block. A family over two letters attains $\lfloor(s-1)/2\rfloor$ crossings for every $s\ge3$, showing that the linear order of this local density bound is optimal.

0.10

A Characterization of $q$-Ary $2$-Designs and a $3$-Design Obstruction for Doubly-Extended Reed-Solomon Codes

cs.IT ·2026-11-08 ·Guanghui Zhang, Zhengbang Zha
theory
Abstract · 18 pages

Let $\mathcal{C}_k$ be the doubly extended Reed--Solomon code with parameters $[q+1,k,q-k+2]_q$, where $q\ge3$ is a prime power and $2\le k\le q$. We prove a conjecture of Wu and Zhou by showing that the minimum-weight codewords of $\mathcal{C}_k$ do not form a $q$-ary $3$-design when $2\le k\le q-1$. Using shortening, we extend this obstruction to every nonempty weight class with $3\le w<q+1$. We also establish that each individual nonempty positive-weight class of $\mathcal{C}_k$ forms a $q$-ary $2$-design if and only if $\gcd(k-1,q-1)=1$. Consequently, every nonempty positive-weight class below full support has exact $q$-ary design strength $2$ when this coprimality condition holds, and $1$ otherwise.The three-design obstruction follows by applying a classical subset-product character identity to a marginal covering count for three prescribed nonzero symbols. The comparison between weights rests on a covering-count formulation of the classical MDS coset-weight relations. For every linear $[n,k,d]_q$ MDS code with $2\le k\le n-1$, this formulation shows that the $q$-ary $2$-design property at any one nonempty positive weight is equivalent to the same property at every nonempty positive weight.More generally, for $2\le t\le\min\{k,d\}$, we prove that the $q$-ary $t$-design property at any $t-1$ distinct nonempty positive weights forces it at all nonempty positive weights.

0.07

Iterative Interference Cancellation for Near-Field XL-MIMO Networks with LoS Channels

cs.IT ·2026-11-08 ·Nayyab Haider, Rodrigo C. de Lamare
Abstract · 6 pages, 2 figures

In extra-large multiple-input multiple-output (XL-MIMO) systems, near-field users experience spherical wavefront propagation, offering higher spatial resolution than far-field plane-wave models, but complicating signal detection and interference cancellation. Under line-of-sight conditions, we propose a weighted-rate successive interference cancellation (WR-SIC) scheme whose ordering metric exploits the enhanced spatial degrees of freedom of near-field propagation to prioritize users with stronger channels. An iterative detection and decoding (IDD) scheme with LDPC coding is then developed. Simulations in hybrid near- and far-field deployments show that the proposed WR-SIC and IDD schemes improve bit-error and sum rates over existing schemes

0.06

Robotics, Control & Signals 55

HARP-Bench: Benchmarking Robot Manipulation Around Active Humans

cs.RO cs.CV ·2026-11-08 ·Qinhong Zhou, Riley Zong, Chunru Lin, Diyi Hu, Shivraj Singh Bhatti +4 more
publishedbenchmarkcode/data
Abstract · Accepted at CoRL 2026

Recent robot manipulation benchmarks have enabled systematic evaluation of general-purpose robot policies across diverse objects, scenes, instructions, and embodiments. However, most existing benchmarks still assume that the robot is the only active agent and that task semantics are determined primarily by the instruction and object state. This assumption overlooks a common source of real-world variation: humans actively sharing the workspace. In such settings, humans may reveal an implicit goal, complete part of the task, temporarily change object availability, or constrain safe robot motion while working nearby. We introduce HARP-Bench, a benchmark for evaluating robot manipulation policies in workspaces shared with active humans. HARP-Bench treats human activity as part of the manipulation problem, rather than as background scene variation. It organizes human-present manipulation into four conditions: Cueing, Contribution, Interruption, and Co-presence. These conditions are instantiated through everyday manipulation tasks, including paired human-present variants for controlled comparison with standard manipulation. HARP-Bench also provides a demonstration-generation pipeline for dynamic human-present settings. We evaluate VLA models, imitation-learning methods, and heuristic policies on HARP-Bench. Human-present variations challenge current policies, and higher task success does not consistently improve safety. Quantitative failure analysis identifies manipulation and recovery as the dominant remaining challenges. Training with HARP-Bench demonstrations improves human-present performance in simulation. A real-world study on one task family shows improved performance under human contribution, but not interruption, providing initial evidence of the data's utility and identifying limits of transfer. Code is available at https://github.com/UMass-Embodied-AGI/harp.

0.62

An In-Ear Sleep Modulation SoC Featuring a CNN-LSTM Accelerator with Long-Kernel Memory Reuse and Runtime Dynamic Quantization

eess.SP ·2026-11-08 ·Yaqian Xu, Xinyu Chen, Yuhan Hou, Jianxiong Xu, Hao You +6 more
published
Abstract · Author accepted manuscript. 4 pages, including a one-page copyright and publication information cover sheet. Published in the 2026 IEEE/JSAP…

This work presents a 65 nm SoC for closed-loop in-ear sleep modulation. The SoC performs low-noise in-ear EEG sensing, CNN-LSTM-based sleep stage classification, and phase-specific auditory stimulation under 1 mW. A dedicated accelerator is developed, featuring a memory reuse technique to efficiently process long kernels required for low-frequency EEG feature extraction. A runtime dynamic quantization is incorporated, reducing model size by 3.88 times with only a 0.28% loss in accuracy. A sensitivity of 98.3% is achieved for deep sleep detection, enabling effective sleep modulation.

0.55

K-PWM: Control-Oriented Structured Bilinear World Models under Partial Observation

cs.RO cs.AIcs.LGeess.SY+1 ·2026-11-08 ·Santosh Mohan Rajkumar, Sriram Narayanan, Samuel E. Otto, Debdipta Goswami
top venue
Abstract · Accepted at NeurIPS 2026

World models learn internal simulators of environment dynamics for prediction, planning, and control, yet learning them from limited, noisy, and partial observations remains challenging. While much progress has focused on pixel-based observations, many physical systems instead provide vector-valued measurements that reveal only part of the underlying state. We introduce K-PWM, a probabilistic world model for prediction and control under partial observation. K-PWM couples Koopman-bilinear latent dynamics with a nonlinear observation decoder and performs probabilistic latent-state inference through the learned model via extended Kalman filtering and smoothing, without a separately trained inference network. Its bilinear structure enables exact propagation of the maintained Gaussian belief through the latent transition and efficient rollouts for uncertainty-aware model predictive path integral (MPPI) control. For training K-PWM, we present a generalized expectation conditional maximization (GECM) procedure with closed-form updates for the generators, noise covariances, and latent prior, complemented by gradient-based decoder learning. A Koopman-consistent initialization selects the latent dimension from data and initializes the action-dependent dynamics. With limited training data, K-PWM achieves lower long-horizon prediction error than the evaluated baselines on partially observed Gymnasium MuJoCo systems, while substantially reducing training time relative to the strongest prediction baseline. On high-dimensional MuJoCo Menagerie robots, it achieves competitive tracking accuracy with low per-step computation times, demonstrating potential for real-time deployment. These results support structured probabilistic dynamics as a data- and compute-efficient approach to world modeling and control under partial observation.

0.53

ObjectWorld: Scaling Object-Centric World Models with A Large-Scale 3D Full-Object Flow Dataset

cs.RO ·2026-11-08 ·Haonan Chen, Sicheng Isabella Lu, Haitian Li, Jiajin Wu, Hexiang Wei +7 more
dataset
Abstract

Generative world modeling from large-scale visual data offers a promising path toward scalable robot manipulation. This calls for an intermediate representation that can be supervised at scale and remains actionable for downstream control. We propose 3D full-object flow, an object-centric motion representation that assigns dense 3D trajectories to points across the complete surface of a target object, including its occluded regions. By extending visible-surface object flow to the full object geometry, this representation makes complete object motion explicit and applies uniformly to both rigid and deformable objects. To realize the potential of this representation at scale, we introduce a 1210-hour full-object flow dataset that addresses the scarcity of large-scale supervision for complete 3D object motion. The dataset integrates diverse real-world videos and physical simulations, with a scalable annotation pipeline for recovering full-object motion from real-world observations and quality metrics for systematically assessing annotation quality. We then train a world model that takes an image and a task description as input and predicts dense full-object flow over the reconstructed object geometry. Experiments show reliable annotation quality, data-scalable flow prediction, and effective downstream manipulation across novel scenes and robot embodiments. Project website with additional results and videos: https://nus-lins-lab.github.io/ObjectWorld/.

0.49

Off-Policy RL with Online Inference Tuning: My Solution to the CoRL 2026 RoboEval Challenge

cs.RO cs.CVcs.LG ·2026-11-08 ·Ilia Larchenko
code/data
Abstract · Technical report for the RoboEval Challenge at CoRL 2026. Code: https://github.com/IliaLarchenko/RoboEvalSolution ; checkpoint: https://hugg…

The CoRL 2026 RoboEval challenge has eight bimanual tasks, and it ranks a policy not only on whether it succeeds but on how: every stage of the task, done safely, efficiently and with both arms in coordination, scored against human demonstrations. My solution has two parts. The generic part, carried over from my LeHome challenge solution, is off-policy RL on a $π_{0.5}$-based policy that is also its own value function. It is trained with AWR and RECAP on advantages of the exact official score, in a loop where every rollout is at the same time a Thompson-sampling trial of about 20 inference parameters. Out of the box, it took the policy to 0.85 RDS on held-out seeds in two days (and it was still very far from convergence). The competition-specific part adds data that local exploration cannot find: generated demonstrations that reach the full reward, and recovery episodes that start from the policy's own failures. The submitted policy reaches 0.990 RDS on held-out seeds and 0.987 on the official seeds.

0.47

Tutorial: Vision-Based Formation Control with Differential-Drive Robots

cs.RO ·2026-11-08 ·Christian Marowski, Byung-Cheol Min
code/data
Abstract

This tutorial presents a step-by-step introduction to multi-robot formation control using differential- drive robots, an overhead camera, and Python. ArUco markers are used to estimate robot positions and orientations, while desired pairwise displacement vectors define a circular formation. By comparing measured and desired relative positions, the controller computes movement vectors that are converted into differential-drive wheel commands, with basic proximity-sensor-based obstacle avoidance. The tutorial guides readers through hardware and software setup, robot pose estimation, formation geometry, feedback control, system integration, and formation-error evaluation. Motivated by the need to bridge the gap between theoretical understanding and practical implementation in robotics education, we emphasize how mathematical concepts are translated into coordinated behaviors on physical robots. Although the implementation uses a specific robotic platform, the underlying perception and formation-control principles can be adapted to other differential-drive robots and simulation environments. The accompanying open-source GitHub repository provides the Python implementation, setup instructions, and demonstrations to support hands-on learning and further exploration of multi-robot systems.

0.42

Streaming Failure Forecasting: VLA Failure Monitors That Keep Learning After Deployment

cs.RO cs.LG ·2026-11-08 ·Amit Jadhav, Shaurya Beriwala, David Q. Liu, Yanfei Liu
publishedcode/data
Abstract · 25 pages, 13 figures, 9 tables. A short version was accepted at the Sim2Real2Field and EEAI workshops at CoRL 2026. Project page: https://sf…

Failure monitors for vision-language-action (VLA) policies are trained offline and then frozen. After deployment, lighting, objects and cameras drift, and the policy itself is updated or moved to new tasks. On recorded real-robot runs, after a switch of task or policy version, a frozen monitor fit on the old task ranks failures little better than chance. We present Streaming Failure Forecasting (SFF), a monitor that keeps learning from the one label every episode provides: whether it succeeded. SFF keeps the offline monitor's network fixed and refits only its last linear layer after each episode, at 2.18 ms per policy call on a Jetson CPU core. Its alarm threshold moves only after successes, which in simulation kept false alarms near target except when two drifts came at once. With every setting locked before testing, SFF ranks failures better than the same monitor kept frozen on 4 of 5 simulated test suites under unannounced drift, and above resetting on detected drift on every suite. Carried with its locked learner settings to 434 recorded runs of four VLAs on real SO-101 arms, replayed across unannounced switches of task or policy version, it ranks failures better than the frozen monitor for every VLA and, in an exploratory comparison, above a full reset at the true switch. A deployed failure monitor need not stay frozen: refitting its readout from outcomes is cheap and helps after the changes we tested.

0.39

Enhancing Balance and Posture in Material Handling Tasks with a Wearable Robotic Tail

cs.RO ·2026-11-08 ·Eisa Anwar, Sajeeva Abeywardena, Stuart C. Miller, Ildar Farkhatdinov
Abstract · 16 Pages, 5 Figures

Frequent heavy lifting is linked to musculoskeletal strain, with lower back pain affecting up to 40% of manual labourers. Inspired by animals using tails for balance, we developed a wearable robotic tail as a dynamic counterbalance. Ten participants performed an imitated shelf loading/unloading task across three conditions: active tail, passive tail, and no tail. Kinematics, centre of pressure, and muscle activation (gastrocnemius, erector spinae, tibialis anterior, biceps brachii) were measured. Compared with no tail, the active tail achieved >20 mm reduction in centre of pressure peak displacement, 4.3 degree reduction in trunk motion, and >9% and >20% reductions in erector spinae and gastrocnemius activity, respectively. The passive tail produced smaller reductions, suggesting passive counterbalancing helps while active movement augments it further. Biceps brachii activation increased with the active versus passive tail. The ankle kinematics and tibialis anterior activation were unchanged. These findings support wearable robotic tails for improving balance during material handling.

