# Research Papers — 2026-10 · bundle page 20 of 156 > Full-text markdown of papers #951–1000 of 7799 published in 2026-10, newest first. AI-readable mirrors of the full 7799-paper 2026-10 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-10/page/20 [← Previous bundle](https://aipod.dev/papers/archive/2026-10/page/19.md) [Next bundle →](https://aipod.dev/papers/archive/2026-10/page/21.md) ### LintEval: Benchmarking LLMs on Linter-Guided File Refactoring Authors: Tongwei Deng, Jinyu Zhu, Yue Cao, Chenxu Wang, Yanzhen Zou, Bing Xie Published: 2026-10-14 Categories: cs.SE, cs.CL arXiv: https://arxiv.org/abs/2610.20027v1 Linters are widely used in code maintenance and real-time IDE editing to surface lint alerts related to performance, style, error-prone, and other patterns in source files. However, existing code benchmarks do not explicitly evaluate linter-guided file refactoring, where models must reduce lint alerts in source files with correct initial functionality while satisfying functionality preservation. We introduce LintEval, a multilingual benchmark for linter-guided file refactoring across C++, Python, and Java, comprising 13,061 programs with correct initial functionality and 171,662 lint alerts extracted using language-specific linters. LintEval jointly evaluates functionality preservation, lint alert reduction, new alert avoidance, and edit minimality. Experiments reveal a persistent trade-off: closed-source commercial models reach up to 90.4% Reduction@P but achieve only 67.1%-68.3% Pass@1, while several open-source small models show low or negative Reduction@P. A controlled fine-tuning study further shows that supervision format matters: refactoring-oriented SFT raises Qwen2.5-Coder-7B Reduction@P from 24.7% to 76.0%. LintEval enables controlled evaluation of models that improve source files under functionality preservation. --- ### Memorizing Who, Knowing How: Benchmarking Personal and Procedural Memory for Persistent Assistants Authors: Maosheng Qin, Qi Zhang, Mingxuan Xia, Jinglei Chen, Jiexiang Wang, Sheng Guo, Cheng Peng, Haobo Wang Published: 2026-10-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2610.20022v1 Memory systems for persistent personal agents must move beyond isolated memories, maintaining both personal memory about whom they serve and procedural memory about how to act. Crucially, in harness-based personal assistance, procedures or skills are inherently user-conditioned: personal context determines which procedures apply, how they are instantiated, and when they should change. Existing benchmarks typically emphasize either personal or procedural memory, leaving their interaction underexplored. We introduce PraxisMemBench, a harness-based benchmark for persistent personal assistance across daily-life and work trajectories, where personal and procedural memory evolve together and later tasks depend on both. Experiments show that maintaining two memory types separately is insufficient, with completion dropping from single-type to interdependent dependencies (84.6% to 77.9%). More broadly, across existing methods, relation-related failures increase from early to late sessions as interaction history grows (59.8% to 72.9%), revealing a continual-maintenance challenge: writing correct memories is not enough if their evolving content and organization cannot be preserved for later retrieval. To address this, we propose PraxisMem, which uses privileged joint-state supervision to internalize the ability to maintain memory content and organization into the memory manager, with location-aware GRPO and dynamic residual-routed OPSD. PraxisMem reaches 40.78% overall Session Success Rate, compared with 20.60% for Claude Code's native memory. --- ### MCMC Can See Can't See Authors: Daniel Matsui Smola Published: 2026-10-14 Categories: stat.CO, math.ST, stat.ML arXiv: https://arxiv.org/abs/2610.20020v1 It is folklore that passing "convergence" diagnostics does not guarantee that a Markov chain has converged. We prove this as an impossibility for any finite checklist of diagnostics. Running more chains, longer chains, and smarter diagnostics does not avoid this impossibility in certifying convergence. Under regularity assumptions, we determine how the worst-case undetected non-convergence decays as more diagnostics are added, and how to choose diagnostics optimally. Folklore suggests that the sampler's slowest mixing modes are the "dangerous" ones that must be monitored; we prove that the slow modes yield nearly optimal diagnostics as sampling proceeds. Given a source class, we also give an anytime-valid bound on the remaining discrepancy and show how undetected bias caps the effective sample size that a finite checklist can certify. --- ### MorphAct: Robot Policies Morphing with the Worlds They Act In Authors: Ziyan Liu, Jiting Liu, Weihong Zhong, Wenda Yu, Yeqiu Chen, Jason Li, Xiaojiang Peng, Shuyuan Hu, Kai Wang, Bo Zhao Published: 2026-10-14 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2610.20019v1 Progress toward general intelligence in the physical world requires scaling robot learning across diverse environments and interactions. However, existing robot policies use shared parameters that remain fixed across interaction contexts. When different operations favor competing parameter adjustments, fitting their mixture can leave individual operations insufficiently adapted. Our atomic-task diagnostics reveal competing update directions in both vision-language and action parameters, with source-improving updates increasing loss on other tasks. This motivates MorphAct, a contextual parameterization framework that dynamically adjusts robot policy parameters as interaction contexts change. Two coupled generators produce context-dependent low-rank updates for the vision-language and action modules. Their coupling follows the policy's computational dependency: vision-language updates shape the representations that condition action-side parameter generation. Both generators are trained jointly through the native action objective, without operation labels or expert-adapter targets. MorphAct improves performance across three VLA backbones on LIBERO and Meta-World, with gains of up to 16.30%. In real-world evaluation, it achieves gains of up to 14% over $π_{0.5}$. Ablations support the benefits of contextual parameter generation and representation coupling. --- ### On the Convergence of Semi-Gradient TD: A Mean Field Perspective Authors: Haruki Settai, Tadashi Kozuno, Shinji Ito Published: 2026-10-14 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2610.20014v1 Semi-gradient temporal-difference learning is a core algorithmic component of reinforcement learning, but bootstrapping with function approximation generally induces a non-conservative vector field and can lead to instability. We analyze this effect for two-layer neural networks in the mean-field regime and identify a simple stability mechanism in which entropy-induced relaxation dominates the sensitivity of the moving bootstrap target. Under this condition, we prove exponential convergence to a unique fixed point and quantify its bias relative to the unregularized Bellman solution. We also show that the same mechanism extends to entropy-regularized off-policy control through soft Q-learning. --- ### ChemArgus: Automated Scientific Diagnostics for Lab-Grade Multimodal Chemistry Reasoning: A Frontier Benchmark Authors: Hanyao Guo, Mo Cui, Zihan Tan, Leixin Sun, Yichun Wang, Yinan Shen, Guojun Zhu, Jia Jiang, Erhao Chen, Zui Chen, Wenzhe Gao, Zhihao Yu et al. Published: 2026-10-14 Categories: cs.LG arXiv: https://arxiv.org/abs/2610.20012v1 Chemistry evaluation has moved past multiple-choice probes toward open-ended expert problems, yet the benchmarks that define the current frontier still score answer endpoints alone. Such scoring conflates memorized facts with stepwise derivation, treats visual and textual inputs as interchangeable, and confounds long-horizon planning with stepwise execution. These blind spots prevent any diagnosis of where the reasoning chain breaks in a model. We present ChemArgus, a competition-level chemistry benchmark drawn from chemistry competition material, textbooks, and research papers, whose scoring replaces the answer endpoint with three measurement properties. Fine-grained scoring grades every subquestion point by point with partial credit, so a score lands at the level of the step rather than the outcome. Multimodal capability is priced on every problem containing a real image, paired with a controlled text description, so visual understanding is measured against text on identical tasks. Diagnostic attribution reads paired metrics that share one scoring unit and differ only in context construction, so a gap between two scores isolates a single cause, namely integration, propagation, or local reasoning. GPT-5.6-Sol scores 23.2% on unassisted whole-problem solving against 28.8% and 30.4% under the two sequential metrics, a gap measuring the integration burden. Through rubric partial credit and these cause-isolating score gaps, ChemArgus shows not only whether a model answers chemistry correctly but where its reasoning chain fails, so that every score it reports doubles as a diagnosis, a readout of the chain rather than a verdict, aiming to guide language models step by step on their path toward chemistry expertise, one corrected step at a time, each step priced by the rubric, point by point. --- ### Outperforming 1000 Euclidean Layers With Hyperbolic Goal-Conditioned RL Authors: Timo Klein, Thomas Lang, Andrii Shkabrii, Lukas Miklautz, Yllka Velaj, Sebastian Tschiatschek Published: 2026-10-14 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2610.20009v1 Scaling self-supervised contrastive reinforcement learning (CRL) to 1000-layer Euclidean neural networks has enabled state-of-the-art goal-reaching performance, yet wall-clock time scales linearly with depth. We find that changing the geometry achieves stronger performance with far shallower networks. This matters in complex navigation environments, where walls can make route distance deviate substantially from Euclidean distance. In a U-shaped maze, we show that