# Research Papers — 2026-09 · bundle page 28 of 266 > Full-text markdown of papers #1351–1400 of 13298 published in 2026-09, newest first. AI-readable mirrors of the full 13298-paper 2026-09 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-09/page/28 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/27.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/29.md) ### PE-OPSD: Internalizing Prompt Enhancement into Flow-matching Models via On-Policy Self-Distillation Authors: Mingfeng Lin, Chengfei Cai, Lin Xu, Chengqian Ma, Yuxiang Wei, Liang Han Published: 2026-09-29 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2609.36638v1 Text-to-image users often provide concise and underspecified prompts, whereas generative models benefit from detailed textual conditions for reliable instruction following. Existing systems bridge this gap with Prompt Enhancers (PEs) that rewrite raw prompts at inference time, introducing additional latency and leaving prompt elaboration external to the generator. We instead view enhanced prompts as privileged training information and ask whether their benefits can be internalized. We propose Prompt-Enhanced On-Policy Self-Distillation (PE-OPSD) for text-to-image flow-matching models. During training, a raw-prompt student follows its own generation trajectory, while an enhanced-prompt teacher provides vector-field targets at the states visited by the student. This on-policy supervision distills the behavior induced by enhanced prompts into the raw-prompt student without requiring additional text--image pairs. At inference, both the PE and teacher are removed, and the student generates directly from raw prompts. Across multiple model families, PEs, and benchmarks, PE-OPSD achieves the strongest aggregate prompt fidelity among the evaluated baselines, yields positive aggregate visual appeal gains, and retains the base-model inference efficiency. --- ### What Makes Recurrence Effective in Looped Language Models? Authors: Xinlin Zhuang, Siyuan Wang, Imran Razzak, Weiyang Liu Published: 2026-09-29 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2609.36636v1 Looped language models (LoopLMs) increase computational depth through parameter sharing, offering a path to scale inference computation without adding parameters. However, it remains unclear when additional recurrence is beneficial and how architectural choices affect its effectiveness. Through controlled experiments, we systematically examine (1) when recurrence helps, (2) where it should be applied, and (3) how its conditioning affects performance. Our evaluation covers inference budgets below, within, and beyond the training horizon under knowledge and reasoning tasks. (1) We find that recurrence can improve reasoning beyond the training horizon while degrading knowledge performance, but harder reasoning instances do not consistently benefit more. (2) Performance also depends on how distinct layers and recurrent iterations are allocated, showing that effective depth alone is insufficient to predict behavior. Non-recurrent output layers improve robustness to under-unrolling, while the preferred placement of input and output layers varies with inference budget. (3) Finally, we find that conventional initial-state injection offers limited robustness to varying recurrence depth. We therefore propose history-state injection as an alternative, and show that channel-wise history-state injection combined with timestep conditioning offers a low-cost and more effective design, better preserving knowledge under extended unrolling while improving robustness across inference budgets. Overall, our results clarify when recurrent computation helps, where it fails, and offer practical guidelines for designing LoopLMs across variable inference budgets. --- ### WitnessGym: Benchmarking Coding Agents on the Construction of Bug Witnesses Authors: Haomin Qi, Xiangzhe Xu, Yiming Huang, Jingbo Shang, Chengpeng Wang Published: 2026-09-29 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2609.36635v1 Bug validation asks a coding agent to produce an executable witness for a reported bug. The witness combines a concrete input with a testing harness and exposes faulty behavior during execution. Such evidence makes audit findings actionable, yet benchmark evaluation is difficult when cases reuse public historical bugs and witnesses or require manual construction. We present WitnessGym, an automated framework for constructing bug-validation benchmarks through bug injection. It injects bugs into test-reached paths of real projects, rebuilds each project, and retains cases exposed by a construction-time witness. Bug specifications and execution adapters allow extension to additional bug types and languages. Bug-preserving transformations vary the surrounding structure while preserving the witness behavior. Based on real-world Java projects with test suites, WitnessGym automatically constructs 1,300 benchmark cases. The injected patches resemble historical bug patches and are difficult for the two evaluated models to distinguish in blinded comparisons. We evaluate four coding agent frameworks in six framework/model pairings across bug types, execution contexts, and transformation depths. Witness construction remains difficult even when the bug pattern is known. Our framework, benchmark cases, and evaluation scripts are available. --- ### Physics-Aware Machine Unlearning for Cyber-Physical Systems Authors: Mohammad Zakaria Haider, Muhammad Nadeem, Mohammad Ashiqur Rahman Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.36633v1 This paper proposes a physics-guided gradient-ascent-based machine unlearning