0.39

Kinora: A Versioned Workflow for Text-to-Motion Generation, Robot Retargeting, and Interactive Validation

cs.RO ·2026-11-08 ·Fangcheng Jin
code/data
Abstract · 10 pages, 1 figure, 3 tables. Technical report on Kinora server 0.5.1 and Gadget 0.15.2. Code and release artifacts: https://github.com/Poir…

Motion generation tools produce candidate trajectories, but interactive robot authoring also requires recoverable execution, traceable edits, explicit geometry, and model-specific validation. We present Kinora, a system that connects text-to-motion generation, a motion library, local agent tools, robot retargeting, and trajectory refinement through versioned artifacts. The current release uses an independently specified 52-joint authoring rig and procedural skeleton previews, separating the distributed human geometry from externally installed generation models and robot assets. Human-motion edits preserve frames outside a selected interval, while robot edits create immutable candidates whose reports govern acceptance and export. We evaluate server 0.5.1 and Gadget 0.15.2 with 156 core regression test executions, 49 Gadget executions, four CUDA generation/editing cases, a 30-frame G1 retargeting check, and a 3,600-frame synthetic G1 fixture paired with a suspended negative control. All four CUDA cases passed their numerical gates; the turning splice nevertheless retained a 14.3 cm support-foot drift proxy warning. The G1 positive fixture completed bounded simulation replay, whereas the suspended control was quarantined before replay. These results support selected integration and preservation contracts. They do not establish semantic-quality improvements, population-level feasibility, or hardware safety.

0.36

A Multistage Extension of Witsenhausen's Counterexample with Linear Signaling

eess.SY math.OC ·2026-11-08 ·Quanyan Zhu, Tamer Başar
benchmark
Abstract · 8 pages, 3 figures

We study a multistage scalar extension of Witsenhausen's counterexample in which controller 1 observes the initial state exactly, whereas every subsequent controller observes only a local noise-corrupted state and does not inherit its predecessors' observations, so that every nonterminal action both regulates the state and shapes the signal available to the next controller. Restricting the signaling controllers to linear policies while leaving the terminal controller unrestricted, we compute the exact value within this class and reduce the optimization to a deterministic dynamic program in a single propagated variance. The optimal intermediate gains are shown to have a shrinkage form, and the value functions of that program are concave. For two stages, the benchmark admits a closed form on a matching locus, where for small effort weights it has two distinct minimizers, and at the standard parameter instance on that locus a two-point nonlinear policy strictly improves upon it. For three stages, numerical continuation at a reference instance reveals three distinct local minima, and two values of the initial variance at which the lowest of the three changes abruptly. Under the terminal-cost normalization we further prove, for every finite horizon, that the optimal linear value becomes asymptotically equal to the signaling-idle benchmark as the initial variance tends to infinity.

0.35

GFM Control Automatic Compliance Toolbox

eess.SY ·2026-11-08 ·Tejas Kunbi, Rahul Rane, Dan Wu, Aleksandra Lekić
Abstract

The increasing penetration of converter-interfaced generation reduces system inertia, short-circuit strength, and voltage support. European grid codes, including the ENTSO-E Network Code on Requirements for Generators and the ACER Demand Connection Code, now define disturbance-based validation procedures for grid-forming (GFM) generation and converter-based demand. Commercial RMS tools, however, lack an integrated workflow aligned with these recent requirements. This paper presents an automated Python-PowerFactory compliance toolbox and an RMS test bench for a co-located system comprising a 1 MW GFM photovoltaic (PV) plant, a 0.3 MW GFM battery energy storage system (BESS), and a 1 MW converter-interfaced dynamic load at a 20 kV point of common coupling fed from a 110 kV grid. Extended virtual synchronous machine models and a DCC-oriented load model are tested in batch, with results automatically extracted and plotted for assessment. Results show compliant RMS performance for the slower, electromechanical tests, while the paper identifies which tests require complementary electromagnetic transient (EMT) verification.

0.34

Grid-forming Control of Offshore Wind through a Diode-Rectifier HVDC System: a 2 GW Case Study

eess.SY ·2026-11-08 ·Ishi Nigam, Saif Alsarayreh, Tuanku Badzlin Hashfi, Miloš Cvetković, Joris van Doorn +1 more
Abstract

Diode-rectifier-unit (DRU)-based high-voltage direct current (HVDC) transmission considerably reduces the volume and weight of the offshore converter platform for remote wind farms and, hence, its cost, but shifts the task of forming and regulating the offshore AC network to the wind-turbine converters. This paper examines which grid-forming control strategy is the most suitable for DRU power transfer. The DRU commutation relationship shows that the delivered active power depends on the offshore voltage magnitude (but not on the converter voltage angle), so the angle-based active-power loop of virtual synchronous machine control is not closed through the rectifier. A $P-V/Q-ω$ grid-forming architecture, which regulates active power via the voltage magnitude and derives frequency from the reactive power error, is therefore adopted. Electromagnetic transient simulations of a 1GWaggregated grid-forming turbine on a 66 kV collector network with a $\pm$525 kV bipolar DRU link show that the commanded power is delivered over 800 to 1000MW with the offshore voltage held near nominal. The paper also examines how grid-forming operation changes the allocation of control duties within the Type-4 turbine: once the grid-side converter forms the network, DC link regulation must move to the machine-side converter, which limits the achievable bandwidth of the DC-link control. Extending the system to 2 GW with two 500 MW grid-following units achieves full power transfer in simulation, but with the grid-forming converter loaded beyond its continuous rating and a steady-state offshore frequency deviation of approximately 1.8 Hz.

0.34

Analysis increments for model bias diagnosis and correction: Theoretical foundations

eess.SY physics.ao-phstat.ME ·2026-11-08 ·Eviatar Bach
Abstract · 15 pages, 5 figures

A common method for diagnosing and correcting model bias in weather and climate prediction is based on analysis increments: the difference between the analysis and the forecast at each data assimilation cycle. Their time average is used to estimate and correct the bias, and machine learning is increasingly used to map the forecast state to the increments, giving a state-dependent correction. Despite its widespread use, the method lacks theoretical foundations. We first analyse a Kalman filter whose forecast model has a constant bias. We show what the averaged increment measures, and prove that iterating the correction removes all of the bias seen through the observations; for linear dynamics, no method can identify the rest. We then extend the analysis to state-dependent biases. For linear dynamics, each increment records the model error over the filter's memory, as seen through the observations. We use this to determine what the increments measure when averaged over a long period, averaged over periods with similar slowly varying conditions, and used without averaging, as in machine-learned corrections. Numerical experiments confirm the theory for a linear model and suggest that its main conclusions carry over to the nonlinear Lorenz '96 model, including for a correction learned by a neural network.

0.34

Scenario-Based Assessment of Transport Electrification for Low-Carbon Urban Planning: A Case Study of Phu Quoc, Vietnam

eess.SY ·2026-11-08 ·Tien-Khai Nguyen
Abstract

Urban net-zero strategies require a defensible emissions boundary and quantitative estimates of the contribution achievable through individual mitigation measures. This study assesses road-transport electrification within a planned residential development in Phu Quoc, Vietnam, using the assumptions and scenario data of an urban-planning study originally prepared in 2022. A vehicle-class inventory combines population-dependent fleet projections, assumed travel activity, gasoline and diesel consumption, fuel-specific carbon dioxide (CO_2) factors, and electricity-related emissions from battery-electric vehicles. The reference case projects annual road-transport emissions of 59.41, 73.86, and 87.23 tCO_2 for 2022, 2025, and 2030, respectively. Three year-specific substitution scenarios, corresponding to nominal electric-vehicle shares of 25%, 50%, and 100%, yield source-reported avoided emissions of 9.43, 13.64, and 55.41 tCO_2/year, or 15.9%, 18.5%, and 63.5% of their respective baselines. However, a cross-scenario audit identifies an inconsistency in the 2025 mixed-fleet results under time-invariant vehicle-class emission-intensity ratios: under that additional assumption, 2025 avoided emissions would be 23.45 rather than 13.64 tCO_2/year. The original detailed vehicle-energy input tables are not included in the available source document, preventing definitive reconciliation. The study therefore provides a transparent, preliminary planning assessment rather than evidence of citywide net-zero attainment. Its principal value is a reproducible accounting structure and an explicit identification of the data, electricity-system, and infrastructure constraints that must be resolved before policy or investment decisions.

0.34

Karhunen-Loève Transforms of Autoregressive Windows Based on Matched Symmetry Groups

eess.SP cs.ITmath.NA ·2026-11-08 ·Mitchell A. Thornton
theoryunusual
Abstract · v1: 38 pages, 7 figures

The Karhunen-Loève transform of a window of a stationary autoregressive process of any order $p$ is described through its poles, extending the Algebraic Diversity framework, which selects a transform by the matched group. For simple roots every basis vector is a sum of at most $p$ sinusoids, at the frequencies where the spectrum equals the eigenvalue, plus modes decaying from the ends. The window covariance differs from a member of the class diagonalized by the type-II discrete cosine transform by a closed-form matrix of rank $2p$, exactly when no pole is zero, so the eigenvalues interlace spectrum samples on the cosine grid, and the coding-gain loss of the cosine transform is an exact function of that matrix with computable bounds: for the $8\times8$ Markov block of image coding at pole $0.95$ it is below 1.2 percent. By Kronecker's theorem the rank is at most $2\max(p,q)$ for rational spectra; a moving-average window departs in its Hankel corners, and an integrated (ARIMA) window departs by low rank from the type-VII sine class, exact for the random walk. The exact symmetry is generically the reflection and contains a wreath product for interleaved windows; the dihedral group of the even extension is matched approximately, the dominant pole's sign choosing cosine or sine. The two groups give the transform a fast form: a parity fold, half-length cosine cores and a cascade of at most $p$ first-order Cauchy stages, exactly $p$ for a minimal AR($p$) model. The first-order case recovers results obtained by other means.

0.33

Floquet-Parameterized Koopman-MPC for Tracking Periodic Orbits

cs.RO eess.SY ·2026-11-08 ·Kartik A. Pant, Vishnu Vijay, Yufei Zhu, Xintong Deng, Taewoo Nam +1 more
Abstract · 6 pages, 2 figures

Designing real-time control for nonlinear systems to track periodic orbits can be challenging, especially for complex and coupled dynamical systems (e.g., tethered kites), where the true model parameters are hard to obtain. To address this, we propose a physics-informed Koopman learning approach that incorporates periodic-orbit stability directly into the learned linear predictor. The approach combines the underlying periodic structure of the dynamics along with periodic state-input trajectory data to learn physically consistent linear predictors (in lifted/observable space). We leverage Floquet theory, which studies the stability of periodic orbits in nonlinear systems, to obtain the periodic stability information of a given orbit via a monodromy matrix (i.e., the state transition matrix over an entire period). Subsequently, we impose open-loop monodromy constraints to learn Floquet-parameterized linear predictors using a least-squares Extended Dynamic Mode Decomposition (EDMD) formulation, thereby constraining the search space and aiding in Koopman learning. We then utilize these linear predictors to design a model predictive control (MPC) law for tracking periodic references. Finally, we evaluate the designed framework on a cartpole system under periodic forcing and compare it against a purely data-driven baseline.

0.32

Compute Regeneration through Re-Productization: Reorganizing Retired Edge AI Silicon into a Delivered Multilingual Service Robot

cs.RO cs.AR ·2026-11-08 ·Hongyu Luo, Sun Tie, John Wang
code/data
Abstract · Code and data: https://github.com/gravity-link/gedge_bot_paper (release v1.0.0); archived at https://doi.org/10.5281/zenodo.24152092. v2 rep…

Useful edge hardware can lose product support while retaining substantial computation. We treat this as systems design: compute regeneration here means re-productization, retaining the compute module while redesigning the surrounding product organization. Our case is GEdgeBot, a multilingual service robot built around a retired 32 GB Jetson AGX Xavier. A source-grounded inventory shows the box alone supplies inference, while the combined system exposes five on-device AI services, a seven-language speech chain, three sensing domains, and one deterministic seven-byte control boundary. AI and orchestration stay on the Xavier; finite motor-pulse generation on an STM32F407. Audits cover the six-mode control state machine, readiness barriers and sound orientation. Historical single-prototype measurements report 1,782 MB GPU memory with YOLO, Whisper-small, and NLLB-200 resident, a direct-control p99 deviation rising from 0.24 to 49.93 ms under concurrent AI load, and six of six trigger-to-follow demonstrations at 4.35 s mean latency, with conversational voice unavailable. A November campaign restores an isolated inference stack, re-measures AI workloads, and repeats control-frame runs. Driving two execution organisations with the same 6,000 measured 50 ms frame intervals shows the finite-pulse boundary removes host jitter from the pulse train but not the command start phase; its stopping bound is host-enforced, while the MCU 16-bit pulse field is unclamped. A six-check static audit exposes a host--MCU queue mismatch, and a 14-scenario replay executes the audited main() dispatcher across six modes with zero physical device opens and 54 contract-checked frames. Results support an existence claim and a reproducible architectural hypothesis, not fleet-level reliability. The 208 direct-control ratio remains unreplicated; MCU waveform capture and on-site fault injection remain open.