the longer the route, the more it deviates from Euclidean distance. Hyperbolic CRL addresses this mismatch by replacing Euclidean distances in CRL's contrastive objective with geodesic distances in hyperbolic space. On hard navigation tasks, inductive bias beats scale: depth 64 hyperbolic agents outperform 1024-layer-deep Euclidean networks while training more than $12 \times$ faster and using about $13 \times$ fewer parameters. We observe that hyperbolic agents behave qualitatively better in hard humanoid mazes, where they follow corridors to reach goals, whereas Euclidean agents try straight-line shortcuts over walls. Our results show that matching the representation geometry to the task can substitute for substantial network depth. Code is at https://github.com/Probabilistic-and-Interactive-ML/HypGCRL . --- ### Minimax estimation of the pseudo-spectral gap of continuous Markov chains, with applications to empirical concentration bounds Authors: Vahe Karagulyan, Mikołaj Kasprzak, Pierre Alquier Published: 2026-10-14 Categories: stat.ML, cs.LG, math.ST arXiv: https://arxiv.org/abs/2610.20007v1 We study the minimax estimation of the pseudo-spectral gap $γ_{ps}$ of a Markov chain from the observation of a single trajectory. Extending earlier works on finite-state space chains by Wolfer and Kontorovich (2024), we provide an estimator in the case of a compact, continuous state-space. Our estimator appears to be minimax-optimal on the class of chains with a transition operator with Lipschitz density, up to logarithmic terms. Our estimator, combined with Paulin's version of Bernstein inequality, allows to derive empirical concentration inequalities for Markov chains. This unlocks many applications, from confidence intervals, to active learning and generalization bounds. In particular, for rested bandits with Markov rewards, we provide a version of UCB with a regret bound similar to the one in the iid case, without assuming any knowledge on the pseudo-spectral gap of the chains. --- ### Provable Transformer Context Compression for In-Context Reinforcement Learning Authors: Yajie Zhao, Yingjun Mo, Zhiyong Wang, Bo Tang, Chao Xu, Fang Kong Published: 2026-10-14 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2610.20004v1 Transformers can perform in-context reinforcement learning without parameter updates. However, existing theoretical constructions often require context sizes that grow with the number of sampled transitions, leading to increasing memory costs in large-scale problems. This raises a fundamental scalability question: how much context is intrinsically necessary to carry out such learning? We answer this question for fixed-policy TD evaluation. We show that, with a shared linear decoder, the minimum context dimension required to exactly reproduce the first $L$ TD iterates is the Krylov dimension induced by the sampled transitions. When the available context dimension $m$ falls below this threshold, exact reproduction is no longer possible. In this regime, we characterize the optimal approximation error and show that it decays as $Θ(γ^m\max\{1,1/(m(1-γ))\})$ where $γ$ is the discounting factor. Moving from representation complexity to actual memory cost, we characterize the optimal retained memory for exact continuous representations as $Θ(nk_L)$ and show experimentally that on a sampled task whose exact rank is full but whose numerical rank is small, a context accurate to $10^{-2}$ is $88.7\%$ smaller than the sparse sufficient statistics of TD. Finally, we establish the learnability of the shared executor through pretraining by proving global convergence of an auxiliary objective over effective attention parameters under suitable task diversity. Experiments across diverse TD evaluation problems validate parameter recovery and the predicted context--accuracy tradeoff. --- ### What Coding Agents Leave Unsaid: Measuring Task-State Gaps Between Execution and Reports Authors: Guoyang Xu, Rui Zhang, Xiuyu Wu, Hao Chen Published: 2026-10-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2610.20003v2 Human supervisors rely on coding agents' reports to assess progress, verify repairs, and decide when to intervene. But a supervisor cannot investigate a failure they were never told occurred. Failed attempts, workarounds, and unresolved risks may disappear from routine reports, leaving supervisors unaware of issues that warrant scrutiny. We conduct CTXGAP, a measurement study quantifying the gap in task-state recoverability between agents' accumulated user-facing reports and their full recorded execution contexts. Across 37 coding tasks and 4,799 valid probes, agents answer 93.6% of task-state questions on demand; full-context readers recover 92.7%, while report-only readers recover 52.0%, a 40.7-point gap. At the final checkpoints of 30 verified fixes, the gap remains 34.8 points. Human annotation identifies states mentioned without enough supporting detail as the largest category of unrecovered information, and model-assessed supervision-relevant states retain substantial gaps. Full terminal scrollback improves recovery to 73.6% but still trails full context. Under matched report lengths, context-informed rewrites improve recovery by about eight points, without a detectable targeted-over-natural advantage. Positive gaps persist across harnesses and independent DeepSeek execution and measurement. CTXGAP establishes report-based task-state recoverability as a complementary evaluation dimension. --- ### A Strong Harness Narrows the Model Gap: A Comparative Evaluation for Legislative Document-to-XML Conversion Authors: Palvika Bansal, Rittika Jindal, Ruichong Wang Published: 2026-10-14 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2610.19999v1 Legal-content reproduction requires matching the source exactly in section hierarchy, verbatim text, citations, and amendment spans; a dropped citation or mislabeled amendment carries into the products lawyers rely on. We report an industrial pipeline converting U.S. state legislation into strict, schema-constrained XML, and hypothesize that its accuracy comes from harness engineering, the deterministic tooling built around a model, more than the model itself. Our controlled ablation crosses three harnesses (zero-shot, a single tool-using agent, and a decomposed multi-agent pipeline) with three Claude models on 153 gold bills from two jurisdictions. The harness narrows the model gap: the weakest model in the full pipeline beats the strongest model alone (0.948 versus 0.806, paired bootstrap p less than 0.001). Because the two agentic harnesses share identical tools, their gap comes from orchestration: decomposition plus the persisted state and per-agent context it brings. Across four scored dimensions, the model spread at zero-shot (0.68 to 0.81) narrows to 0.95 at the full pipeline; the residual sits almost entirely in citation normalization, where the canonical identifier is retrieved from an external authority rather than generated from the weights, so a deterministic lookup tool supplies the exactness that a larger model does not. --- ### Empirical Boundaries of Reservoir Computing for Few-Shot Sequence Pattern Tasks Authors: Mario Mallea, Angela Nebot, Franscisco Mugica Published: 2026-10-14 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2610.19997v1 Efficient sequence modeling remains a fundamental challenge in machine learning. Despite the success of Transformers, recurrent neural networks (RNNs) have recently re-emerged as efficient alternatives. Particularly, Echo State Networks (ESNs) offer a compelling framework of randomized RNNs that have been extensively studied for time-series forecasting and dynamical systems. This paper presents an empirical analysis of distributional, architectural, and optimization variants of ESNs evaluated on few-shot sequence-to-sequence (seq2seq) pattern tasks. These synthetic tasks utilize discrete symbol sequences, characterized by delays, inactive phases, and triggered markers to isolate and evaluate memory and pattern recognition capabilities. The minimality of these tasks is crucial to understand the strengths and limitations of machine learning models. Furthermore, when the input signal utilizes a one-hot encoding, ESN dynamics are governed strictly by their weight distributions, offering an ideal lens through which to study reservoir computing. Our findings map the performance boundaries of ESNs relative to state-of-the-art (SOTA) models. We show that preserving a linear state regime is vital for memory tasks, such as pattern completion or simple copy. We propose a data-dependent modulation mechanism for nonlinear input injection into the reservoir (G-ESN). G-ESN functions as a statistical filter capable of solving tasks that conventional ESNs cannot (the adding problem and bracket matching). Finally, for Deep ESNs, we evaluate a shared readout mechanism across layers that outperforms conventional concatenated readouts on structural tasks, including selective copy, sorting problem, and bracket matching. --- ### Gray-Box Hallucination Detection in LLMs with Token Identities Authors: Shaked Goldshtein, Yitzchak Grunbaum, Fabrizio Frasca, Guy Bar-Shalom, Haggai Maron Published: 2026-10-14 Categories: cs.CL arXiv: https://arxiv.org/abs/2610.19996v1 Large language models often produce fluent but factually incorrect output. Gray-box detectors identify such hallucinations from the token-probability statistics emitted during generation, avoiding the need for white-box access to hidden states that frontier models do not provide. These detectors, however, typically rely only on probability-derived representations and ignore the identities of the candidate tokens. They can therefore observe how uncertain the model is, but not what it is uncertain about. Consequently, they cannot distinguish between a model that is torn between "big" and "small" and one that is torn between "big" and "huge" when the two pairs receive the same probability distribution. We introduce GURI (Gray-box Uncertainty Representations with Identity), a lightweight gray-box hallucination detector that combines candidate-token probabilities with per-token vector