method that couples the forgetting signal with the physical residual of the target cyber-physical systems, ensuring that weight updates during unlearning are steered toward physically feasible regions of the weight space. The physics residual acts as a safety fence during gradient ascent: the model is steered away from the poisoned behavioral basin and simultaneously toward physics-compliant territory, rather than toward an arbitrary alternative that may still violate domain constraints. We evaluate the proposed method against four baselines: naive gradient ascent, exact unlearning, SISA, and full retraining on an IEEE 34-bus distribution system, driven by two physics-informed neural network-based distribution energy resource controllers and validated through high-fidelity OpenDSS power-flow co-simulation. From the evaluation, we found that our proposed physics-guided model simultaneously removes poison and restores physical compliance, which are essential for the safe deployment of safety-critical cyber-physical systems --- ### Distilling Agentic Systems: A Roadmap across Models, Artifacts, and Harnesses Authors: Ziluowen Luo, Senzhang Wang, Chaozhuo Li, Jun Yin, Hao Yan, Ming Cheng, Chenxu Wang, Songyang Liu, Litian Zhang, Qiwei Ye, Zheng Liu, Philip S. Yu Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36630v1 Modern agents increasingly rely on memories, tools, and execution logic, so their competence extends beyond model parameters. This shift exposes a limitation of conventional knowledge distillation, which asks how a student model imitates a teacher model. We define Agent Distillation as the persistent transfer of task-solving knowledge from a teacher agent to a student agent. Our study organizes the field by where transferred knowledge is retained: within the model, as artifacts, through the execution harness, or across substrates. This perspective separates transfer evidence from its outcome and clarifies how knowledge moves between agent components. We develop an evaluation framework that relates retention to causal contribution and deployed utility. Together, these contributions establish a foundation for the reliable, maintainable, and safe development of increasingly complex agentic systems. --- ### Beyond Binary Preferences: Graded Preference Optimization for Limb-Motion Captioning Authors: Yanan Wang, Tingsong Li, Kaixun Jiang, Chongyang Zhong, Chenwei Xoe, Zhaohe Liao Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.36628v1 Vision-Language Models (VLMs) can generate rich video captions, yet often misidentify which person performs an action or which limb is involved, particularly across camera cuts. Improving these details requires evaluation and training that distinguish missing information from incorrect assertions. We introduce FlexBench, a benchmark spanning 3,105 shots and 18,161 evaluation queries, with human-verified identities and systematic per-person coverage of fine-grained limb actions and states. Its reference-derived checklists support automated assessment of complete captions in their person and shot contexts. Our Graded Physical Alignment score (GPA) awards credit for correct content and deducts points for incorrect or fabricated actions, making these errors explicit in the aggregate score. Building on this rubric, we propose Graded Margin Direct Preference Optimization (GM-DPO), which assigns stronger preference margins and greater training weight to more severe action errors. Across three VLM backbones, GM-DPO achieves the highest substantive-action and GPA scores among the evaluated preference objectives, improving GPA over DPO by 2.02-3.40 points. On Qwen3-8B, it reduces the weighted hallucination rate by 21.3% relative to DPO. These gains accompany sustained long-form output, improved shot structure, and competitive performance on three additional multimodal benchmarks. --- ### Semantic Projection for Continual Self-Evolution of Language Agents Authors: Ziyu Liu, Jun Chen, Lixu Wang Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36626v1 Language-model agents increasingly rely on persistent natural-language skills to adapt beyond their frozen model parameters. When a shared skill is repeatedly revised from a non-stationary, heterogeneous task stream, however, improvements for new tasks can overwrite procedures needed for earlier ones. In continual learning, Orthogonal Gradient Descent (OGD) addresses analogous interference by projecting a new-task gradient onto a subspace that locally preserves prior predictions. Natural-language skill revisions, however, have neither gradients nor a canonical vector space in which such a projection can be performed. We introduce \emph{Semantic-Scope Projected Evolution} (SSPE), which transfers the functional principle of gradient projection from parameter space to behavior space. SSPE treats an unconstrained skill revision as a proposed update, identifies acquired capabilities with which it may interfere, and uses the observed gains and regressions to construct a compatible revision rather than merely rejecting the update. This enables one shared skill to evolve across latent and recurring task contexts without exposing semantic domain identities to the evolution model. Across controlled synthetic streams and heterogeneous real-agent benchmarks, SSPE improves final cross-domain competence and mitigates forgetting relative to strong skill-evolution baselines. The evolved skill also retains the strongest average performance after transfer to a different executor model. These results establish semantic projection as a promising principle for stable and adaptive self evolution of