0.32

Predictive Neuromorphic Control of a Scanning Mirror System: An Event-Centric Framework

eess.SY math.OC ·2026-11-08 ·Changrui Liu, Robin Wroblowski, Anil Alan, Shengling Shi, Rodolphe Sepulchre +1 more
Abstract · 8 pages, submitted to ECC27, this work is the first, foundational work that proposes Event-Centric Model Predictive Control (EC-MPC), a fram…

Organizing control around events can reduce sensing and actuation demands while directly linking control decisions to the occurrence and timing of desired system behaviors. This paper proposes a novel event-centric model predictive control (MPC) framework that unifies event sensing, event prediction, and event regulation. Events define the feedback information, prediction horizon, and event-timing objectives, placing them at the center of predictive control design. The framework is developed for a scanning mirror system equipped with a baseline neuromorphic controller composed of spiking neurons that supports sparse actuation. The control objective is to achieve prescribed zero-crossing times and a desired scanning range while minimizing the mechanical energy transferred between the actuator and the mirror. A sampling-based algorithm is presented to obtain suboptimal solutions of the event-centric MPC optimization problem without requiring an explicit event-to-event prediction model. By grounding predictive optimization in events, the proposed framework opens a new direction for MPC design and its integration with neuromorphic control.

0.32

Nonlinear Interference Estimation in Coherent Optical Systems: Principles and Applications

eess.SP ·2026-11-08 ·Dario Pilori
Abstract · 10 pages, 7 figures. Extended version of an invited talk at ECOC 2026. Submitted to the Journal of Lightwave Technology

Accurate estimation of nonlinear interference (NLI) is important for monitoring and optimizing coherent optical networks. This invited paper reviews the main techniques proposed for receiver-side NLI estimation, with particular focus on longitudinal power monitoring (LPM). We discuss the principles, advantages and limitations of LPM-based NLI estimation, including recent experimental demonstrations. We then experimentally investigate the impact of the spatial step size and the number of averaged constellation blocks on estimation accuracy. The results show that accurate NLI estimation does not necessarily require a highly accurate power profile, with limited improvements obtained by increasing the number of averaged blocks. Finally, we discuss the main practical limitations and applications of NLI estimation, with particular attention to end-to-end and span-wise optical power optimization.

0.31

Brain-Aligned Latent Actions: Imposing Premotor Cortex Geometry on a Robot's Action Vocabulary

cs.RO cs.LG ·2026-11-08 ·Amit Jadhav
Abstract · 16 pages, 3 figures, 6 tables. Robot manipulation results (CALVIN) will be added in a later version

Robot policies increasingly act through a small vocabulary of discrete latent actions learned from unlabeled video, yet training leaves open which motion receives which code. We find this choice highly unstable: two independently trained runs of Moto's latent motion tokenizer assign the same robot clips to nearly unrelated codes (adjusted Rand index 0.001), and evaluating one tokenizer in bf16 instead of fp32 reassigns 74% of its tokens. We use this freedom to impose the representational geometry of macaque ventral premotor cortex (PMv), recorded while a monkey drew shapes, on the tokenizer's code assignment, with a straight-through similarity loss trained alongside its reconstruction objective on robot video. In one training run per target, on five shape categories the loss never saw, the PMv-trained tokenizer reaches a representational similarity of 0.52 with PMv, against 0.20-0.26 for released and no-prior tokenizers and 0.42 for the 95th percentile of random kinematic encoders. Kinematic, language, and random targets do not transfer (-0.17 to -0.03). Two results limit the claim: a hand-picked kinematic descriptor, the stroke's initial direction, matches PMv better (0.65), and a label-shuffled PMv control scores 0.41. We specify the downstream manipulation test, on a CALVIN pipeline that reproduces Moto's published results, before running it.

0.30

ScanDreamer: A Map-Grounded World Model for Quadrotor Navigation

cs.RO ·2026-11-08 ·Yunyi Gao, Yixing Shi, Yize Liu, Hai-Duong Tran, Nhat-Hoang P. Nguyen +6 more
code/data
Abstract · 8 pages, 6 figures, 3 tables. Submitted to AICAP 2027; decision pending

Quadrotors inspecting buildings and tunnels must navigate unknown clutter with a forward-facing depth camera whose limited field of view constrains scene understanding and collision-avoiding navigation. Recent world-model agents learn navigation through imagined experience (e.g., AirDreamer), but remain data-intensive and often lack explicit geometric memory and awareness of what the camera can observe. We propose ScanDreamer, a map-grounded world model that constructs an online 2D scan-memory map from depth images and vehicle pose. Our world model, actor, and critic are conditioned on this map, which also defines a geodesic progress reward. Moreover, a forward-cone penalty discourages horizontal motion outside the camera's viewing direction. During imagination, ScanDreamer propagates the quadrotor pose over a frozen map and derives progress and wall-contact signals from the mapped geometry. We train ScanDreamer for 0.8 million environment steps (5.5 h on one GPU) and evaluate it in NVIDIA Isaac Sim over 1560 episodes spanning unseen layouts and parameter ranges. Across three seeds (mean $\pm$ SD), ScanDreamer achieves 96.3 $\pm$ 0.3% success with 3.6 $\pm$ 0.3% collision, compared with 59.6 $\pm$ 3.0% success and 28.8 $\pm$ 2.8% collision for AirDreamer at the same online-interaction budget. Removing the forward-cone penalty increases collisions to 9.9 $\pm$ 3.1%. These results show that combining explicit map memory, map-grounded imagination, and perception-aware motion substantially improves navigation performance and data efficiency. The source code is available at https://github.com/WANGSHUAIzzzz/ScanDreamer.

0.29

Numerical Stability Issues and a Mitigation Method in Switching Events Calculation Considering State Variable Discontinuities for EMT Simulation

eess.SY ·2026-11-08 ·Sheng Lei
Abstract · Copyright may be transferred without notice, after which this version may no longer be accessible

Some existing methods for switching events calculation considering state variable discontinuities for EMT simulation can encounter numerical stability issues, either over-stabilizing or destabilizing the studied power systems. This paper documents these issues in detail with a concrete example. The underlying mechanism of the issues are revealed via systematic analysis. A mitigation method for the issues is proposed. The convergence behavior of the method is analyzed. Numerical examples demonstrate the effectiveness of the method. It is observed that the method also helps to improve the accuracy of the simulation results.

0.29

Analysis of the Effect of Doppler on Direct-to-Cell: Real-World Performance and Proposed Improvements

eess.SP cs.IT ·2026-11-08 ·Yuxi Chen, Muhammad Iqbal Rochman, Kyler McCormack, Jakob Kotz, Monisha Ghosh
Abstract

Direct-to-cell (DtC) satellites serve unmodified LTE phones today from low Earth orbit, where the satellite's motion creates Doppler shifts of tens of kHz. Since large offsets such as these cannot be compensated adequately by the user equipment (UE), the satellite pre-compensates the Doppler shift of each beam, but it can do so for only one reference point in the beam. A UE elsewhere in the beam still has to compensate a residual offset; previous field studies report residuals of up to 10 kHz, beyond the $\pm$7.5 kHz that the 15 kHz subcarrier spacing can resolve. In this paper we corroborate this effect with our own detailed measurements of Starlink transmissions and further correlate ineffective handover performance to inadequate residual offset compensation. We analyze the Doppler compensation problem with detailed standard-specific simulations and propose a compensation method at the UE, based on the location of the UE with respect to the beam reference point used for the Doppler compensation and the known satellite location, that improves the efficiency of handover. With UE cell-specific pre-compensation, the neighbors are identified, read correctly and handed over to in time rather than lost to radio-link failure.

0.29

From Festival Video to the Stage: A Sim-to-Real Ladder for Long-Horizon Expressive Whole-Body Dance on a Commercial Humanoid

cs.RO ·2026-11-08 ·Hesham Shehata
Abstract · 19 pages, 12 figures. Four supplementary videos are linked in the manuscript

We present a pipeline for transferring Awa Odori, a traditional Japanese festival dance, from monocular video to a commercial Unitree G1 EDU humanoid. The pipeline combines human mesh recovery, kinematic retargeting, reinforcement learning for motion tracking in simulation, and onboard deployment via ONNX. Training ran on a single RTX 5070 GPU. An eleven-version deployment ladder evaluated on hardware examines three interventions. We detect and repair silent corruption of deployment artifacts, model state-estimator drift during training, and edit the reference motion to support stable execution and a controlled ending. The final policies achieved three consecutive rope-free completions of a 43-second routine, a 46-second performance with automatic handoff, and a continuous 9.7-minute rope-free performance composed from the captured motion. In a live handover, a single remote request let the robot start from and return to the manufacturer's walking controller on stage, with a measured torque-free window of 45-140 ms. We report the training dynamics and hardware outcomes and discuss failure boundaries involving compliant flooring and occlusion in monocular capture.

0.29

Ultrawideband Polarization-Rotating Antenna for Enhanced Ground Penetrating Radar Detection

eess.SY ·2026-11-08 ·Ming Li, Yuki Gao, Shu-Lin Chen, Hai-Han Sun, Y. Jay Guo
unusual
Abstract

Ground penetrating radar (GPR) represents a powerful and widely adopted tool for nondestructive subsurface detection and imaging. To overcome the impaired detection performance caused by polarization mismatch in conventional single-channel GPR systems, this paper presents an innovative ultrawideband polarization-rotating (UPR) antenna that achieves rotating linear polarizations (LPs) by varying the source frequency for enhanced detection. The developed UPR antenna achieves a continuous polarization rotation from $0^\circ$ to $170^\circ$ across 1.2 to 3.2 GHz, as verified by both full-wave simulations and experimental measurements. The antenna is subsequently integrated into a single-channel stepped-frequency GPR system and evaluated against targets of varying geometries in a sandy environment. Compared with a conventional fixed-LP GPR system, the UPR antenna-enabled GPR system delivers markedly more consistent detection performance across all target types, with particularly pronounced enhancements for polarization-sensitive targets. By enabling ultrawideband frequency-dependent polarization agility, the proposed GPR system enhances subsurface target detection while maintaining the simplicity and low cost of a single-channel measurement architecture.

0.29

Descriptor: UWB Anchor Self-Calibration Dataset (UWB-ASC)

eess.SP ·2026-11-08 ·Paul Schwarzbach
survey
Abstract

Anchor self-calibration estimates the positions of fixed nodes of an indoor radio positioning or localization system from observations among the nodes and to mobile devices. It replaces the survey of every anchor, which is costly and goes stale when anchors move. The UWB Anchor Self-Calibration Dataset (UWB-ASC) is a testing ground for such methods, recorded in one industrial hall with ultra-wideband (UWB) two-way ranging. It has three parts. The first is two measured deployments. In the line-of-sight (LOS) deployment, 10 anchors range with a walking tag tracked by motion capture. In the non-line-of-sight (NLOS) deployment, 12 nodes range among obstructions, and one of them roves over surveyed markers in 20 quasi-static sessions. Both deployments also hold static sessions, and a total station surveyed every fixed node. The second part is a seeded range simulator that draws synthetic ranges on the surveyed geometry with adjustable error components, from noise-free ranges to NLOS errors, for benchmarks at controlled error levels. Its 81 frozen draws in 5 conditions serve as the reference. The third part is a set of readers that convert the ranges among the static nodes of two public UWB datasets, TWR-CLOUD-Bern and selfloc, into the same format without redistribution. All parts share open Parquet and YAML files, in which topology views select the ranges for each class of calibration method. Raw recordings with rebuild scripts, quality reports, and field labels of the line of sight complete the dataset.

0.28

Spot Markets for AI Inference: Dynamic Capacity Allocation and Pricing on a Shared Accelerator Fleet

eess.SY ·2026-11-08 ·Quanyan Zhu
Abstract · 39 pages, 12 figures, 4 tables

Inference providers size their accelerator fleets for the interactive peak and sell spare capacity through fixed discounts, such as batch tiers at half price. However, inference differs from conventional cloud capacity in four respects: capacity depends on the latency tier through batching, moving weights between models is slow and costly, spot demand can wait until its deadline, and power can bind. In this paper, we develop a framework for dynamic capacity allocation and pricing of AI inference on a shared accelerator fleet. We consider a provider that serves several models under a power cap and sells each model in an interactive tier with a speed promise and in a spot tier for preemptible work with a deadline. In each period, the provider chooses how many replicas of each model to keep loaded and what spot price to quote. We show that one interactive token displaces a fixed number $ρ_m>1$ of spot tokens, and that the optimal spot price lies midway between the interactive price and the effective cost of capacity in the deadline window, which is close to the prevailing half price only when the fleet is slack. We further show that a spot tier raises the optimal reliability of the interactive tier, and that placement should follow persistent demand shifts while prices absorb transients. Building on these results, we design a model predictive controller based on a linear program with explicit approximation bounds, which solves each period in milliseconds. Numerical experiments on a simulated week with a measured daily demand profile, a flash crowd, and a grid curtailment show that the controller raises profit by 7.4% over fixed half-price practice, captures 93% of the gain available with perfect forecasts, keeps the interactive promise through both events, and leaves spot buyers better off.