identifiers. This exposes which alternatives the model considered while preserving the accessibility and transferability of the gray-box setting. Across three open-weight models and three question-answering benchmarks, GURI increases mean test ROC-AUC from 72.09% to 79.09% over a probability-only baseline. On the closed-source GPT-4.1 model, mean ROC-AUC increases from 67.39% to 71.84%. Among the three open-weight models, GURI outperforms transferred probability-only baseline in 17 of 18 directed cross-model transfers without fine-tuning. For the three open-weight models, cross-dataset transfer between HotpotQA and TriviaQA with the model held fixed improves mean ROC-AUC over transferred LOS-Net by 3.76 points. --- ### Why 3D Reconstruction Breaks Down on Under-Ice AUV Video, and How to Predict It Authors: Ana-Ştefania Kogălniceanu, Klára Orbán, Ashwin Nedungadi, Stefan Lüdtke, Kuderna-Iulian Benţa Published: 2026-10-14 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2610.19993v1 Reconstruction of 3D structures using under-ice AUV video footage has been inaccurate, although it is not certain whether this is due to shortcomings in the methods or an inherent deficiency of the capture system. We test four different models for reconstruction on the same Antarctic under-ice AUV video (Icefin, Kamb Ice Stream): classical feature matching SfM, differentiable photometric splatting, volumetric depth fusion, and a pose-free feed-forward foundation model. All four fail, each in a distinct way: near-total registration yielding almost no structure, geometrically broken output despite healthy photometric metrics, a persistent coverage artifact, and unstable cross-chunk scale. We show that all four are consistent with the same capture regime, a small camera-baseline-to-scene-depth ratio. Following our analysis, we propose a detection system that flags this type of failure in minutes, rather than hours of wasted compute. --- ### SoK: Grading Open-Text Legal Answers, from Law School Exams to Trustworthy Rubric Judges Authors: Sebastian Nagl Published: 2026-10-14 Categories: cs.CL arXiv: https://arxiv.org/abs/2610.19992v1 Law students are examined through open-text answers: case solutions, essays and memos that apply the law to a set of facts. Such answers are expensive to grade, human graders disagree widely on them, and large language models (LLMs) now offer to grade them at scale, both in teaching and in the benchmarks that test legal AI. This paper systematizes how open-text legal answers can be evaluated. We derive eight desiderata from educational measurement and from the law that governs exams, among them validity, reliability, reasons, credit for defensible alternative solutions, robustness to gaming and human oversight. We then compare eight evaluation paradigms against them, from unstructured human grading to LLM judges with instance-specific rubrics and pairwise comparison. Three findings stand out. A rubric written for the task is the strongest lever for agreement between an LLM judge and human graders, but the reported figures answer different questions and rarely come with the human agreement they must be read against. The central validation gap lies where examination law is most demanding: defensible alternative solutions, follow-on errors, and the examiner's correction of model judgments. And rubric-centred grading is the most auditable family, because it makes the standard, its application and the final decision checkable. From this we derive the trustworthy rubric judge: an instance rubric that credits arguments and alternative paths,evidence checked in code, probes, repeated runs, calibration against independent human grades,logging, and an examiner who reads every script. We state what each safeguard rests on, what remains to be tested, and the uses the evidence supports today. --- ### LatentLoop: Scaling Looped MoE Models with Asymmetric Compute Allocation Authors: Zefeng He, Yunkai Chai, Xiaoyang Chen, Tong Zhu, Zongliang Wu, Yun Luo, Yafu Li, Yu Cheng Published: 2026-10-14 Categories: cs.LG arXiv: https://arxiv.org/abs/2610.19975v1 Looped Mixture-of-Experts (MoE) models combine parameter reuse with sparse computation to improve performance under matched budgets. Prior work demonstrates that different loop visits can play different roles even when they share parameters. To probe the contributions of different loop visits, we learn separate residual coefficients in a model that applies a shared block twice, observing a smaller scale in the first visit. Fixed coefficient experiments show that reducing the first visit's scale preserves or improves performance, whereas the same reductions to the second hurt it. This asymmetry motivates us to explore asymmetric compute allocation. We propose \textbf{LatentLoop}, which compresses the first visit's sequence, retains full resolution in the second, and reallocates saved computation to increase the hidden size. We further learn how strongly the first visit updates the full sequence, and share MoE weights across visits while using separate attention parameters to accommodate their different sequence resolutions. Across 15 configurations spanning 1.3B--7B parameters, with the largest budget exceeding $10^{21}$ FLOPs, LatentLoop consistently lowers validation NLL. At a fitted NLL of 1.988, predicted compute savings are 19.0\% and 7.9\% relative to vanilla MoE and the looped baseline. It also improves average scores on nine benchmarks at all sizes in 0-shot and 5-shot settings. These results suggest asymmetric compute allocation is a promising direction for looped MoE models. --- ### Safe-FCL: Continual Risk Awareness and Fine-Grained Safety Aggregation for Federated Continual Learning of LLMs Authors: Shuaijie Ma, Yongkang Hu, Yu Cheng, Jiuan Zhou, Yi Xie, Yushuo Zhang, Zhaoxia Yin Published: 2026-10-14 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2610.19968v1 Federated fine-tuning enables large language models (LLMs) to learn collaboratively from distributed client data, but malicious client updates can compromise global model safety. As federated tasks evolve, changing malicious update patterns challenge the ability of existing defenses to maintain reliable risk detection over time. To address this challenge, we propose Safe-FCL, a framework integrating continual risk awareness with fine-grained safety aggregation for federated continual learning of LLMs. Specifically, the Continual Risk Awareness (CRA) mechanism learns emerging risk patterns across sequential tasks and constrains detector updates using a protective subspace derived from historical features, mitigating forgetting of previously learned malicious patterns while adapting to new risks. To translate risk awareness into aggregation-level safety control, the Fine-Grained Safety Aggregation (FGSA) mechanism combines client-level risk assessments from CRA with module-level deviations from historical update behavior to adaptively regulate aggregation contributions, suppressing potentially malicious updates while minimizing unnecessary restrictions on benign updates. Extensive experiments demonstrate that Safe-FCL outperforms current SOTA methods, improving global-model Rule and MD-Judge scores in federated continual learning by 0.41% and 0.64%, respectively, while increasing average risk recognition accuracy by 2.01%. --- ### Choosing the Number of Attention Heads from Intrinsic Dimension: A Small-Scale Study Authors: Arjun Boya, Charan Sai Bandaru, Masthanaiah Pillimetla, Kavana Chidanandappa Published: 2026-10-14 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2610.19966v1 The number of attention heads in a transformer is usually set by convention or grid search. We test a simple heuristic for setting it in advance: estimate the intrinsic dimension of token representations without labels, and use roughly that many heads. We estimate intrinsic dimension with a nested-dropout VAE trained on DistilBERT embeddings of WikiText-2, choosing the KL weight at the knee of the VAE's rate-distortion curve rather than by hand. This gives $k^*=9.2$ (95% CI (8.6, 9.8)). In a small 4-layer transformer trained from scratch, 10 heads and 32 heads are statistically equivalent on SST-2 and on MRPC (paired TOST over 10 seeds, $\pm1$ point margin), with 69% fewer attention parameters for the 10-head model. Training 10 heads from the start also slightly outperforms training 32 heads and pruning to 10 (0.6 points, $p=0.008$). The evidence is narrow: two classification tasks, one small architecture, no pretrained models, and no formal link between intrinsic dimension and head count. We present the result as a practical heuristic worth testing at larger scale, not as a general principle. --- ### Complementary Future Token Pruning for World Action Models Authors: Baekseung Kim, Junhoo Lee, Sohyeon Hong, Suhyun Jeon, Minkyu Kim, Seungyeon Kim, Donghoon Han, Nojun Kwak Published: 2026-10-14 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2610.19957v1 World action models (WAMs) jointly predict future visual states and robot actions, but processing dense future-frame tokens incurs substantial inference latency. We propose CF-Prune, a training-free framework for pruning future-frame tokens in pretrained WAMs. Our key insight is that future visual predictions support action generation and therefore need not preserve every visual detail required for high-fidelity video synthesis. CF-Prune combines instruction and action attention to identify task-relevant tokens during the initial denoising forward pass, even when future-frame and action representations remain highly noisy. An entropy-based phase identification mechanism further retains dense computation during precision phases to mitigate pruning-induced performance degradation. Experiments on RoboLab, RoboCasa, and real-world manipulation tasks demonstrate consistent inference speedups. At a 35% pruning ratio, CF-Prune achieves average speedups of 1.19-1.27x while improving task success rates. At a 50% pruning ratio, it achieves 1.29-1.33x average speedups on both simulation benchmarks, with overall success rates within 1.0 percentage point of the baseline. --- ### Neurosymbolic Learning of