language agents. --- ### DualTrack: Synchronized speech-gesture generation via symmetric coupling of pretrained priors Authors: Yuanzhuo Hu, Zehan Liu, Xiaoyi Qin, Ming Li Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36624v1 Joint speech-gesture synthesis must coordinate two modalities despite limited paired data. Existing approaches often lack bidirectional interaction, have limited language coverage, or simplify body and finger representations. We present DualTrack, which couples pretrained speech and motion priors on a shared 12.5 Hz timeline. Causal adapters exchange previous-packet information, while current-state fusion coordinates the streams before they separately complete sixteen-codebook packets. We evaluate 43 BEAT2 recordings in four languages, with speakers held out from joint training and validation. On the shared English/Spanish inputs, without speech or motion prefixes, DualTrack achieves lower word error rate and full-motion Fréchet Gesture Distance, higher beat consistency and speech naturalness than the evaluated GELINA baseline. --- ### Neural Structural Reasoner: A Brain-inspired Architecture for Reasoning over Structured Knowledge Authors: Zixing Jia, Yuhang Pan, Ni Ji Published: 2026-09-29 Categories: cs.AI, q-bio.NC arXiv: https://arxiv.org/abs/2609.36620v1 Structural reasoning, the ability to recognize and make inferences over the relational structure between objects and concepts, is a hallmark of human cognition, yet prevailing methods often collapse relational topology into flat embeddings, cannot discover hidden structure and lack interpretability. We introduce Neural Structural Reasoner (NSR), a brain-inspired network that preserves relational structure directly in the connectivity and dynamics of coupled neuronal populations. NSR draws inspiration from three biological mechanisms: multi-layered architecture for encoding hierarchical knowledge, stable representations of entity and concepts, and path integration for input-driven state inference. At query time, NSR parallelizes computation over candidate relational structures and leverages confidence-weighted scores to perform link prediction. Across standard knowledge-graph benchmarks, NSR achieves competitive accuracy without leading on every dataset, and has lower reported training times than several neural baselines. Because reasoning is implemented through sequences of human-readable neuron activations, NSR affords native interpretability by tracking intermediate inference steps. The model further extracts latent relational hierarchies and compositional rules, demonstrating the brain-inspired architecture as an effective, efficient, and highly interpretable substrate for structural reasoning. --- ### SemPSG: A Semantic Channel-Aware Foundation Model for Polysomnography Analysis Authors: Junyu Chen, Chenxi Liu, Shiqin Tang, Hao Miao, Wanyun Ling, Ziyue Li, Hongbin Liu, Gaofeng Meng Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.36619v1 Polysomnography (PSG) integrates multiple physiological signals to provide a comprehensive characterization of human sleep, yet its heterogeneous channel configurations across centers pose substantial challenges for transferable representation learning. Existing foundation models mainly focus on physiological modeling or temporal learning, while channel identity is often treated as a fixed structural index, overlooking the physiological semantics encoded by signal modality and reference configuration. To this end, we propose SemPSG, a Semantic channel-aware foundation model for heterogeneous PSG analysis. SemPSG explicitly represents the physiological semantics of channel identity and incorporates them into both signal representation learning and channel aggregation, enabling flexible modeling across diverse data configurations. Specifically, a semantic-conditioned time-series encoder captures signal-specific temporal dynamics and cross-signal interactions, while a multi-view image encoder extracts complementary time-frequency and morphological patterns from the same physiological recordings. We evaluate SemPSG on sleep and health-related tasks, including sleep staging, sleep-disorder breathing analysis, disease prediction, cognition and emotion recognition, and demographic estimation. Extensive experiments demonstrate consistent improvements over both general-purpose time series foundation models and PSG-specific foundation models, together with generalization across heterogeneous datasets across diverse channel configurations. --- ### Generating Edit-Inducing Questions for AI Research Manuscripts Authors: Sebastian Joseph, Zichao Wang, Jennifer Healey, Alexa Siu, Junyi Jessy Li, Ani Nenkova Published: 2026-09-29 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.36617v1 We study the ability of LLMs to generate edit-inducing questions whose answer will improve a paper draft. On a dataset of paired submission and camera-ready papers from ICLR and NeurIPS, we compare the helpfulness of questions from GPT models with or without full paper context to that of human reviewers. GPT produces more edit-inducing questions and its questions are associated with more extensive edits and cover a broader range of edited content compared to questions from reviewers. However, a much smaller percentage of the GPT questions are edit-inducing. Our analyses confirm that automated questions can be beneficial to authors and highlight an example task where proper attending to long context deteriorates reasoning model ability to produce helpful output. --- ### CrossTimeEdit: A Decade-Spanning Cross-View Dataset and Reward-Guided Editing for Historical Street-View