0.28

Kimodo-Policy: Learning Kinematic Action Models for Humanoid Loco-Manipulation

cs.RO cs.AIcs.CV ·2026-11-08 ·Yunheng Wang, Jiaxi Zhang, Wei Lu, Yuetong Fang, Taowen Wang +10 more
code/data
Abstract · 23 pages, 14 figures, 4 tables. Project page: https://kimodo-policy.github.io/

Humanoid loco-manipulation requires continuous coordination of high-dimensional whole-body motion to realize task objectives. While vision-language-action models and world action models transfer semantic and visual-dynamics priors to robot control, these priors do not necessarily capture the structure of robot motion. We introduce Kinematic Action Models (KAMs), a paradigm that treats whole-body motion as a direct pretraining target and uses a pretrained motion generator as the primary action-generation backbone. KAMs ground learned temporal and cross-body coordination priors through task conditioning, visual observations, and execution feedback. We present Kimodo-Policy as an initial instantiation, combining a frozen Kimodo text-to-motion backbone with lightweight ControlNet-based visual conditioning. A root-body-hand hierarchy coordinates locomotion and manipulation, while motion-history conditioning and repeated action-chunk prediction bridge motion synthesis and closed-loop control. Following large-scale loco-manipulation pretraining, the policy achieves the highest overall success among compared policies on multi-task HumanoidArena (72.14%), competitive performance on SIMPLE, and 60% mean success across eight real-world tasks, with 678M parameters (92M trainable). Ablations support the contribution of the pretrained motion backbone. Together, these results provide initial evidence for motion generation as a complementary foundation for humanoid action learning.

0.28

Local Certificate Repair for Asynchronous Controller Updates in Recommender Systems

eess.SY math.OC ·2026-11-08 ·Simone Mariano, Paolo Frasca
Abstract

We study how to adapt an existing distributed recommendation controller when local dynamics or objective weights change. Starting from a quadratic stability and cost certificate, we construct positive semidefinite corrections with finite spatial support and an explicit residual. A joint admission test allocates the available dissipation margin to these residuals and to interactions among changes and corrections, including pending updates. The test covers complete declared parameter ranges, allowing different operators to progress independently. For a fixed compatible batch, the corrections define a common storage that guarantees exponential stability and bounds the cost accumulated throughout deployment. When the admitted collection evolves, a reservation protocol preserves the certificate and a rate bound on storage increases ensures exponential convergence. Regional linear-quadratic regulator (LQR) design proposes new gain rows for the operators' own users, and a separate residual-based calculation can certify that the final gain lowers the expected cost. A recommendation-network example illustrates the tradeoff between repair depth, data access, and the benefits of deploying an update.

0.27

Radiation-Pattern-Based Channel Estimation for Reconfigurable Pixel Antennas

eess.SP ·2026-11-08 ·Daniel Dinis, Leo M. Lahti, Pasi Ylä-Oijala, Risto Wichman
Abstract

Reconfigurable pixel antennas provide spatial diversity through multiple radiation patterns generated by a single antenna aperture, but exploiting this diversity can require substantial pilot overhead for exhaustive channel estimation. This paper jointly considers beam-steerable pixel-antenna design and its communication-side channel estimation. A pixel antenna is designed to generate diverse radiation patterns, whose electromagnetic correlation is then exploited to interpolate unobserved channel responses from a limited number of measurements. The proposed approach is evaluated in terms of SNR gain and NMSE, demonstrating that effective pattern selection can be achieved with significantly reduced channel-observation overhead and can even outperform exhaustive scanning at low SNR.

0.27

Breaking survivorship bias in driving world models

cs.RO ·2026-11-08 ·Zihao Sheng, Zilin Huang, Yuhao Luo, Zhengyang Wan, Boyue Wang +3 more
Abstract · Project Website: https://zihaosheng.github.io/NWM/

World models allow autonomous vehicles to imagine the consequences of candidate actions before execution, but their reliability depends on the futures represented in their training data. We identify a survivorship bias in driving world models: because the futures recorded in prevailing driving datasets almost always end safely, existing models often imagine benign outcomes even when a collision is imminent. We introduce a neutral world model (NWM) that learns both routine and safety-critical futures from a mixture of routine driving and real accident videos with a bounding-box-guided objective. The NWM more faithfully anticipates imminent collisions, while maintaining a low false collision rate and competitive generation quality in routine driving. We further use the model as an imagination engine for decision-making. Given multiple ego trajectories, the NWM predicts the future associated with each candidate, and a reward function evaluates these imagined outcomes before selecting an action. On challenging NAVSIM scenarios, this procedure selects the better trajectory more reliably than the baseline world model. Zero-shot experiments with real-world pedestrian dummy crossing data collected using an autonomous research vehicle further demonstrate transfer to unseen sensor observations. These results show that correcting survivorship bias enables world models to more faithfully reason about futures and support safety-critical decision-making in autonomous driving.

0.26

A Distributed Observer with Random Agent Activations over Random Communication Networks

eess.SY ·2026-11-08 ·Yong Joo Do, Jin Gyu Lee
Abstract

This paper develops a distributed observer for the forward-Euler discretization of a continuous-time linear time-invariant system under random local operations and random communications. Under uniform stochastic strong connectivity, we establish almost sure exponential convergence of all local state estimates for sufficiently small discretization step sizes. We further examine how the underlying parameters and components affect the admissible range of the discretization step size and the guaranteed convergence rate bound. Unlike related approaches, the proposed method does not require each local unobservable subspace to admit an $A$-invariant complement.

0.26

Qubit-to-Motion: Traceable Action Provenance in Hybrid QRL for Robotic Bin-Packing Sequencing

cs.RO cs.AI ·2026-11-08 ·Hassen Nigatu, Howard Li
code/data
Abstract

Hybrid quantum reinforcement learning has been explored for robotic decision-making, but evaluations often emphasize task success, return, or convergence, leaving unresolved whether quantum circuit outputs influence constrained actions and remain traceable through robot execution. We present \texttt{Qubit-to-Motion}, an evidence-centered framework for support-constrained sequencing and simulated robotic placement. Four Pauli-$Z$ readouts provide the sole learned input to the actor and critic, while a support mask enforces admissible object selection. Each decision is recorded and linked via step indices and SHA-256 fingerprints to a sealed route, deterministic Cartesian and inverse-kinematics waypoints, prescribed joint motion, and MuJoCo contact replay. A frozen checkpoint is evaluated across five deterministic 30-object scene variants. All 150 placements satisfy support, containment, and settling criteria. A fixed-state zero-readout intervention changes $90.0\pm0.0\%$ of decisions, demonstrating dependence on the joint circuit representation. Mean macro travel is $34.537\pm0.206$~m, inference requires $8.60\pm1.27$~ms per route, and replay achieves $8.764\pm0.220~μ$m mean final position error with 0.356~mm maximum interpenetration. These results establish reproducible decision-to-contact traceability. The implementation is available at \href{https://anonymous.4open.science/r/Qubit-to-Motion-Quantum-Reinforcement-Learning-F7DC}{\textcolor{linkpink}{the anonymous repository}}.

0.26

Synergizing CKM and ISAC for Environment-Aware Wireless Networks

eess.SP ·2026-11-08 ·Juncong Zhou, Yilong Chen, Yuan Guo, Zixiang Ren, Han Hu +2 more
survey
Abstract

Channel knowledge map (CKM) and integrated sensing and communications (ISAC) have emerged as two key technologies for sixth-generation (6G) and beyond wireless networks, enabling a new paradigm toward environment-aware system design. CKMs provide static or stochastic wireless environment information a priori by learning location-dependent mappings from transceiver positions to channel knowledge of various types, whereas ISAC exploits wireless signals to sense environmental objects in real time. CKM and ISAC have primarily been studied separately in prior works, while their integration offers a promising approach for enhancing the environment awareness of 6G networks and further improving both sensing and communication capabilities. On the one hand, sensing can support CKM construction by providing location-tagged measurements and richer environmental observations, while also enabling real-time CKM refinement in dynamic environments. On the other hand, CKM can enhance sensing by offering prior environmental information for clutter mitigation, and CKM and sensing information can be jointly leveraged to improve communications. This article provides an overview of the complementary interactions between CKMs and ISAC, with particular emphasis on three design approaches: sensing-assisted CKM construction and refinement, CKM-assisted sensing, and joint CKM- and sensing-enhanced communications. We also discuss future research directions for achieving environment-aware wireless networks.

0.26

ColNav: Uncertainty-Aware Spacetime Collision Checking with Bounding Volume Hierarchies

cs.RO ·2026-11-08 ·Sanjeev Ramkumar Sudha, Lars Ivar Hatledal, Erlend M. Coates
Abstract

Safe navigation for mobile robots in dynamic and uncertain environments requires planning paths that avoid collisions with other moving targets. Typically this involves forward propagating the robot motion and checking for collisions at sampled points in space and time. Such methods can miss collisions between sampled points and are not robust to uncertainty in the state estimates. We propose ColNav, a robust collision check formulation with geometric volumes. The key aspect of our approach is encoding the uncertainty in the targets' and the ego robot's positions and velocities into spacetime tubes, in two variants that consume bounded-set and Gaussian uncertainty descriptions, respectively. The tubes are organised in bounding volume hierarchies, so that a single tree traversal covers the entire horizon, rather than testing sampled time instants, and returns a robust geometric certificate together with an estimate of the time to collision. Static and dynamic obstacles share the same representation. Our approach reaches the goal with a 100% success rate under prediction model mismatch, where the best baseline reaches 65%, and maintains up to three times the clearance in dense traffic. We also validate real-time operation on a physical robot.

0.26

Feasibility-Aware Learning for Network-Constrained Unit Commitment

eess.SY ·2026-11-08 ·Hien Thanh Doan, Yeji Lim, Hongseok Kim
Abstract

Learning-based approaches to network-constrained unit commitment (NCUC) usually still rely on an optimization solver to repair or complete their predictions, which limits the speedup they can deliver. This paper proposes FALCON-UC, a feasibility-aware graph learning framework that predicts commitment and dispatch jointly and produces final schedules without solving a mixed-integer program at inference time. It combines an LSTM--GraphSAGE predictor over a fully connected generator graph, a differentiable correction that embeds commitment-dependent dispatch bounds, power balance, and PTDF-based branch flows in training, and deterministic commitment, dispatch, and ramp-coupled reserve repairs at inference. On the 2383-bus system, FALCON-UC achieves $99.09\%$ commitment accuracy and a $0.4350\%$ mean absolute cost gap with all evaluated constraints within tolerance at $0.0562$ s per instance, a $2.27\times10^{3}$ speedup over the solver. MTLR R-SCUC-FL is slightly more accurate but still needs an optimization solver to complete each schedule, which takes $262.2$ s per instance, $4.67\times10^{3}$ times the FALCON-UC runtime. A label-only baseline without repair violates the balance and branch limits on every 2383-bus test instance, and FALCON-UC also meets all evaluated constraint tolerances on a 1354-bus system.

0.26

Learning the NMPC with Stability Guarantees

eess.SY ·2026-11-08 ·Shuyou Yu, Yan Sun, Jun Ai, Qifang Liu, Hong Chen
published
Abstract · 13 pages, 7 figures, this paper was not presented at any IFAC conference

Although nonlinear model predictive control (NMPC) is widely used, its online optimization is computationally expensive. Neural network approximations of NMPC can effectively reduce the online computational burden but generally lack stability guarantees. This paper proposes an approximate NMPC framework with guarantees of constraint satisfaction and closed-loop stability. The approximation error is addressed through constraint tightening, and the resulting tightened NMPC is used to generate the training dataset of the approximate NMPC. The maximum allowable approximation error is prescribed a priori for the design of tightened NMPC and satisfied in a probabilistic sense through an iterative trial-and-error procedure alternating between network training and a posteriori verification. Finally, the effectiveness of the proposed framework is demonstrated through a numerical example of a constrained nonlinear system.