Interpretable World Models Authors: Riccardo Andreoni, Massimiliano Ronzani, Chiara Ghidini, Alessandro Daniele, Emile van Krieken Published: 2026-10-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2610.19955v1 World models aim to learn compact representations of an environment from observations, enabling agents to predict future states and plan their behaviour. However, most existing approaches learn black-box representations, with limited ability to incorporate prior knowledge about the world. We introduce Concept-Grounded World Models (CGWMs), a neurosymbolic world model design that instead learns interpretable concept representations. We learn these concepts from observations using explicit and complete knowledge, expressed in temporal logic, that specifies the valid world states and the temporal relations governing their dynamics. Furthermore, we directly use this knowledge by formulating planning as differentiable reasoning, optimising the agent's actions by gradient ascent to satisfy a goal. We experimentally compare CGWM with a purely neural baseline and a neurosymbolic baseline on two game environments. We find that CGWMs ground concepts much more accurately, even over long horizons, resulting in significantly more robust planning both in and out of distribution. --- ### AtomSearch: Evidence-Grounded Tree Search for Agent Skill Optimization Authors: Jia Tong, Senyou Deng, Min Peng, Jianfeng Pan Published: 2026-10-14 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2610.19951v1 Agent skills adapt frozen models through editable external instructions. Revising several rules at once makes changes difficult to inspect, while judging each local edit only by its immediate score can discard useful intermediate states. We introduce AtomSearch, which keeps revision actions local while using subsequent skill states to guide search. Contrastive atomic editing links each primitive rule edit to failed and successful executions, a behavioral hypothesis, and an intended scope. Skill-State Tree Search (STS) evaluates short edit compositions and selectively retains them as executable states that can be revisited and extended. Outcome memory supplies later proposals with observed repair and regression statistics. We evaluate AtomSearch across four tasks and five target models. With deepseek-v4-flash, it exceeds SkillOpt on three of four tasks, including a 9.68-percentage-point gain on LiveMathematicianBench, and SkillOpt's final skills contain 2.3-19.7 times as many characters. In 67% of completed searches, the best validation state passes through an ancestor that was not highest-scoring at its depth. These results show how local rule edits can support the exploration of useful revision paths and yield compact skills, with additional offline evaluations used to explore alternative states. --- ### When Protection Overrides Consequences: Understanding Protection Mode in Large Language Model Decision-Making Authors: Yuanfan Li, Junyi Chen, Qi Zhou, Rui Yang, ZhangLiujie, Weihang Chen Published: 2026-10-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2610.19948v1 Frontier language models excel at complex tasks yet may prioritize protection over larger physical consequences in simple, urgent decisions. GPT-5.6 Sol, for example, moves a girl playing in shallow water before retrieving a submerged camera. We call this Protection Mode and study it with ProtectionBench: 285 prompts, 34 physical families, and nine models. At lowest effort, at least one of four responses prioritizes the subject on 34.74% of Sol prompts and over 83% of MiniMax M3 and Qwen3.5-27B prompts; none of four independent human raters does so. Controls show that choices usually track physical risk. We elicit both risks before requesting an action. Some models assess risk incorrectly; others state the correct ordering but still choose the subject. Qwen3.5-27B with thinking shows 100% correct risk ordering but 57.81% subject choice among eligible conversations on eight selected scenes. Attention audits find no stable reduction in attention to relevant facts; removing the stated risk direction reduces object-first choice in dots3-note-prev but not detectably in Qwen3.5-27B. Adding a general "people or animals before property" rule increases subject choice by 19-38 percentage points, whereas factual reminders favor the object by 0.7-5.2 points. Removing naturally generated versions of that rule lowers subject choice by 10-15 points. We also examine whether post-training introduces this pattern. Base models already show Protection Mode; post-training changes justifications more than choices. --- ### Quiet Triggers, Loud Footprints: A Tri-Space Measurement Study of Backdoor Stealthiness Authors: Behrad Tajalli, Henk Berendsen, Xiaoyun Xu, Stjepan Picek Published: 2026-10-14 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2610.19947v1 Backdoor attacks on deep neural networks are commonly evaluated by attack success rate and benign accuracy, yet their stealthiness remains inconsistently defined and measured. Existing work often studies only one observable aspect of stealthiness, such as trigger imperceptibility, feature-space separability, parameter-space concentration, or evasion of particular defenses. This creates a fragmented picture: an attack may appear stealthy under one metric while leaving conspicuous footprints elsewhere. This paper presents a tri-space measurement study of backdoor stealthiness in image classification. We distinguish latent stealthiness---the ideal but unobservable difficulty of distinguishing a backdoored system from a benign one---from measurable footprints in three spaces: input, feature, and parameter space. Under this framework, we evaluate representative backdoor attacks across four datasets and three model architectures using a broad suite of footprint metrics. Our results show that stealthiness is not monolithic: attacks that are quiet in one space are often loud in another, and existing attacks rarely achieve consistently small footprints across all three spaces. We further analyze agreement and disagreement among stealthiness metrics and study how footprint measurements interact with attack success rate. Based on observed blind spots in existing measurements, we introduce two diagnostic metrics: Class-Specific Davies--Bouldin Index (CDBI), which captures class-conditional feature overlap, and TAC--UCLC Product (TUP), which identifies channels that are both trigger-responsive and downstream influential. These metrics are intended to augment, not replace, the broader measurement suite. Finally, we provide practical guidance for future stealthiness evaluations. --- ### Learning Graph Energies for Finite-Time Generation Authors: Amartya Banerjee, Xingyu Xu, Caroline Moosmüller, Harlin Lee Published: 2026-10-14 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2610.19946v1 Energy-based models can represent distributions over discrete structures such as molecular graphs, but an energy that is accurate at equilibrium need not yield accurate samples when the sampler stops after a fixed number of steps. We introduce finite-horizon energy learning, which adapts the acceptance energy of a Metropolis--Hastings sampler to the distribution it returns after a prescribed number of steps, with initialization, proposals, and annealing schedule held fixed. For graph generation, a frozen pretrained model supplies initialization and proposals, and a learned energy governs acceptance. A training-only discriminator scores completed samples, and coupled continuations of the accepted and rejected branches of a proposal give an acceptance-gradient estimator that is unbiased for a fixed discriminator under stated conditions, without equilibrated negative samples or differentiation through discrete states. Generation uses the learned energy in the original single-trajectory sampler, without discriminator evaluations or reranking. On twelve pairwise binary targets with exactly evaluated terminal laws, learned energies have lower median terminal total variation than the equilibrium-correct energy in every tested short-horizon setting, and than its numerically tuned scalar multiple in most settings. On MOSES, a learned two-coefficient combination of pretrained Graph Energy Matching (GEM) potentials lowers FCD by 3.7--10.5\% relative to scale-calibrated GEM at matched sampling steps, with validity gains at the shortest budget and a sub-point validity trade-off at higher budgets. --- ### Does Machine Translation Retain Media Bias? Authors: Silvia V. Gargova, Tim Menzner, Jochen L. Leidner Published: 2026-10-14 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2610.19945v1 Within the most recent two decades, machine translation (MT) has been advanced from a research area to a mature, deployed technology for prose text and most language pairs. However, language has different dimensions of meaning, and the alluded progress has primarily been demonstrated for propositional meaning. In this small exploratory study, we investigate whether one type of affective meaning, namely different types of linguistically expressed media bias, are retained under automatic translation for two language pairs, English and German as the source languages, and Bulgarian as the target language. We evaluate four state-of-the-art translation methods to answer the question whether and to what degree a loss of non-neutral language (specifically, 27 types of media bias) can occur as a result of applying MT. Our empirical findings are consistent with the claim that MT does not fully retain existing news bias of its input source text, but also may add extra bias of its own. To the best of our knowledge, this is the first experimental study of MT media bias retention. --- ### MorphingNet: Symbolic Regression Where the Network Is the Formula Authors: Jincheng Gao, Jiahe Yi, Shuo Sun Published: 2026-10-14 Categories: cs.LG arXiv: https://arxiv.org/abs/2610.19942v1 We introduce MorphingNet, a network for symbolic regression (SR) whose units change their function continuously. Instead of searching over expressions built from a fixed operator library, MorphingNet trains the form of each