0.25

Single-Radar Bilateral 3D Near-Field Imaging for Mobile Robots

cs.RO cs.CV ·2026-11-08 ·Yingqi Wang, Tingwei Chen, Zhongqin Wang, Min Xu
Abstract

Mobile robots operating in homes need reliable 3D perception of nearby furniture and surrounding structures, yet recovering this information on both sides of the robot's path with a single forward-mounted millimeter-wave radar remains challenging because of near-field wavefront curvature, viewpoint-dependent target visibility, and limited elevation diversity. This paper presents a motion-assisted framework for reconstructing structures on both sides of a robot's trajectory. Our near-field model relates scene reflectivity to stepped-frequency multiple-input multiple-output (MIMO) observations across frequencies and robot positions using spherical paths from transmitter (Tx) to scatterer to receiver (Rx). Bilateral Tx combining provides wideband directional weighting, while sliding-aperture back-projection coherently focuses the measurements using near-field propagation distances. Top-K power fusion combines local images with different visibility. The reconstruction provides lateral-forward support for multi-layer height extraction, followed by guarded adaptive height refinement using the original MIMO channels. Experiments in controlled, office/study, and corridor environments produce mean lateral-forward localization errors of 4.65, 5.63, and 5.98 cm, respectively, with peak-to-background ratios (PBRs) exceeding 13 dB. The framework recovers bilateral 3D scattering structure from one forward pass using a single radar, without relying on learned representations of the scene.

0.25

A Physical Testbed for Self-Driving Laboratories: Comparing Bayesian and Agentic Optimization in a Minimal Closed-Loop Color Formulation Task

cs.RO cs.AI ·2026-11-08 ·Tony Qin, Ziheng Wu, Maximilian Schommer, Roya Amini-Naieni, Aldair Gongora
Abstract

Self-driving laboratories (SDLs) integrate laboratory robotics and automation with AI-driven decision-making to close the design-build-test-learn (DBTL) cycle. However, physical testbeds that combine off-the-shelf hardware and AI into a functioning closed-loop workflow remain in early stages of development, and their integration with commercial laboratory automation is comparatively less developed. In this work, we develop a physical testbed that assembles and integrates commercially available laboratory automation instruments and demonstrate continuous, closed-loop, autonomous operation over 24 cumulative hours of unattended experimentation. We validate this platform on a minimal closed-loop discovery task in which the system autonomously formulates aqueous dye mixtures to match a target absorbance spectrum. The system completed six independent campaigns, each consisting of 48 closed-loop iterations over approximately four hours. In every campaign, the system autonomously selected, dispensed, and measured each mixture, with an AI-driven decision-making algorithm determining subsequent formulations without human intervention. Using this testbed, we compared the performance of two AI-driven decision-making algorithms, traditional Bayesian optimization (BO) and an agentic approach, in selecting mixtures under a fixed total-volume constraint. The agentic optimizer matched target absorbance spectra using fewer iterations than BO, with reductions up to 52%, and converged on a solution for one target that BO failed to match within 48 iterations. These results demonstrate a reference physical platform for SDLs that can evaluate both established and emerging AI strategies, including large language model (LLM)-directed optimization, before their deployment in real-world applications where experiments may be costly and time-consuming to prepare.

0.25

WiFo-R: Wireless Foundation Model with Recurrent Channel Representation

eess.SP ·2026-11-08 ·Liu Xuanyu, Cao Keyi, Cheng Xiang
Abstract

Wireless foundation models learn transferable channel state information (CSI) representations across heterogeneous wireless systems. However, conventional grid-aligned representations couple feature sequence length to CSI dimensions, restricting the allocation of representation capacity to different propagation conditions and task requirements while increasing inference overhead. This paper proposes WiFo-R, a wireless foundation model that learns progressive channel representations with selectable capacity. A channel-query cross-attention encoder aggregates spatial, temporal, and spectral CSI information, while parameter-shared recurrent updates refine existing channel features and introduce additional latent tokens. Multi-mask, multi-depth self-supervised pretraining equips intermediate states with strong reconstruction capabilities, followed by task-aligned post-training for channel estimation and CSI feedback. WiFo-R supports temporal and frequency-domain channel prediction, channel estimation, and variable-rate CSI feedback, and transfers to localization and LoS/NLoS classification. Experiments across in-distribution and out-of-distribution channel configurations demonstrate accurate reconstruction, progressive performance improvements, and strong transferability. A single feedback encoder--decoder supports six payload lengths at fixed quantization precision, while cross-attention and selectable recurrent depth improve the accuracy--complexity trade-off for large-scale CSI observations.

0.23

DeepSoMo: Deformable Object Motion Estimation under Proprioceptive Partial Observability via Predictive Latent Dynamics

cs.RO ·2026-11-08 ·Hanwen Zhao, Weicheng Dai, Luoyao Chen, Chen Feng
position
Abstract

Estimating the motion of deformable bodies in soft robotics from proprioceptive sensing is challenging due to partial observability, as internal observations provide only indirect access to global motion, contact, and shape evolution. In such settings, instantaneous observations are insufficient, and accurate estimation requires temporally consistent latent representations. This paper studies deformable motion estimation under proprioceptive partial observability from a latent dynamics perspective. We argue that a key requirement for long-horizon estimation is latent temporal consistency, where latent states remain predictive of future observations. To this end, we present DeepSoMo, a learning-based framework that enforces predictive structure in latent dynamics through a self-supervised latent prediction constraint. We evaluate the approach in a canonical deformable dynamics setting using high-fidelity finite-element simulation as a controlled source of ground truth. Results demonstrate consistent improvements in trajectory estimation, orientation recovery, shape reconstruction, and contact inference over architecture-matched recurrent baselines.

0.22

DexForesight: Grounding Dexterous Policies with Implicit Spatial Foresight

cs.RO ·2026-11-08 ·Zhongxi Chen, Yi Yang, Xinyi Chen, Mingzhe Ye, Yaping Li +4 more
code/datacommunity pick
Abstract · 13 pages, 4 figures

Dexterous manipulation requires rich spatial foresight over both the surrounding scene and the robot hand. Existing world-action models often acquire such predictive knowledge through explicit future prediction. We introduce DexForesight, a framework that learns spatial foresight implicitly by injecting dynamics-aware world representations into the latent space of a dexterous policy. During training, DexForesight learns an action-conditioned spatial world model from offline scene depth, kinematic hand geometry, and demonstration trajectories. The world model predicts 3D hand-point motion and visual feature changes, capturing both articulated hand dynamics and scene changes under demonstrated actions, including the motion of fingers occluded from the cameras. We then freeze the world model and align policy queries with its predictive world queries, transferring the learned spatial dynamics into policy latents and using them to condition action generation. At deployment, only the policy is retained, operating on RGB observations, language, and robot state without depth input or explicit future prediction. On 11 DexJoCo simulation tasks, DexForesight improves the overall success rate of the base policy from 49.4% to 60.5%. On three real-robot dexterous manipulation tasks, DexForesight achieves 58.3% success, compared with 50.0% for the strongest baseline. Project page https://dexforesight.github.io/

0.21

MOSU2: Long-range Navigation System with Multi-modal Scene Understanding

cs.RO ·2026-11-08 ·Jing Liang, Zhenyu Hou, Senthurbavan Kirubaharan, Md Safwan Mondal, Ashok Gudivada +2 more
Abstract

Robots that travel long distances through public outdoor spaces must do more than avoid obstacles: they must remain on terrain suited to their mobility and decide when to proceed, yield, or detour around pedestrians, cyclists, vehicles, traffic signals, and temporary restrictions. These decisions are difficult because their correctness often depends on how a scene evolves, such as whether an approaching car is slowing or a cyclist will cross the robot's path. Vision-language models (VLMs) can interpret such cues, but they reason over images that lack metric geometry and motion. Supplying this evidence at every step is too slow for real-time navigation, and hand-crafted triggers cannot anticipate the open-ended variety of outdoor situations. We present MOSU2, a long-range navigation system with two-stage trajectory selection, in which first-stage contextual reasoning determines when additional evidence is needed and what to retrieve. Only when an interaction remains unresolved does the first stage trigger the second stage to retrieve geometric and dynamic information and align it with scene context to make navigation decisions. Building on MOSU, MOSU2 follows routes from GPS-based route planning and filters candidate trajectories through multi-modal traversability analysis before selecting a trajectory to move the robot. In real-world experiments spanning stop gestures, construction sites, crosswalks, and roads shared with vehicles, cyclists, and pedestrians, MOSU2 raises mean navigation success from 61.8% to 94.2% and condition-detection accuracy from 72.2% to 91.4% over its single-image predecessor, at about 2.2 times its mean decision latency. Project website: https://jinggm.github.io/mosu2/

0.19

Beyond Memory Limits: A Hierarchical Spatio-Temporal Map Representation for Large-Scale 3D Autonomous Exploration

cs.RO ·2026-11-08 ·Yizhe Zhang, Jianping Li, Yilun Gao, Liangliang Yin, Pengfei Wan +1 more
Abstract · 16 pages, 12 figures, preprint submitted to Elsevier

Autonomous UAV exploration is increasingly used for large-scale surveying and inspection in inaccessible or disaster-affected environments, where onboard memory and computational resources are often severely constrained. Most existing UAV exploration systems maintain globally queryable geometric map representations for motion planning and exploration decision-making, causing memory consumption and spatial-query costs to increase with the explored area. Robot-centric local mapping bounds dense-map memory and computation but discards global task context needed for long-horizon exploration.We present FLAVOR, a hierarchical spatio-temporal framework that confines dense metric geometry to a robot-centric local map while retaining compact global observation and task context. At the observation level, lightweight LiDAR-keyframe visibility records preserve ray-wise evidence for frontier reasoning. At the task level, unresolved frontiers beyond the local-map extent are retained as compact pending tasks associated with odometry anchors. A sparse topological graph connects these anchors through retained routes for future revisits.Upon returning to a task region, FLAVOR reconstructs the frontier and exploration viewpoint from observations, while the topological route is instantiated and validated against the current local map before execution. Dense-map capacity remains fixed as exploration expands, with exploration-dependent memory growth confined to compact observation and task context.FLAVOR achieves 31.98--41.88% shorter exploration time and a 94--99% smaller map memory footprint than the best baseline that completes all three simulated scenes. In real-world deployments, it reduces this footprint by 92.67% in the forest and 62.42% at the construction site relative to an explicit point-cloud ikd-tree while preserving long-term task continuity under constrained onboard resources.

0.17

Energy-Harvesting-Aware Cell-Free LEO Satellite Networks: A Lyapunov Framework for Sustainable User-Centric Beamforming and Clustering

eess.SP ·2026-11-08 ·Wali Ullah Khan
Abstract · 16, 10

We study battery-aware downlink control in user-centric cell-free low Earth orbit satellite networks with intermittent solar harvesting and finite onboard batteries. A perturbed Lyapunov controller coordinates clustering and hybrid beamforming using measured battery states and causal harvest forecasts. The per-slot solver combines group-sparse association, rate bounds, a quadratic transform, and successive convex approximation. We establish robust battery invariance for feasible implemented actions and give a utility bound conditional on a comparison policy and the per-slot implementation gap. Multi-orbit control-layer simulations show that queue-aware scheduling preserves a larger battery reserve in the stressed scenario at the expense of effective energy efficiency. A separate fixed-cluster matrix snapshot illustrates recovered hybrid beam patterns and the per-slot objective trace.

0.17

Control-Aware Command Generation for Asynchronous Control and Action Inference with Vision-Based Policies towards Fast Execution

cs.RO ·2026-11-08 ·Takayuki Osa, Kazuki Ota
published
Abstract · 8th Robot Learning Workshop: Is Physical AI Going Zero-Shot? at NeurIPS 2026

Vision-based control policies, including vision-language-action models, have enabled robots to perform increasingly complex tasks. However, their execution speed is often limited by the high computational cost of action inference. In this work, we investigate what can be achieved at the control level without modifying the underlying policy learning process. We introduce an asynchronous inference and control framework that enables fast task execution while maintaining smooth and stable robot motions. By formulating the generation of continuous target positions from discontinuous model predictions as a tracking control problem, our method makes asynchronous inference practical for real robots. We further accelerate execution through adaptive action skipping based on the variance of predicted positions. Experiments with Action Chunking Transformer and OpenVLA-OFT on a bimanual robot demonstrate substantial speedups while preserving high task success rates.

0.16

Concurrent Learning-Based Inertial Parameter Estimation and Control of Quadrotors

eess.SY cs.RO ·2026-11-08 ·Sandesh Thapa, He Bai
theory
Abstract · 6 pages, 4 figures. To be submitted to IEEE Control Systems Letters (L-CSS) with the CDC 2027 option

In this letter, we consider trajectory tracking control and inertial parameter estimation of a quadrotor UAV whose center of mass (CoM) does not coincide with its geometric center. Onboard payloads like an IMU, a camera, or a battery are not necessarily placed at its geometric center. Moreover, during the flight these sensors might move because of external disturbances and cause a CoM offset. Thus, these events and equipment introduce an offset and cause the inertia matrix to be non-diagonal. We develop an integral concurrent learning (ICL)-based geometric controller that estimates the mass, the CoM offset, and the full inertia matrix online, without an accelerometer or numerical derivatives, and tracks a given trajectory corrected by the offset estimated online. We prove that the estimated parameters converge to their true values under a finite excitation condition. We validate the effectiveness of the proposed algorithm using a simulation example.