function as a parameter, so SR becomes one continuous optimization problem. Its units are either pole-free rational functions or complex exponential--logarithm functions, which move between exponentials, sinusoids, damped oscillations, power laws and logarithms as their parameters change. The network is a readable formula at every step of training, and no operator library is needed to train it; a structure-recovery step afterwards rewrites the trained formula into short elementary expressions. Training is dense linear algebra, so it is GPU-friendly and scales to large data. MorphingNet achieves state-of-the-art accuracy on SRBench, recovers most generating formulas on synthetic benchmarks, and on large data is up to two orders of magnitude faster than the strongest search method. By turning the search for a formula into the training of a network, MorphingNet offers a new paradigm for SR and for scientific discovery from data. --- ### To Reason but Not to Stop: Understanding and Recovering Pruned Large Reasoning Models Authors: Minh-Anh Nguyen, Nguyen T. Ha Lan, Xiyan Fu, Dung D. Le, Wei Lu Published: 2026-10-14 Categories: cs.CL arXiv: https://arxiv.org/abs/2610.19934v1 Weight pruning can degrade the accuracy of large reasoning models (LRMs), yet it remains unclear whether it damages reasoning itself. We find that much of this degradation stems from a failure to stop rather than to reason. As sparsity increases, pruned LRMs increasingly fail to terminate within the token budget, often after the correct answer has appeared in their reasoning, instead falling into repetitive loops, even though they still recognize where the dense model would stop. We further find that the hidden-state difference between terminating and looping continuations is consistent across pruned checkpoints, suggesting a shared direction associated with committing to an answer. Building on this, we propose Termination-Guided Activation Steering (TGAS), a training-free method that adds this direction to the residual stream at inference time. Across six LRMs, three pruning pipelines, and math and code benchmarks, TGAS improves termination and overall accuracy while reducing repetitive reasoning and generation length, showing that this failure is actionable. --- ### Instruction Mining From Images: Constructing a Synthetic Dataset for Multimodal Learning ILSP DCVLR Competition Submission Authors: Eleni E. Kasoura, Dimitrios Damianos, Georgios Paraskevopoulos, Vassilios Katsouros Published: 2026-10-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2610.19932v1 This work presents a compact and reproducible pipeline for synthesizing and curating image-based instruction-response pairs for DCVLR. We draw from the Cauldron collection and use Magpie-style instruction and response synthesis with Qwen2.5-VL-7B Instruct to create concise, image-conditioned instruction pairs. The pipeline consists of instruction synthesis, conditioned on input images and controlled via a system prompt, sanitization, tagging, and filtering of the instructions based on the score obtained by a reward model (DeBERTa-v3). We retain samples with positive scores to maximize the quality of the generated instructions. We observe that score distributions have a median and mean reward score of ~ 1.53 and ~ 1.75 respectively, indicating that most generated instructions are of good quality. The resulting dataset contains diverse instructions in the math, spatial, and social reasoning domains. --- ### SAM3RL: Memory Control via Reinforcement Learning for Visual Object Tracking Authors: Tomáš Čížek, Matej Straka, Illia Volkov, Matěj Kripner, Ervin Macić, Jiri Matas, Klara Janouskova, Martin Schmid Published: 2026-10-14 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2610.19931v1 Segment Anything Model 3 (SAM 3) is a foundation model for image and video segmentation that sets the state of the art in visual object tracking. Its tracker, inherited from SAM 2, stores information from past frames in a memory bank, managed by a simple First-In-First-Out rule, allowing temporal consistency in complex videos. Several recent methods improve SAM 2 with hand-crafted memory update rules, each targeting a specific failure mode, such as distractors, occlusions, or object motion. However, their gains transfer only partially to the stronger SAM 3 backbone. We instead present a more general, data-driven approach that frames memory control as a sequential decision-making task. SAM3RL is an extension of the SAM 3 tracker with a lightweight memory controller, trained by reinforcement learning, that decides which frames enter the memory bank in order to maximize tracking performance. SAM3RL sets a new state of the art on 10 of 12 tracking and video object segmentation benchmarks, outperforming all hand-crafted rules without any runtime overhead. The same approach applied to SAM 2 gives SAM2RL, which outperforms hand-crafted rules on 9 of the 12 benchmarks, showing the generality of our method across model generations. --- ### MemDuet: Learning Dual-Process Memory for Long-Context Reasoning Authors: Jiangcheng Song, Haodong Jing, Yulin Mu, Lesong Tao, Jixin Wang, Jin Cui, Tianjing Ren, Runxuan Xie, Yongqiang Ma, Nanning Zheng Published: 2026-10-14 Categories: cs.CL arXiv: https://arxiv.org/abs/2610.19928v1 Reasoning over infinitely long documents is becoming a reality with recurrent memory agents. However, increasing context length exacerbates the trade-off between reasoning performance and inference efficiency. Inspired by Kahneman's account of fast and slow thinking, we propose MemDuet, a dual-process memory framework that learns to coordinate a small language model for fast, parallel evidence selection from raw context chunks (System 1) with a large language model for memory-update verification and memory updates in document order (System 2). We train MemDuet in three stages: supervised fine-tuning of System 1 for evidence selection, Group Relative Policy Optimization (GRPO) of System 2 for verification and memory rewriting, and joint optimization of both policies to coordinate selection and updating for long-context reasoning. Experiments on diverse long-context reasoning benchmarks (RULER-HQA, LongBench v2, and 2WikiMultiHopQA) demonstrate a favorable accuracy-efficiency trade-off as context length grows. Averaged across evaluation settings on the three datasets, MemDuet achieves an end-to-end inference speedup of 9.74x and an output-token reduction of 73.91% relative to RL-MemAgent. --- ### Convergence Rates of Gaussian Process Thompson Sampling for Efficient Global Optimization Authors: Haowei Wang, Yilin Zheng Published: 2026-10-14 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2610.19925v1 Efficient global optimization aims to identify a global optimum of an expensive black-box function within a limited evaluation budget $N$. We investigate the convergence rates of Gaussian process Thompson sampling (GP-TS) under noiseless observations, assuming a fixed objective function in a reproducing kernel Hilbert space (RKHS). For Matérn kernels with smoothness parameter $ν$ in dimension $d$, we establish a worst-case expected simple-regret upper bound of $O(N^{-ν/d}\sqrt{\log N})$ for both discrete GP-TS on sufficiently fine grids and its continuous counterpart. For the continuous implementation, we develop a nested-grid argument to control posterior sample paths over the entire domain. We further establish matching lower bounds of $Ω(N^{-5/2}\sqrt{\log N})$ for both implementations under a fixed one-dimensional Matérn-$5/2$ kernel, demonstrating that the logarithmic penalty is intrinsic to standard GP-TS in this setting. Motivated by this limitation, we propose Gaussian Process Capped Thompson Sampling (GP-CTS), which caps posterior sample paths using a valid RKHS norm upper bound while preserving optimism. This modification successfully eliminates the logarithmic penalty and achieves the minimax-optimal rate $Θ(N^{-ν/d})$ for both discrete and continuous implementations. --- ### MolHop: Scaffold-Hopping Molecular Design with Agentic Reinforcement Learning Authors: Yanyi Su, Chenyan Wang, Yulei Gong, Yeyun Chen, Xiaohui Yang, Jun Cheng Published: 2026-10-14 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2610.19920v1 Molecular design is often formulated as an optimization problem over existing molecular spaces. However, molecular optimization and the discovery of structurally distinct replacement molecules represent two fundamentally different search problems. LLM-based molecular design agents typically perform the former, conducting local refinement within the neighborhood of existing structures to improve target properties through structural modifications. In contrast, the latter requires exploration of a broader chemical space to discover molecules with entirely different scaffolds while preserving the functional characteristics of reference molecules. We propose MolHop, an agentic framework for cross-scaffold molecular discovery that enables models to navigate and edit within a pre-organized chemical space. MolHop organizes approximately 1.6 million molecular scaffolds from commercial molecules into a coarse-to-fine hierarchical action space. The agent selects new exploration regions through scaffold navigation actions and adjusts candidate properties through molecular editing. The framework maintains historical trajectory memory to support long-horizon search processes. On drug molecule and commercial compound tasks, connecting Grok-4.6 and DeepSeek-Flash-V4.1 to the MolHop framework substantially improves their cross-scaffold joint success rates. We develop MolHop-9B through agentic reinforcement learning, and MolHop-9B outperforms generative baselines including GPT-5.6-sol, Claude-5-Opus, and Gemini-3.1-Pro, while generating candidate molecules with larger scaffold differences from the reference molecules. These results demonstrate that effective molecular discovery beyond the traditional optimization paradigm requires structured chemical and action spaces, enabling agents to explore new structural regions rather than merely refine existing molecules. --- ### Beyond DFT electronic band structures via