0.16

FlowEval: How Much Adaptation Does an Embodied Foundation Model Need at Deployment?

cs.RO ·2026-11-08 ·Zhang Xi-Jia, Roman Yakunin, Shivang Chopra, Chengyue Huang, Che Chen +2 more
Abstract

Large-scale pretraining is reducing the amount of adaptation required to deploy embodied foundation models, but it is not guaranteed that a pretrained model already provide the level of performance required in every deployment. Practitioners therefore need to understand how well a pretrained model matches a target deployment and how much adaptation data is left to be collected. We study whether a small set of successful target-domain demonstrations can reveal whether and how much target-domain data will be required before any adaptation. We introduce FlowEval, a set of training-free metrics that use the flow-matching objective to probe an embodied foundation model with a set of demonstrations. FlowEval quantifies predictive disagreement on successful target demonstrations to assess how well a pretrained model fits a target deployment. Across four embodied foundation model families: Pi-0.5, MolmoAct2, GR00T, and DreamZero, FlowEval distinguishes seen from unseen deployments across real-world and simulated manipulation. Further, by finetuning Pi-0.5 and GR00T on 17 RoboCasa365 manipulation tasks, we showed that FlowEval correlates with how success rate improves as additional adaptation data is provided. Together, these results suggest that FlowEval can serve as a lightweight, training-free diagnostic of how much model--deployment adaptation may be required.

0.14

PA-SaR: Perception-Aware Learning, Control, and Coordination for Multi-Robot Search and Rescue

cs.RO cs.MAeess.SY ·2026-11-08 ·Jokūbas Dimša, Maria Grouev, Filip Keerberg, Nicolas Laplagne, Tigo Schutgens +2 more
Abstract

Autonomous SaR requires robots to explore unknown or partially known environments, avoid hazards, coordinate team activities, and guide discovered victims to safety. Improving individual capabilities of robots does not necessarily improve mission outcomes: safer motion can restrict the exploration behavior, faster discovery can increase the rescue demand, and effective guidance depends on victim responses. We introduce PA-SaR, a mathematical and architectural framework for coordinating heterogeneous local control, learning-based exploration, supervisory task allocation, and adaptive approach-lead victim guidance around mission-level objectives. PA- SaR distinguishes environmental ground-truth, robot-specific perceptions, and simulated victim behavior. This provides a common basis for integrating complementary mechanisms and analyzing their interactions across mission stages. We demonstrate the framework through single- and multi-robot simulation studies incorporating model-predictive and learning-based exploration, perception-based safety shielding, event-driven task allocation, and adaptive victim guidance. The results show trade-offs between collision avoidance and search completion, indicate that supervisory allocation improves search completion beyond information sharing alone, and demonstrate that increased exploration does not necessarily yield more rescues. Adaptive guidance increases rescues relative to continuous leading, without a detected improvement in mean or final victim trust or a reduction in mortality. These findings support evaluating local control, team coordination, and evacuation as related but distinct aspects of autonomous SaR within the considered simulation environments.

0.14

Scalable Neural Safety Certification via Monotonicity

eess.SY cs.LGmath.OC ·2026-11-08 ·Amirreza Alavi, Majid Zamani, Saber Jafarpour
Abstract

Learning-based safety certification offers a promising route for verifying complex dynamical systems when explicit models are unavailable, but existing data-driven methods often scale poorly: they treat the system as a black box, rely on dense discretizations or Lipschitz bounds, and consequently suffer from sample complexity that grows exponentially with dimension. This paper shows that a common structural property of dynamical systems--monotonicity--can be used to break this curse of dimensionality. We develop a data-driven framework for robust safety verification and safe controller synthesis for unknown monotone systems using only simulator queries. Our main theoretical result proves that, for monotone systems with upper-closed unsafe sets, safety is completely characterized by the existence of monotone inductive barrier certificates. This structure reduces certificate verification over continuous state spaces to localized boundary checks over finitely many cells. To learn such certificates, we introduce Max-Monotone Networks, a monotone neural architecture with universal approximation guarantees for monotone functions. We then propose a training and refinement algorithm that, upon successful termination, returns formally valid neural barrier certificates and, under mild partition assumptions, requires only $O(n)$ simulator queries. Across high-dimensional benchmarks, including oscillator networks with up to $13{,}659$ states and traffic networks with $1{,}000$ states, our method certifies safety in regimes where prior approaches either do not converge or exceed memory limits. These results demonstrate that exploiting order structure can make neural safety certification both scalable and formally sound.

0.13

Large-Scale Hierarchical Decision-Making in Robot Control via ADMM

cs.RO math.OC ·2026-11-08 ·Kai Pfeiffer, Jia Wang, Gordon Owusu Boateng, Yuquan Wang, Vincent Bonnet +1 more
Abstract

Sparse hierarchical nonlinear programming (SH-NLP) based on the $\ell_0$-norm provides a powerful framework for integrated, efficient hierarchical decision-making in robot kinematic control, for example in the presence of a large set of possible end-effector candidate locations. While Sequential Sparse Hierarchical Quadratic Programming (S-SHQP) with interior-point method based SHQP sub-problem solvers exploit the hierarchical, sparse structure of such problems, they are less amenable to effective warm-starting between subsequent, coherent control steps. In this work, we formulate an Alternating Direction Method of Multipliers (ADMM) to efficiently solve related SHQP sub-problems. We introduce efficient substitutions for auxiliary variables representing $\ell_0$-norm surrogates. The solver efficiently handles separable decision functions by evaluating linear algebra expressions in the appropriate order. Results in large-scale kinematic and dynamic control show reduced computation times of up to 23.5% compared with a structure-exploiting interior-point method in low- to moderate-accuracy regimes. Furthermore, hierarchical decision-making is extended to whole-body dynamic control for humanoid walking, where the need for high-accuracy solutions is highlighted.

0.13

CAFE: Closed-Loop Autonomous Task Formulation and Execution without Instance-Specific Instructions

cs.RO ·2026-11-08 ·Mingjia Chen, Xuewen Zhao, Philip Dames, Qi Mao, Fei Xie +1 more
Abstract

Scene-driven robot autonomy requires selecting tasks from observations as well as executing them. We study scene-driven task formulation and execution under predefined cleaning, safety, and storage/organization categories, without instance-specific goals or action sequences. CAFE maintains online memory of objects, functional zones, and execution states. A rule-guided Trigger-LLM admits formulation from accumulated evidence; Draft, Complete, and Audit propose goals, resolve execution details, and check dependencies. Outcomes update memory for correction and subsequent discovery. Evaluation in 12 AI2-THOR scenes compares three backbones and task-complexity tiers using coverage, constraint, and execution metrics. Tri-LLM records higher GCR scores than the evaluated baselines across three backbones and task-complexity tiers. Controls examine completion, auditing, zones, feedback, triggering, and prompt examples. LimoCobot traces illustrate storage, cleanup, and recovery from a failed grasp. The study examines how shared scene state connects task discovery to execution and outcome-driven revision.

0.12

When Coding Agents Get Arms: Safety Risks Across the Digital-Physical Boundary

cs.RO cs.CRcs.MM ·2026-11-08 ·Yiming Sun, Mi Zhang
benchmark
Abstract · Technical report, work in progress

Advances in spatial understanding have enabled frontier models to power coding agents that control a real robot arm zero-shot and end-to-end, without task-specific demonstrations or retraining. This brings two new safety challenges. First, the extensive knowledge of frontier models lets these agents perform tasks that earlier VLA and WAM paradigms could not, opening the door to new forms of capability misuse. Second, coding agents can access digital resources such as web pages and emails. Threats in these resources, such as indirect prompt injection, can carry over into the physical world when the agent controls a robot. We therefore introduce ArmedAgent, the first systematic benchmark to evaluate both risks for coding agents performing zero-shot robot control. It covers robustness, malicious use, and indirect injection across 36 tasks in 5 scenarios, evaluated in simulation and validated on a physical robot arm. A distinctive portion of the tasks target harmful behaviors that earlier VLA paradigms could not perform. Testing several frontier models, including GPT-6-Astra, under widely used agent frameworks, we find substantial safety risks in current systems.

0.12

HINT-VLM: Humanoid Interactive Navigation over Rough Terrain with Embodiment-Grounded Vision-Language Model

cs.RO ·2026-11-08 ·Ziwon Yoon, Seok Hwan Kim, Yiyu Chen, Quan Nguyen, Lu Gan +1 more
Abstract

Humanoid robots have achieved significant progress in locomotion and manipulation, enabling traversal over rough terrain and richer interaction with the environment, such as moving obstacles. However, most traditional navigation methods assume non-interactive environments and focus only on traversal. While Navigation Among Movable Obstacles (NAMO) approaches relax this assumption, prior efforts have largely focused on flat environments and reactive interaction, rather than predicting object movability and planning ahead. These limitations become critical in outdoor rough-terrain scenarios, where robots must jointly reason about terrain traversability and object movability. In this work, we address NAMO for humanoid robots in cluttered outdoor rough terrain by proposing a hybrid planning framework that combines semantic priors from a Vision-Language Model (VLM) with embodied robot interaction data through a Gaussian process (GP) residual model. The GP corrects the VLM-derived movability prior to produce embodiment-grounded probabilistic movability estimates, which are then incorporated into hierarchical uncertainty-aware planning. The proposed approach allows humanoids to jointly leverage locomotion and manipulation for efficient and safe navigation, representing the first solution to NAMO in rough terrain through predictive movability reasoning and probabilistic VLM integration.

0.11

Review the Episodes, Refine the Lessons: Simple Memory Evolution for Embodied Agents

cs.RO cs.AIcs.LG ·2026-11-08 ·Yifei Yao, Yude Zou, Xing Gao, Guanbo Jiao, Yukai Wang +3 more
benchmark
Abstract

Recent studies of embodied agents based on frontier multimodal large language models (MLLMs) develop harness engineering around the MLLM, such as harnesses with post-trained policies and recursive self-improvement loops. Yet whether these additions exceed the native capability of the MLLM remains unexplored. We introduce SimplEvo to explore how far an MLLM can improve under direct control when it retains its execution experience in self-written text memories. It requires no weight updates, post-trained policies, or privileged information. The MLLM writes a coordinate-free task memory once from a single demonstration. Under a strict train-test protocol, it then creates a bounded working memory on training episodes only and refines it after each episode with lessons on procedure, completion, and recovery. Both memories stay fixed at test time. On EBench, SimplEvo lifts the success rate (SR) of GPT-6-Astra under direct control from 46.7% to 81.7%, and tops the RoboDojo leaderboard with 69.63% SR and a score of 75.32. Compared with memories that also store past visual frames, SimplEvo reaches a similar SR with text memory alone while using fewer tokens. With its memories fixed, SimplEvo also generalizes to unseen layouts and randomized scene configurations. Moreover, the working memory transfers across models: written by GPT-6-Astra, it can be reused as is by weaker MLLMs and improve their performance without any adaptation of their own.

0.09

People & Society 20

Exploring Mixed-Reality for Enhancing Driver Warning Systems: A Preliminary Study on Attention-Shifting Methods and Hazard Perception

cs.HC ·2026-11-08 ·Shih-Yu Ma, Nolan Robert Brady, Xu Han, Neng-Hao Yu, Tom Yeh
published
Abstract · 8 pages, 3 figures

Modern vehicles are equipped with advanced sensors and capabilities, yet engaging human drivers effectively in hazard perception remains a challenging research area. This paper investigates the potential of mixed reality (MR) to enhance driver warning systems. The study examines the impact of attention-shifting methods, including visual cues, audio cues, and a combination of both, on drivers' hazard perception. A preliminary study involving six participants was conducted, and the NASA-TLX analysis did not yield significant differences. However, through the use of functional near-infrared spectroscopy (fNIRS), we unveil distinct brain activation patterns associated with visual and sound cues. Further research with larger sample sizes and diverse driving scenarios is required to validate and expand upon these preliminary results.

0.64

From Sketch to Design: A Cross-scale Workflow for Procedural Generative Urban Design

cs.HC cs.GR ·2026-11-08 ·Jin Gao, Sayjel Vijay Patel
published
Abstract · 7 pages, 6 figures. Author version of a paper published at CAADRIA 2024

Urban designers use sketches and symbols to express spatial intentions, while many generative urban design tools rely on numeric parameters that are difficult to relate to this visual process. We propose a cross-scale workflow and a web-based prototype that translates symbolic sketches into controllable procedural urban generation. A vocabulary of shapes, informed by examples of urban design diagrams and sketching exercises with graduate urban design students, encodes attributes such as network structure, density, and program. The prototype uses the \$1 Unistroke Recognizer to map gestures to symbols and translates their locations and sizes into tensor-field attractors and weighted parameter grids. Grid and radial tensor fields guide street-network generation, while the parameter grids control building heights and programs. Implemented in JavaScript and TypeScript with React and Three.js, the prototype supports interactive sketching, visualization, and urban massing generation. We describe the methodology and illustrate prototype interactions, then discuss downstream integration with architectural-scale generative tools for further refinement and analysis. The work connects designers' symbolic input with an explicit procedural generation process.

0.56

Who Benefits from Wikipedia's Growth Mentorship? The Role of Newcomer Information Needs

cs.SI cs.HC ·2026-11-08 ·Yubo Zhou, Pengyu Chen, Ceren Budak
published
Abstract · 25 pages, 10 figures. Accepted at ICWSM 2027

Evaluations of Wikipedia's programs for new editors usually ask whether a program works on average; we ask which newcomers benefit when their question to a mentor gets a reply. English Wikipedia's Growth mentorship system assigns every new account a volunteer mentor, and about a quarter of the questions newcomers send never get a reply. Using the first questions of 34,073 newcomers (2021 to 2026), we compare newcomers who received a reply with newcomers who did not, within groups that are similar on 164 characteristics recorded before the question (propensity score stratification). (1) Does a reply help newcomers keep editing? Yes, in the short term: in the 14 days after the reply, newcomers who received one were 1.8 percentage points (pp) more likely to edit an article (22% vs. an estimated 20% without a reply) and 1.8pp more likely to edit on two or more days. Later in the first year, the only difference that remains significant after correction for multiple testing is in the share who edit an article at least once between six months and one year after the reply (+1.5pp). (2) Does the effect depend on what the newcomer asked? Yes. We sort questions by Morrison's (1993) information-seeking types, which earlier studies measured mostly through self-reports, using a language model checked against human labels. A reply has significant effects for questions about rules and norms (+3.4pp), questions that show the newcomer's earlier editing (+3.4pp) and specific how-to questions (+2.4pp). For newcomers who do not know what to work on, a reply has no detectable effect, and the effect is 4.2pp smaller than for newcomers with other questions. (3) Which replies are followed by more editing? Among newcomers who received a reply, those whose reply came sooner or stayed closer to the topic of their question edited more.

0.45

TALK: One-Shot Batch Design for Protein Variant Effect Prediction

cs.CE cs.LG ·2026-11-08 ·Qiao Huang, Ao Shen, Jie Du, Manning Wang
top venuecode/data
Abstract · Accepted at NeurIPS 2026 (Oral)

Protein variant effect prediction is central to understanding protein function and guiding protein engineering, yet experimental labels remain costly. In many protein experiments, the variants that can be assayed are dictated by protocol constraints, while the variants that need accurate prediction may lie in a different or broader region of the landscape. This assay-prediction mismatch makes candidate-local criteria such as predicted function, uncertainty, or exploration value insufficient for measurement design. Given a protocol-defined accessible space and a limited budget, we ask which variants should be assayed so that the resulting data best improves landscape prediction. We propose TALK, which estimates each candidate's contribution through posterior coupling between accessible and target variants and reduces redundant information within the batch. Across retrospective ProteinGym and FLIP protocols, TALK generally improves target-space prediction under limited measurement budgets. These results point to a practical two-stage workflow: first, spend a limited assay budget on variants selected to be informative for the target landscape and train a predictive model from the measured labels; then use this model to score and prioritize high-potential variants. Code is available at https://github.com/joehuang618/TALK.

0.44

What Changed in Engineering Writing After Generative AI Chat Assistants: Evidence from 1.8 Million Articles

cs.DL cs.CL ·2026-11-08 ·Taner Cokyasar
code/data
Abstract · 26 pages, 5 figures, 7 tables. Code and data: https://doi.org/10.5281/zenodo.24028663

This study asks what changed in engineering articles after conversational language models became publicly available on November 30, 2022, and develops methods for measuring such change in a field without a uniform abstract index. It assembled 1,839,962 articles published from 2015 to 2025 in 256 engineering journals, tracked twelve measures of wording, style, and article structure, and compared each with the projection of its own pre-2023 trend. Building the monthly series showed that records dated only by year fall on January 1 and create false January peaks unless removed. In abstracts, 31 words associated with language model output appeared 82% more often than projected (55.4 against 27.0 per 10,000 words in 2025), while four control words of the same formal register fell 16%. Words that soften a claim, such as may, fell 13%, words that strengthen one, such as clearly, rose 20%, and long words rose 11%. Author counts, reference lists, page counts, and publication volume stayed within 4% of trend. The change appears in all fourteen subject categories, and false release dates, dropping any one journal or word, and word lists chosen from the data support it. A random sample of 12,866 full texts from the same corpus, including subscription articles and weighted back to it, shows the same words rising 99% in article bodies. Statements disclosing AI writing assistance, measured apart from mentions of language models, went from none to 3.7% in 2025, four in five of them at one publisher. The design is descriptive, not causal.

0.34

Evolutionary games on networks with negative edge-based environmental feedback

cs.SI physics.soc-phq-bio.PE ·2026-11-08 ·Katherine Betz, Feng Fu, Naoki Masuda
published
Abstract · 8 figures

Environmental feedback can fundamentally alter the game payoff structure underlying the evolution of cooperation. How feedback-evolving games interact with population structure under non-weak selection remains incompletely understood. We study an evolutionary game on networks in which each edge has a binary environmental state that determines the payoff matrix used by the two connected individuals. By assumption, cooperation tends to improve the local environment, which in turn favors defection, thereby creating negative feedback between strategies coupled with the environmental condition. We analyze the model under strong selection using a mean-field approximation, a pair approximation that tracks both strategy pairs and edge environments, and agent-based simulations. In well-mixed populations, the feedback can generate bistability between full cooperation and full defection, as well as sustained oscillations arising through a Hopf bifurcation. Sparse networks (i.e., with low nodes' degrees) can eliminate bistability, suppress oscillations, and produce stable intermediate levels of cooperation. The pair approximation captures these effects and generally agrees more closely with simulations on sparse networks than the mean-field approximation. Our results show that edge-based environmental feedback and network structure, such as density of edges, interact to determine whether the fraction of cooperators stabilizes or oscillates over time.

0.33

Feel My Focus: Gaze-Triggered Digital Twin Creation for Remote Collaboration

cs.HC cs.CVcs.GR ·2026-11-08 ·Shih-Yu Ma, Suibi Che-Chuan Weng, Ming-Yuan Pai, Tom Yeh, Ellen Yi-Luen Do
top venue
Abstract · 5 pages, 3 figures

We present Feel My Focus, a gaze-triggered mixed reality (MR) demonstration that enables rapid creation and sharing of lightweight 3D digital twins during remote collaboration. Unlike systems that use gaze primarily as a referential pointer, Feel My Focus treats sustained visual attention as a generative action: a fixation on a physical object initiates single-image capture, attention-guided segmentation, and one-shot 3D reconstruction, allowing a remote partner to inspect and discuss the object in situ. The system integrates gaze-tracking glasses, modular segmentation and reconstruction services, and an MR client for runtime streaming and interaction. This design prioritizes responsiveness and experiential continuity over photometric fidelity. We report observations from a public demonstration, highlighting design trade-offs, failure modes, and user control considerations for deploying gaze-triggered generative systems in real-world collaborative settings.

0.32

Dashboard Remixing: Goal-oriented Organization and Enrichment of Visualizations and Insights

cs.HC ·2026-11-08 ·Junran Yang, Leilani Battle, Jane Hoffswell
published
Abstract

Dashboard creation requires iteratively authoring multiple views and organizing them into a structured format. However, creators often struggle to step back and assess their own work for organization and refinement due to cognitive limitations. In this work, we propose a pipeline to provide computational support for self-reflection during the design process. Our pipeline automatically infers analytical objectives, discovers implicit patterns, and structures those patterns to surface coherent analytical pathways. It further enriches existing dashboard designs by generating complementary visualizations and aggregating natural language insights based on relationships across data entities (e.g., contrasting, drill-down, and relatedness) to reveal chart connections. Our rule-based pipeline provides objective, verifiable feedback, offering an external perspective while enabling creators to tell their own data stories. Four case studies and six expert interviews demonstrate how this approach helps creators reflect on organizational alternatives and surface hidden analytical opportunities in a principled way, reducing the cognitive burden while encouraging deeper analytical exploration.

0.28

Wording Matters: Understanding How GenAI Policy Rules Shape Teachers' and Students' Workflows and Perceptions

cs.HC ·2026-11-08 ·Sabiha Mehreen, Xiangzhe Yuan, Rishab Nithyanand, Naimul Hoque
Abstract

The growing use of generative AI (GenAI) in higher education has led institutions to develop policies governing its applications. However, we have little knowledge about how teachers and students adapt and adhere to the policies. In this paper, we analyzed GenAI policies from 101 U.S. universities, identifying nine recurring rules for GenAI use, and interviewed 20 students and 20 faculty members about the rules. We find that the wording of the rules can change how students perceive and act on the rules while introducing different challenges that teachers face in enforcing them in practice. For example, rules such as "Passing off AI-generated content as your own constitutes academic dishonesty" and "Students are accountable for any inaccurate, biased, or unethical content" both outline the acceptable use of AI, but the former can make students hide their use of AI out of fear of unwanted scrutiny, whereas the latter can encourage students to take responsibility and ownership of their use of AI and be more transparent. Students described uncertainty about acceptable use, fear of penalties and unfair treatment, and stigma that discouraged acknowledging AI use. Teachers took on approval workload, had to interpret acceptable use, and lacked reliable ways to verify compliance. We summarize these impacts in a policy-impact matrix linking each rule to its effects on teacher and student workflows and perception.

0.28

Cable construction effects on the lateral forces and wakes of helically stranded cables under oblique inflow

cs.CE physics.flu-dyncs.RO ·2026-11-08 ·Tielin Ma, Junlin Zhang, Jingcheng Fu, Chongwen Jiang, Chuan Ye
Abstract

Helically stranded cables in aerial towing systems experience oblique inflow, but are often modeled as smooth cylinders, neglecting three-dimensional cross-stream coupling and surface texture. We perform large-eddy simulations and spectral proper orthogonal decomposition for a smooth cylinder and four stranded cables at four inclinations from $22.5^\circ$ to $90^\circ$, all stationary, with $Re \approx 1.0 \times 10^4$. Aerodynamic differences are governed mainly by cross-sectional symmetry and strand-scale groove depth. Force-source decomposition shows that surface pressure contributes 94.9%--98.3% of the mean normal force and dominates the mean lateral force. High-symmetry, quasi-circular envelopes maintain balanced circumferential pressure, whereas deep-grooved, low-symmetry tri-lobed geometries produce asymmetric separation and a persistent lateral pressure imbalance. In the wake, high-symmetry cables sustain spanwise-synchronized Karman shedding at moderate-to-high inclinations, concentrating fluctuation energy in a dominant low-rank mode. Deep grooves disrupt shedding coherence, deflect the recirculation zone laterally, and spread energy across broadband scales. Mean lateral force and lateral-force fluctuations are therefore decoupled: the quasi-circular $(1 \times 19)$ cable has minimal mean lateral force but strong periodic fluctuations, while the deep-grooved $(3 \times 7)$ cable has weak fluctuations but appreciable mean lateral force. These results support construction-dependent cable selection and aerodynamic load modeling.

0.27

Rules, Choice, and Chance: A Welfare Evaluation Framework for Student-School Matching

cs.CY ·2026-11-08 ·Reetahan Mukhopadhyay, Clara Strasser Ceballos, Augustin Plantureux, Christoph Kern, Julia Stoyanovich
code/data
Abstract · 42 pages, 26 figures, with 14 page main section

Centralized student-school matching systems, adopted in the United States, Chile, the Netherlands, and the United Kingdom, assign students to schools based on their submitted preference rankings and schools' priority policies. Yet welfare in these systems is typically measured through a narrow lens: a few aggregate statistics published by education authorities, such as the share of students unmatched or assigned to a top-ranked choice. A single aggregate cannot reveal how outcomes vary across students, or why. Welfare depends on three interacting forces: the rules policymakers set (algorithm, tie-breaking, list length limits); the choices students make (which and how many schools to rank); and chance (lottery draws and students' residential location). A rule that reduces unmatched rates may simultaneously lower top-ranked assignment rates for specific groups, and evaluating such trade-offs in advance is challenging because individual-level preference data are rarely available. We present Trace (Trade-offs in Rules, Agency, Chance, and Equity), an open-source framework that synthesizes individual-level preference profiles from aggregate data using mixtures of Mallows models and evaluates alternative rules across multiple welfare metrics, disaggregating outcomes by choice and chance. We apply Trace to two real-world systems: NYC high school matching, where we study minimum list length requirements, and Chilean school assignment, where we compare the current multiple tie-breaking rule with a counterfactual single tie-breaking rule. Both reveal substantial, heterogeneous welfare trade-offs, underscoring the value of simulation-based evaluation before implementing school choice reforms.