dual-space orbital learning Authors: Minhyuk Kang, Beom Seok Kang, Nikolai Argatoff, Miguel Nouman, Danish Khan, William A. Goddard, Anima Anandkumar Published: 2026-10-14 Categories: cs.LG, cond-mat.mtrl-sci arXiv: https://arxiv.org/abs/2610.19919v1 Accurate electronic band structures are essential for predicting material properties, but high-fidelity calculations are computationally demanding. We introduce O-kRystal, a dual-space orbital-learning framework that uses a lower-cost Perdew-Burke-Ernzerhof (PBE) calculation as both an energy baseline and a source of structured orbital information. Combining PBE band energies with real- and reciprocal-space orbital matrices, O-kRystal predicts higher-level corrections for individual electronic states as a function of crystal momentum and spin. For states within 5 eV of the band-edge midpoint, it predicts Heyd-Scuseria-Ernzerhof (HSE06) band energies with a mean absolute error of $0.171 \pm 0.009$ eV across nearly 6,000 three-dimensional materials spanning insulating, metallic, magnetic, and nonmagnetic systems. On a separate benchmark of six strongly correlated materials, it reduces the corresponding PBE band-energy MAE by 68.7% and resolves orbital-dependent changes in the relative placement of transition-metal d and ligand p states. Despite training only at high-symmetry points, it maintains comparable accuracy at unseen points along band paths and reconstructs HSE06 densities of states from dense Brillouin-zone predictions. Finally, training O-kRystal on $G_0W_0$ quasiparticle energies for two-dimensional materials demonstrates its applicability across levels of theory and the potential of orbital learning for higher-fidelity, momentum-resolved electronic-structure prediction beyond density functional theory. --- ### Not All Criteria Matter: Importance-Aware Criteria Weighting for Rubric-Based LLM Evaluation Authors: Haocheng Yang, Licheng Pan, Yiming Dong, Xiaoxi Li, Yuan Lu, Zhenhao Chen, Jiajun Fan, Shuai Shao, Chunyuan Zheng, Hao Wang Published: 2026-10-14 Categories: cs.CL arXiv: https://arxiv.org/abs/2610.19915v1 Rubric-based evaluation assesses the responses of large language models (LLMs) by aggregating the verification results of explicit criteria, yet not all criteria matter equally. However, existing methods commonly aggregate the verification results with uniform weighting, which assigns minor and overlapping criteria the same weight as essential ones and thus produces misleading evaluation scores. Although some methods prompt an LLM to directly assign weights, direct weighting requires a holistic judgment over all criteria that LLMs struggle to make reliably. To address these limitations, we propose Importance-Aware Criteria Weighting (ICW), which derives the weights of criteria from pairwise importance comparisons made by an LLM. Specifically, ICW transforms the holistic judgment over all criteria into a series of pairwise importance comparisons, checks and revises inconsistent comparisons, and derives the weights from the principal eigenvector of the comparison matrix. Extensive experiments demonstrate that ICW consistently outperforms uniform weighting and competitive weighting methods, and translates into better policy models in downstream reinforcement learning. --- ### MUCO: A Mani-Unit-Structured Manipulation Framework for Compositional Long-Horizon Robot Tasks Authors: Xiangyi Wei, Hanzhe Yao, Jingwen Shang, Zhenhong Wu, Yang Li, Changbo Wang Published: 2026-10-14 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2610.19913v1 Learning and executing long-horizon robot manipulation as complete task trajectories raises data-collection costs, lim?its capability reuse, and causes local failures to cascade into later task stages. We introduce manipulation units (MUs) as reusable operations, each defined by a local goal, an initi?ation condition, and observable success and failure criteria. MUs are learned and verified individually. During execution, consecutive MUs remain coupled through shared environ?ment dynamics. We formulate MU-Structured Manipulation (MUSM) and realize it through MU-Centric Orchestration (MUCO). Given language and vision, MUCO constructs an MU sequence, matches each MU to compatible policy models in a frozen capability registry, and dispatches the model with the highest calibrated success rate. Visual verification checks MU outcomes against success and failure criteria and governs transitions, enabling retry or fallback to another model from the current state without re-executing verified MUs. Canonical keys enable checkpoint reuse for equivalent MUs across tasks without new demonstrations or updates. We evaluate MUCO on RLBench, RoboTwin 2.0, and five real-world tasks. MUCO achieves higher complete-task success and verified progress, stronger failure containment, and lower error propagation than baselines. Results demonstrate cross-task MU reuse and het?erogeneous policy composition without joint training, shared backbones, or common action representations. --- ### Quantifying Annotation Agreement in Object Detection with Optimal Transport Authors: Kirill Prokhorov, Alexandr A. Kalinin Published: 2026-10-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2610.19912v1 Reliable validation of machine learning models is a crucial step in designing robust and trustworthy systems. Yet validation relies on annotated test sets treated as ground truth, even though annotations can be noisy or ambiguous. In object detection, sources may disagree on object presence, box location, or class label. Standard agreement metrics usually binarize box overlap before computing a summary score, obscuring the degree and structure of disagreement between annotation sets. Here, we study an Optimal Transport (OT)-based metric for measuring agreement between object detection annotations. Instead of binarizing bounding box overlap, this metric assigns graded costs to matched and unmatched labeled boxes. Through theoretical analysis and experiments on real and synthetically perturbed annotations, we show that OT captures graded changes in agreement and distinguishes annotation pairs that thresholded metrics score identically. Our results suggest that OT provides a richer measure of annotation agreement in object detection, especially for domains with many small or ambiguous objects. Such agreement estimates can support more reliable validation when labels are noisy or conflicting. --- ### CARD: Closing the Teacher-Student Loop for Adversarial Robustness Distillation Authors: Zhi Li, Haowei Liu, Hongchen Yang, Xiaoxuan Wang, Song Gao, Zhenli He, Wei Zhou, Shaowen Yao Published: 2026-10-14 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2610.19911v1 Adversarial robustness distillation typically treats a robust teacher as a fixed knowledge source and transfers robustness to a compact student in a one-way manner. This paradigm terminates teacher-student interaction once the student is distilled, thereby discarding potentially useful feedback from the distilled student. However, exploiting such feedback is nontrivial because indiscriminate weak-to-strong transfer may propagate student errors to the teacher. To address this issue, we propose Cyclic Adversarial Robustness Distillation (CARD), a closed-loop framework that turns the distilled student into a feedback source for refining the robust teacher and subsequently transfers the teacher gains back to the same student. CARD first performs strong-to-weak adversarial distillation and then uses reliability-enhanced student feedback to refine the robust teacher. The refinement combines inverse adversarial student feedback with teacher-student discrepancy maximization, after which the refined teacher is used to continue training the same student under the unchanged base distillation objective. We further establish a sample-level sufficient condition for reliable weak-to-strong feedback transfer and derive a distribution-level upper bound on the refined teacher's risk. Experiments on CIFAR-10, CIFAR-100, and TinyImageNet-10 across multiple teacher-student architectures and base distillation methods show that CARD improves adversarial robustness in most evaluated settings while generally preserving clean accuracy. These results show that the distilled student can provide useful feedback to the teacher, enabling robustness transfer to proceed beyond the conventional one-way teacher-to-student paradigm. --- ### When Non-Adversarial Noise Hurts Learning: Benchmarking Skill Self-Evolution in Agents Authors: Tao Yu, Zhongtian Luo, Yuchen Long, Yiheng Lu, Yuxuan Zhou, Yanpei Gong, Siyi Cao, Yuqing Wu, Haopeng Jin, Zhenghao Zhang, Yujia Yang, Jingmin Zhu et al. Published: 2026-10-14 Categories: cs.LG arXiv: https://arxiv.org/abs/2610.19910v1 Large language model agents increasingly improve by distilling experience into reusable skills, but errors in that experience can persist in learned rules and affect later tasks. Existing evaluations largely examine clean learning or perturbations encountered after skills are frozen, leaving robustness during self-evolution insufficiently understood. We introduce RSE-Bench, to our knowledge the first benchmark for evaluating skill self-evolution under ordinary, non-adversarial noise. It spans four domains and six benchmarks, with controlled perturbations to instructions, environmental evidence, execution records, and outcome--record pairings that preserve the underlying tasks and native scoring rules. The protocol separates noise during learning from input-side noise at evaluation, testing frozen skill stores on held-out tasks. We also introduce \textsc{RSEClaw}, a method-agnostic framework supporting consistent noise injection and process diagnostics of actual exposure and skill updates. Across three methods and two backbones, noisy learning retains only $25.5\%$--$42.1\%$ of the aggregate gain achieved by clean self-evolution, depending on the noise type. Perturbations confined to execution records and outcome--record pairings cause average losses comparable to input-side noise, despite leaving the original executions and evaluations intact. Diagnostics further identify differences in candidate skill text that coexist with small changes in mean held-out performance. --- ### Sparse Transformer: a Controlled Testbed for Mechanistic Interpretability Authors: Anton Korznikov, Alexey Dontsov, Elena Tutubalina Published: 2026-10-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2610.19907v1 Mechanistic interpretability aims to decompose neural network computations into human-understandable features, yet progress is limited by the absence of ground-truth units to evaluate against. We address this with a Sparse Transformer (ST): a 5-layer language model trained on TinyStories whose wide, TopK-sparse MLP forces neuron specialization. We show that ST neurons are highly interpretable: they outperform a loss-matched dense transformer by 43% on AutoInterp, and circuits of a median of 12 neurons suffice to reproduce individual model predictions. This testbed allows us to directly compare feature learning methods: Transcoder, which approximates the MLP input-output mapping, recovers ST's neuron vectors with 52.6% higher cosine similarity than Sparse Autoencoder, which only reconstructs outputs. Furthermore, the ST enables a controlled study of polysemanticity: distilling its sparse MLP into a dense counterpart reveals how compression leads to feature superposition. These results show how a known reference basis sharpens evaluation and understanding, a role we hope the ST can serve for the field. --- ### Beyond Benchmark Accuracy: A Diagnostic Test for LLM Chemical Prediction Authors: Yanyi Su, Yu Zeng, Hongshuai Wang, Chenyan Wang, Zhifeng Gao, Xiaohui Yang Published: 2026-10-14 Categories: cs.CL arXiv: https://arxiv.org/abs/2610.19906v1 LLMs have achieved strong performance on chemical prediction benchmarks. However, existing evaluation methods typically provide target entity identity information directly, making it difficult to determine whether model predictions arise from transferable structure-property reasoning capabilities or rely on target-related contextual information or memory retrieval. We propose the Identity-Blinded Reasoning Diagnostic framework to evaluate chemical prediction models when target entity identity information is hidden. In this framework, the target is no longer directly defined by its entity identity, but instead by a known source entity and an explicit structural edit operation. This setting preserves the information required to determine the target while reducing the opportunity for models to access target identity cues. We evaluate this framework across six tasks spanning odor prediction, molecular property prediction, retrosynthesis, and reaction yield prediction. The results show that models generally achieve better prediction performance when source and target entity properties are consistent. Providing explicit source entity labels further strengthens this dependence. Moreover, while keeping the target entity, structural change, and ground-truth label unchanged, replacing only the source entity label causes some model predictions to systematically shift toward the substituted source labels. These results indicate that high benchmark accuracy in chemical prediction tasks does not necessarily imply that LLMs possess reliable structure-property reasoning capabilities. Our diagnostic framework provides a controlled evaluation approach for testing whether models can genuinely update predictions according to entity changes. The framework can be extended to other AI for Science tasks, particularly those where targets can be defined through transformations of known entities. --- ### Multimodal Medical Image Parsing with Task-Specific Low-Rank Adapters Authors: Taha Sabir, Alberto Montero Molina, Thinh Quang Ly, Veronika Dmitrenko, Mohammad Yaqub, Tausifa Jan Saleem Published: 2026-10-14 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2610.19901v1 Adapting a single vision-language model to medical image parsing requires training choices that reflect the demands of each task. We study task-specific adaptation with Qwen3.5-9B for FLARE 2026 Task 3, covering six task families and seven imaging modalities. Our system couples a shared, frozen 4-bit backbone with LoRA adapters selected from the input question. We customize adapter ranks and prompts, rebalance rare classification targets, and supplement regression training with dermatology classification examples. On the fixed internal holdout, the task-specific adapters achieve a dataset-macro balanced accuracy of 0.7791 for classification, a dataset-macro micro-F1 of 0.6286 for multi-label classification, a counting mean absolute error (MAE) of 11.73, a dataset-macro detection F1 of 0.8914, a dataset-macro regression MAE of 9.78, and a report-generation CRIMSON score of 0.8204. On the organizer's 1,703-example hidden-validation set, the submitted system achieves 0.8496 classification balanced accuracy, 0.5338 sample-averaged multi-label F1, and 14.5950 regression MAE. The final 30.1 GB container averages 3.1 seconds per case, including model loading, and produces identical outputs across repeated verification runs. The benefits vary across tasks, supporting the use of task-specific adapter configurations and image budgets within a shared medical vision-language model. Code is available at https://github.com/TahaSabir0/FLARE26-Hexapus. --- ### How GUI Agents Fail: GraceBench for Execution Stability Diagnosis Authors: Ridong Zhang, Shuxian Liang, Chaoqun Wan, Zhen Huang, Xian-Sheng Hua Published: 2026-10-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2610.19900v1 As GUI agents are increasingly deployed for mobile task automation, their reliability depends not only on final task outcomes, but also on whether they reach these outcomes through stable interaction trajectories. In automation workflows, unnecessary actions, detours, and execution errors can add noise to collected data, introduce irreversible side effects, or undermine trust, even when the agent eventually completes the task. It is therefore important to systematically diagnose the execution stability, including whether an agent deviates from a feasible path, enters a loop, becomes stagnant, or recovers from unstable actions. Existing benchmarks, however, primarily assess final task success or a small set of intermediate milestones, leaving step-level execution behavior unclear. To bridge this gap, we introduce GraceBench, an online evaluation framework that represents the reachable interaction space of mobile applications as UI Transition Graphs (UTGs), providing a structural reference for step-level execution diagnosis. To support reliable alignment between dynamic UI observations and graph states, GraceBench uses video-grounded trajectory evaluation over this graph-aligned representation, assessing process stability beyond final success with metrics such as deviation rate, loop burden rate, and recovery ability. GraceBench contains 169 tasks spanning 10 applications and constructs 10 UTGs with 2,909 canonical GUI states and 3,116 observed transitions. Across 3,718 trajectories, GraceBench shows that outcome scores hide execution instability: GPT-5.6 Sol achieves the highest success rate (SR) but the lowest recovery ability, 51.24% of successful runs still contain a deviation or loop, and trajectories with the same success rate (SR) and completion rate (CR) can differ substantially in recovery behavior. Code and data will be released on https://iaie-physicalai.github.io/GraceBench. --- ### Universal Sample-Based Embeddings of Higher-Order Objects Authors: Radovan Haluška, Christopher Solinas, Nathan R. Sturtevant, Martin Schmid, Michael Bowling Published: 2026-10-14 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2610.19899v1 Neural networks operate naturally on finite-dimensional vectors, yet many problems require reasoning about higher-order objects such as distributions, functions, and graphs. Such objects capture relationships within a ground set and rarely admit an explicit compact representation. Moreover, we typically want a compact representation not of a single higher-order object but of an entire family of such objects, including members unseen during training. While a few approaches exist for particular object types, none is designed to work across object types. We propose a unified framework that represents any higher-order object as a fixed-size embedding vector learned over a family of such objects. An encoder maps an object, seen only through a finite set of samples, to an embedding, and a domain-specific decoder supplies the training signal by modeling the object conditioned on that embedding. Since the encoder works only from samples, one architecture serves different object types. Because the decoder is a model already suited to single instances of the object type, the framework inherits that model's native operations and inductive biases. We instantiate the framework on distributions, functions, and graphs. In each case we show that a single trained model represents an entire family and generalizes to members unseen during training. Furthermore, we demonstrate the downstream utility of these embeddings: interpolating distributions in embedding space, solving differential equations in embedding space, and predicting unseen edges from a graph's embedding. --- ### DeepPV-MRF: End-to-End Partial-Volume Fraction Estimation from MR Fingerprinting Time Series Authors: Tianyi Ding, Yang Gao, Zhuang Xiong, Feng Liu, Martijn A. Cloos, Jin Jin, Hongfu Sun Published: 2026-10-14 Categories: eess.SP, cs.AI arXiv: https://arxiv.org/abs/2610.19894v1 Background: Estimating subvoxel tissue fractions from magnetic resonance fingerprinting (MRF) is challenging. Conventional dictionary matching and segmentation pipelines can propagate errors from intermediate parametric maps into tissue fraction estimates. Purpose: We introduce DeepPV-MRF, a deep learning framework that directly estimates voxel-wise tissue fractions from raw MRF signals. Methods: A cascaded spatial-temporal network was trained on subject-derived Bloch simulations with radial k-space undersampling. The network combines temporal encoding and spatial U-Net refinement to estimate cerebrospinal fluid (CSF), gray