0.27

From MNH to the Friedman--Komogortsev Method---which changes improve eye-movement classification?

cs.HC q-bio.NC ·2026-11-08 ·Lee Friedman, Mehedi Hasan Raju, Oleg Komogortsev
Abstract · The details about the classification is described in the supplementary materials

The Friedman-Komogortsev Method (FKM) is a rule-based classifier of fixations, saccades, and post-saccadic oscillations in 1000-Hz recordings, developed from the modified Nyström-Holmqvist algorithm (MNH) and used in several recent studies. FKM changes MNH in many ways at once: it centres MNH's Savitzky-Golay filter, adds a position-based test for whether an apparent post-saccadic oscillation continues the saccade, adds rejection tests, and changes several thresholds. We asked which of these changes alter agreement with human classification. Both algorithms were scored against twenty 26-s GazeBase reading recordings hand-classified by three trained raters, and each change was reverted in turn. FKM agreed better than MNH as MNH was used: saccade Cohen's kappa rose on all 20 recordings (median +0.034), and saccade boundaries moved from 2-3 ms late to the raters' sample. One change produced this: centring the filter, which removes a three-sample delay. Reinstating the delay returned FKM's saccade kappa to MNH's value, and centring MNH's filter alone made MNH match FKM on saccades and exceed it on post-saccadic events. The position-based test changed how saccades with several velocity peaks were divided, splitting 8% of rater-labelled "wavy" saccades into a saccade and a spurious oscillation against 39% for delay-corrected MNH. FKM's higher post-saccadic velocity floor traded recall for precision. The remaining changes had no measurable effect on these recordings. FKM's improvement over MNH is therefore a filter-alignment correction that MNH users can apply directly, plus a segmentation rule for multi-peak saccades.

0.26

Known Good, Known Bad: A Ground-Truth Calibration Protocol for AI Judgment in Science

cs.CY cs.HC ·2026-11-08 ·Olga Lavinda
Abstract · 18 pages, 3 figures, 6 tables

Scientific use of generative AI requires evaluation against cases with independently checkable answers. We present a ground truth calibration protocol that compares a verified reference case with reference deviant variants through neutral evaluation, an authority framed recheck, and a line by line audit. Human adjudication distinguishes deviation detection, premise accommodation, unsupported doubt, and correctness of the resulting calculation. An instructor generated organic chemistry demonstration comprises six conversations and 18 responses from an unidentified backend served through the free logged out ChatGPT interface in July 2026. Neutral evaluation detected three of five intended deviations. Both initial misses, involving oxidant stoichiometry and experimental scale, were detected at the authority framed second turn; all five detections persisted at the third turn audit. The fixed sequence supports a descriptive account of recovery across turns. In the scale case, the model initially produced a locally coherent peroxide limited yield. After recognizing the unit problem, it incorrectly retained that limiting reagent conclusion for the corrected scale. Detection and numerical repair therefore require separate assessment. The verified reference attracted speculative concerns while retaining its correct yield calculation. The protocol connects documentary verification of scientific claims and citations with empirical verification of laboratory interpretations. It supplies inspectable transcripts, reference calculations, behavior codes, and a reusable calibration architecture for AI literacy instruction. The pilot demonstrates the protocol's measures in one case family; replicated model evaluation and learner assessment are the next validation steps.

0.26

The AI research space shapes country academic performance

cs.SI physics.soc-ph ·2026-11-08 ·Siyi Zhou, Tianxing Pan, Qing Ke
Abstract

Recent advances in Artificial Intelligence (AI) have demonstrated great potential across diverse domains, prompting increasing attention to understanding the AI ecosystem. As knowledge production is often perceived as a process of connecting existing pieces of knowledge, mapping the interconnections of AI knowledge is crucial for unraveling the dynamics of scientific production in AI. However, maps created so far have fallen short of capturing knowledge relatedness, focusing instead on citation relationships between subfields. Here, we build an AI research space by connecting pairs of AI research topics based on their proximity measured through author-topic coupling. Projections of country level performances onto the space reveal that countries exhibit distinct baskets of specializations. We further examine the role of the structure of the AI research space in the scientific development of nations. We find that countries are more likely to gain specialization in topics that have high proximity to already specialized neighbor topics, and that publications whose topics align more with the developed topics in the country are more likely to become high impact. Our results highlight the AI development gaps across nations and the path-dependence in the development induced by the AI research space.

0.24

Hydrograd v2.0: a differentiable two-dimensional shallow-water model on unstructured meshes with steady and unsteady adjoints for calibration and closure learning

cs.CE physics.flu-dynphysics.comp-ph ·2026-11-08 ·Xiaofeng Liu
theory
Abstract

We present \textit{Hydrograd} v2.0, a significantly improved differentiable 2D shallow-water model written in Python with JAX. The model solves the shallow-water equations on the unstructured meshes of production models such as SRH-2D and HEC-RAS 2D with a well-balanced, positivity-preserving finite-volume scheme that handles wetting and drying. Gradients are tailored to the type of problem. Steady states are found by Newton's method with an exact sparse Jacobian obtained by graph colouring, and their gradients follow from the implicit function theorem at the cost of one linear solve. Unsteady flows are differentiated in reverse mode with checkpointing, or with an implicit time integrator whose gradient costs about two forward runs. All gradients agree with finite differences, to about $10^{-8}$ for steady flows. On the Savannah River case, the steady water surface differs from that of SRH-2D by 4.7\,mm (root mean square). Steady closure learning was 96 times faster than differentiating through time marching and reached a lower loss. An inflow hydrograph was inverted from stage records with a peak error of 2\,\%. In an unsteady roughness calibration, a singular value decomposition analysis of the sensitivity matrix showed that the stage records at the gauges are 70 to 160 times more sensitive to the total resistance of the reach than to its distribution among the roughness zones, which explains why only a gauge layout with gauges in every zone recovered all five roughness values. On one GPU, a mesh of 512,000 cells takes 7.6\,s for a forward run and 51\,s for a gradient.

0.23

ReGKG-2025: A GKG-Compatible Reconstruction of the GDELT Global Knowledge Graph for the Missing Period of 14 June to 2 July 2025

cs.DL cs.CL ·2026-11-08 ·Nikita A. Klimin
Abstract · 17 pages, 14 figures, 19 tables, including appendices. Companion methods paper: arXiv:2610.35358. Dataset: https://doi.org/10.5281/zenodo.23…

The June 2025 outage in the Global Database of Events, Language, and Tone (GDELT) collection pipeline left a gap in the Global Knowledge Graph (GKG), interrupting article-level news metadata used in downstream research. ReGKG-2025 is an archive-derived sample of 281,620 reconstructed article records covering all 1,665 missing 15-minute batches. Articles were recovered from Common Crawl and the Wayback Machine using source quotas estimated from surrounding native GKG data. Text was screened and normalized, non-English bodies were translated, and flagged records were manually reviewed. Eligible articles were converted to the 27-column GKG 2.1 layout. The versioned package includes record-level provenance, archive replay and WARC locators, source summaries, source-by-day calibration weights, and the complete manual decision journal. The sample reaches about 14% of modeled native volume and represents sources accounting for 84% of selected reference record mass. In an intact-week test, 41% of recovered articles matched native GKG records; field agreement on matched pairs is close to the processor baseline (tone MAE 1.69). The dataset supports retrospective content analysis and source-level aggregates; its tone and theme distributions differ from native GKG, and it does not support exhaustive counts. Data: https://doi.org/10.5281/zenodo.23964265.

0.20

Retrospective Analysis of Team-Following Trading Profitability in Pumpfun Meme Coins Using Funding Networks

cs.SI physics.soc-phq-fin.TR ·2026-11-08 ·Fail Gafarov
Abstract · 42 pages

If the right token-creation teams had been known, how much could following them have returned, and what is left when only information available before each launch is used? We study ten separately recorded Pumpfun windows (April 2024 to September 2026), each reconstructed independently from real-SOL reserve states and finite-capital accounts that start with one SOL. Teams are directed funding groups ranked in hindsight. When every team exits at its own best reserve threshold, chosen in hindsight, the main account ends above one SOL in all 8 windows that have a qualifying team (1.10 to 11.32 SOL), but free cash exceeds one SOL in only 5, and marks of positions that never reached their threshold make up 17--94% of equity. If teams are recognised online, from tokens whose first hour has completed, 9 of 19 teams become identifiable only after their last launch; the remaining 10 end above one SOL in 10 cases, with free cash above one SOL in 7; funding links in this step still come from the full window, so even these results are not a fully realisable signal. A large model balance, a result above random cohorts that repeat the threshold search, and a statistical rejection in the selection-matched nulls (3 of 40 tests after Holm adjustment) are three different statements, and these nulls condition on the discovered funding graph, so they do not calibrate the whole selection procedure. Midpoint tests describe a further source of apparent opportunity. A detailed analysis of the best and the worst window, with every analysis parameter varied, shows that the worst result, which is close to break-even, rises several-fold with a stop rule or another funding-graph construction. The results are conditional retrospective model accounts, not executable profits, and not evidence of a causal effect of the SOL price.

0.19

Trail Making Tests in Extended Reality: A Critical Review with Implications for Cognitive Ageing

cs.HC cs.CY ·2026-11-08 ·Panagiotis Kourtesis
Abstract · Review article. 22 pages, 1 figure, 4 tables; Supplementary Materials (19 pages); ancillary files: the Supplementary Materials (PDF), File S…

Extended reality (XR) versions of the Trail Making Test and Color Trails Test are promising for assessing and training cognition in older adults. However, each XR adaptation is a new instrument, whose display, interaction technique, timing rules and layouts may change what it measures. This critical narrative review maps these implementations by software lineage and appraises their evidence for assessment and training against common criteria. The Trail Making Test in Virtual Reality (TMT-VR), whose source code was inspected, serves as a measurement case. XR trail-making tests showed convergence with paper and computerised versions, varying by test part and sample, and clinical group differences (e.g., Parkinson's disease, mild cognitive impairment). However, gaps in reliability (e.g., randomised forms), validation (e.g., norms, prospective samples) and reporting (e.g., error handling, build identity) emerged, and no training benefit attributable to the trail-making component was identified. The TMT-VR case shows how interaction technique, timing rule, error handling and unlogged layouts shape a score's meaning. Future tests should log layouts and events, estimate alternate-form reliability and undergo prospective validation, and training trials should isolate the trail-making component. Several XR trail-making tests have moved beyond feasibility, and reported tolerability during supervised sessions supports further evaluation in older adults.

0.13

From Dialogue to Task: How Cognitive Engagement With Generative Artificial Intelligence Relates to Post-Dialogue Activity and Task Outcomes

cs.CY cs.HC ·2026-11-08 ·Xiaoyu Zhuang, Aneesha Bakharia, Dragan Gašević, Hassan Khosravi
Abstract · 31 pages, 12 figures. Submitted to the Journal of Computer Assisted Learning

Generative AI is increasingly embedded within learning tasks, yet interaction frequency reveals little about cognitive engagement in student-AI dialogue. Little is known about how such engagement is situated within task workflows or how dialogue characteristics and subsequent task activity relate to students' behaviour and task outcomes. This study examined cognitive engagement in student-AI dialogue, its position within task workflows, and associations of dialogue and task-process characteristics with post-dialogue activity and task outcomes. Data were collected from 1,681 students across 40 university courses using three AI companions embedded in resource creation, peer moderation, and practice. Cognitive engagement in 829 conversation threads was coded with a dialogue-oriented Interactive-Constructive-Active-Passive (ICAP) framework, and 2,969 task sequences were modelled with first-order Markov models. Regression models linked dialogue and task-process characteristics to behavioural re-engagement, peer-rated artefact quality, and practice performance. Passive and Active engagement predominated, although resource creation involved more Constructive and Interactive dialogue, peer moderation more Passive, and practice more Active, while higher-engagement dialogue was less often followed by session termination. In resource creation, student-composed messages were associated with more dialogue-to-task returns than template-only requests at the same Passive level, whereas deeper cognitive engagement was associated with higher final artefact quality independent of return frequency; in practice, AI dialogue after an incorrect attempt was associated with fewer further attempts but higher next-attempt correctness among continuers. Post-dialogue activity and task outcomes are distinct aspects of AI use that interaction frequency cannot capture and evaluations should assess separately.

0.12

Linking Event-sequence Data Claims to Analytic Provenance in Chatbots: Technique and Evaluation

cs.HC ·2026-11-08 ·Po-Ming Law, Jianheng Ouyang, Nguyen-Truong Thinh, Weizhi Li, Arpit Narechania
Abstract

Chatbots are increasingly used to analyze temporal event sequences of which website clickstreams are among the most common. Explanations are essential for understanding claims about event-sequence data. However, existing systems often require programming expertise to interpret these explanations and fail to explain the data processes underlying the data claims. We introduce a technique for linking a data claim to its analytic provenance (i.e., the analytic steps and results that produced it). It has two components: a reference table that records analytic steps and results in addressable cells, and a citation format that binds each claim to one or more cells. To operationalize this technique for clickstream analysis, we curated a set of analytic operations for clickstream analysis, organized them through a pivot-table abstraction, and developed tools that an AI agent can invoke to create a reference table and assemble a natural language response with cited data claims. To evaluate how citations affect users' assessment of data claims, we conducted two crowdsourced experiments with 90 participants each. The experiments showed that participants perceived correct data claims to be more accurate with citations when compared with no explanations, and that they perceived incorrect data claims to be less accurate with citations when compared with code and no explanations.

0.10