matter (GM), and white matter (WM) fractions without intermediate T1/T2 maps. We compared DeepPV-MRF with partial-volume dictionary matching (PV-DM) and pseudoinverse estimation (PV-Pinv) using simulated and in vivo data. Results: DeepPV-MRF outperformed both methods across acquisition settings. In single-shot simulations, CSF/WM/GM mean absolute errors were 0.0418/0.0744/0.0757, versus 0.1281/0.2443/0.2958 for PV-DM and 0.2021/0.2793/0.2630 for PV-Pinv. Corresponding structural similarity values were 0.960/0.916/0.922. In vivo results showed sharper anatomical boundaries and smoother tissue transitions, while cross-shot analysis indicated improved consistency across volunteers. Conclusions: DeepPV-MRF enables accurate and efficient direct tissue fraction estimation from MRF signals, particularly with limited shots. Further validation is needed before clinical deployment. --- ### Few-Shot Adaptation for Anomalous Sound Detection at Low False-Positive Rate Authors: Simone Lugani, Christof Pichler, Hannes Wegleiter, Christoph Feilmayr, Rodolfo Zunino, Edoardo Ragusa Published: 2026-10-14 Categories: cs.SD, cs.LG, eess.AS arXiv: https://arxiv.org/abs/2610.19893v1 Reliable fault detection is crucial for industrial machinery in large-scale facilities, where unexpected failures can cause costly downtime and production interruptions. Although acoustic monitoring provides a scalable solution for machine condition monitoring, its deployment is often constrained by high false-positive rates. We propose a two-step approach to automatic anomalous sound detection: first, when only normal operating audio samples are available, a regularized unsupervised model is deployed; then, after the first machine fault, a few-shot adaptation strategy exploits the anomalous samples acquired during the fault to improve the system. This adaptation increases the area under the precision-recall curve (AUPR) from 23.0 to 64.0 and improves detection in the low false-positive-rate region, with the partial area under the receiver operating characteristic curve (pAUROC) increasing from 39.6 to 72.8. We also show that covariance regularization improves both prediction accuracy and training stability in the unsupervised model. These results provide a practical strategy for acoustic anomaly detection systems, enabling improved accuracy and reduced false alarms. --- ### On the Limits of Representation-Based Test-Time Adaptation for Multimodal LLMs Authors: Mikołaj Piórczyński, Wojciech Masarczyk, Damian Sójka, Sebastian Cygert, Bartłomiej Twardowski Published: 2026-10-14 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2610.19892v1 Vision-language models suffer significant accuracy drops under common corruptions. While test-time adaptation (TTA) successfully mitigates these shifts in CLIP-like contrastive models, its application to generative multimodal large language models (MLLMs) remains underexplored and uniquely challenging due to their sequential token generation. In this work, we demonstrate that representation-based TTA fails to recover performance in MLLMs. Our experiments reveal that even when violating standard TTA assumptions by leveraging target data offline, sample-blind feature alignment recovers only marginal accuracy. We attribute this failure to the nature of corruption shifts: unlike CLIP-like architectures, which benefit from a shared domain-shift direction within a single pooled classification embedding, MLLM visual token sequence shifts are nearly orthogonal and span the full embedding space.These findings bound what any sample-blind, position-shared correction at a single encoder site can achieve, establishing that representation-space interventions alone might be insufficient to ensure corruption robustness in MLLMs. --- ### Frustratingly Easy Majority Voting Using Length Conditioning Authors: Fiorenzo Parascandolo, Enver Sangineto, Xiaoyi Yu, Wenhui Tan, Jipei He, Ruihua Song, Rita Cucchiara Published: 2026-10-14 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2610.19890v1 Majority Voting (MV) is a simple and widely used strategy which improves the reasoning accuracy of Large Language Models (LLMs) by allocating additional inference-time computation. Specifically, in MV multiple Chain-of-Thoughts (CoTs) are independently sampled and the most frequent final answer is selected. In this work, we propose Length-Conditioned Majority Voting (LCMV), a Test-Time Training approach which improves MV leveraging an empirical observation that has recently been established in several studies: longer CoTs tend to be associated with a lower probability of correctness. We formulate this inductive bias as a length-conditioned prior probability, which is modeled by a simple logistic function. Moreover, we adapt this prior at test time using only unlabeled testing data and without fine-tuning the LLM. Finally, the adapted prior is used to weight the vote of each sampled answer. Extensive experiments across multiple LLMs and reasoning benchmarks show that LCMV consistently improves over both standard MV and existing length-based and confidence-based aggregation strategies, while retaining the simplicity and model-agnostic nature of MV. Our code is available at: https://github.com/aimagelab/LCMV. --- ### Maximum likelihood, neural networks and posterior mean for inferring fractional Brownian motion parameters Authors: Daniel S. Han, Vaishnavi Ananthanarayanan, Matthew J. Simpson Published: 2026-10-14 Categories: q-bio.QM, cond-mat.stat-mech, physics.bio-ph, stat.ME, stat.ML arXiv: https://arxiv.org/abs/2610.19888v1 Fractional Brownian motion (fBM) is a model of anomalous diffusion parameterised by a Hurst exponent $H \in (0,1)$ and a generalised diffusion coefficient $D_H > 0$. We jointly estimate $H$ and $D_H$ using maximum likelihood (MLE) from density measurements with Poisson counting noise and single-particle trajectory with Gaussian localisation error. We show that one density measurement in time cannot infer both $H$ and $D_H$ and that several time points are needed, which matters most for inference precision. For trajectories, we show that a neural network trained with a squared error loss approximates the posterior mean of $H$, which can be computed exactly at negligible cost. Paired with the profile-likelihood estimate of $D_H$, the posterior mean has the lowest root mean squared error in both $H$ and $D_H$, except near $H=0$ and $H=1$ where MLE estimates $H$ better. In vesicle trajectories from two cell types, $H$ and $D_H$ form a continuum rather than distinct modes of motion. The posterior probability that each window within a track is subdiffusive or superdiffusive shows that $26\%$ of MRC5 fibroblast and $21\%$ of HeLa vesicle trajectories contain both subdiffusive and superdiffusive windows, against $2\%$ and $7\%$ expected from estimation uncertainty alone. Intracellular cargoes therefore switch between subdiffusion and superdiffusion over time rather than forming populations with fixed $H$. --- ### Occupancy Modeling via Drifting: One-Step Generation Enables Direct Policy Extraction Authors: Robin Uhrich, Marco Bagatella, Núria Armengol Urpí, Pavel Kolev, Georg Martius Published: 2026-10-14 Categories: cs.LG arXiv: https://arxiv.org/abs/2610.19883v1 The growing interest in behavior foundation models, combined with recent advancements in generative modeling, has led to an appealing recipe for reinforcement learning from unlabeled data: pre-train a generative model to model occupancies, i.e., discounted state visitations, on reward-free trajectories, such that a policy can be easily extracted once a small reward-labeled dataset becomes available. We decompose this recipe along three axes: (i) the degree to which the multi-modal structure of occupancies must be captured, (ii) the class of generative model, and (iii) the method used for policy extraction. Along each axis, we consider existing choices, propose simpler alternatives, and validate each decision experimentally. This analysis culminates in a novel, streamlined method that leverages drifting, a modern one-step generative sampler, to model the occupancy. We show that this enables policy extraction via direct differentiation through the occupancy model, and bypasses the need for critic distillation. Overall, we observe that this nimble setup leads to improvements in performance on standard benchmarks. --- ### BluePixel: Toward Native 4K Agentic Image Editing Authors: Shanyan Guan, Shiyu Li, Xin Li, Zhizhou Chen, Yunzhe Xu, Yuehao Liu, Ruixue Shen, Shiju Zhang, Min Jin, Qiang Liu, Junbo Liu, Penggang Shi et al. Published: 2026-10-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2610.19882v1 Instruction-driven image editing has reached native 4K, yet how such editors are trained remains undisclosed, and pretraining one from scratch is beyond most teams. The task itself compounds the difficulty: a single end-to-end editor cannot make precise parametric adjustments, match specialized experts, or interpret the vague instructions users actually write. We argue that 4K editing is a system problem rather than a single-model problem, and present BluePixel, which couples a native-4K editing model with an agent harness; the complete training recipe is disclosed in this paper. The model makes 4K training practical: it turns a 1K-pretrained editor into a native-4K one through progressive VAE adaptation that keeps 4K training as affordable as 1K, absorbs a long tail of vertical editing skills through continual learning, and refines the aesthetic-fidelity trade-off with reinforcement learning. Distillation via source-anchored target projection (SATP) then compresses 50 denoising steps to 4, completing a 4K edit in 8.7 s on a single Kunlunxin P800 XPU while preserving task success rate and background consistency. The harness, BluePixel Agent, makes editing controllable: it decomposes user instructions into specialist and general operations, verifies every result with the same reward model that guided training, and retries within bounded attempts. On VINS-4KEval++ (3,305 native-4K cases), BluePixel ranks second among seven models, behind only GPT Image 2 in non-edited-region consistency. ---