Generation Authors: Hanwen Lu, Jun He, Mingjia Yang, Hao Wei, Jinhao Huang, Yi Lin, Xiang Zhang Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.36616v1 Historical street-view imagery records urban evolution, but uneven coverage leaves substantial gaps in historical records. Generating plausible past appearances requires restoring changed structures while preserving persistent scene content. We construct VIGOR-his, a decade-spanning cross-view dataset containing 43,653 location-level quadruplets across 11 cities on three continents. Its automated pipeline performs spatial pairing, consistency screening, change classification, and the generation and validation of satellite-based change descriptions and local editing instructions. Based on VIGOR-his, we propose CrossTimeEdit, a model that reformulates historical street-view generation as editing, using recent street views to constrain viewpoint and unchanged appearance and temporal satellite differences as change evidence. Starting from FLUX.2 (Klein) 4B, we train CrossTimeEdit through supervised fine-tuning (SFT) followed by online reinforcement learning (RL). We design three street-view editing criteria, namely Instruction Alignment (IA), Background Preservation (BP), and Quality and Physical Plausibility (QP), as both RL reward dimensions and evaluation metrics. We optimize this multi-reward objective using Within Group Relative Policy Optimization for flow-matching models (Flow-GRPO) with Group reward-Decoupled Normalization Policy Optimization (GDPO), which normalizes each reward dimension before aggregation. CrossTimeEdit improves overall performance across the three editing criteria by 17.12\% over the pretrained baseline and outperforms cross-view generation models in scene consistency, visual realism, and perceptual quality. The implementation code, dataset, and model weights are available at https://luhanwen67.github.io/CrossTimeEdit-release/. --- ### CI-PINN: Causal Integral Physics-Informed Neural Network for Solving Evolution Equations Authors: Xiaodong Feng, Ziyu Sun, Tao Tang, Xiaoliang Wan, Tao Zhou Published: 2026-09-29 Categories: math.NA, cs.LG arXiv: https://arxiv.org/abs/2609.36615v1 Physics-informed neural networks (PINNs) solve partial differential equations (PDEs) by incorporating governing physical laws into the training loss. For evolution equations, however, their conventional pointwise space--time representation does not explicitly encode temporal dependence, which can hinder accurate prediction. To mitigate this limitation, this work proposes a novel neural architecture termed a causal integral neural network (CinNet). The core module of CinNet is a Volterra-type causal integral term, which aggregates historical features to encode temporal dependence, thereby incorporating temporal causality at the architectural level rather than through training-level modifications as in many existing methods. Building on CinNet, we further develop a causal integral physics-informed neural network (CI-PINN) for solving evolution equations. Extensive numerical experiments on benchmark evolution equations demonstrate that the presented method outperforms various baseline PINN variants in terms of solution accuracy, with pronounced superiority under sparse-collocation scenarios. Additional empirical analyses show that CI-PINN exhibits low sensitivity to hyperparameter choices, while ablation studies confirm the effectiveness of the proposed network components. --- ### Selective Elicitation as a Commercial Influence Channel: A Reproducible Synthetic Shopping-Agent Stress Test Authors: Jiapeng Li Published: 2026-09-29 Categories: cs.LG, cs.CY arXiv: https://arxiv.org/abs/2609.36614v1 A commercial incentive need not enter the final ranking algorithm to affect a shopping assistant's recommendation: it may instead influence which preference question the assistant asks. We make this distinction experimentally observable in a deliberately small, synthetic setting. Each task has two products, three verified numerical attributes, a price limit, and a private fixed preference vector. An honest simulated user answers one pairwise question. A separate recommender receives the products and this answer but not the sponsorship assignment. We contrast a neutral question, a soft commercial instruction, and an explicitly adversarial instruction to ask about the sponsor's advantage while omitting the rival's advantage. Across 40 held-out sponsorship-assignment cases (20 distinct catalog-preference contexts), the soft instruction changes no selections. The targeted instruction raises sponsored selection by 0.30 and reduces mean synthetic utility by 0.0547 relative to neutral questioning (95% context-bootstrap interval (-0.0828, -0.0291)) for one language-model recommender. A fixed Bayesian recommender shows a similar effect; a second model makes the same choices on all 120 frozen question-answer inputs. A terminal-answer consistency judge rates all 20 sampled targeted answers consistent, although five have synthetic regret above 0.05; a separate question-coverage dimension flags their one-sided elicitation. A robust partial-preference certificate remains valid under the stipulated synthetic utility but certifies only 16 of 40 targeted cases and is not better than asking a neutral question directly. These results establish neither typical behavior under advertising incentives nor effects on actual consumers. --- ### Do LLMs Really Forget? Hidden-State Leakage in Model Unlearning and How to Fix it Authors: Hadi Reisizadeh, Jiajun Ruan, Sijia Liu, Mingyi Hong Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.36612v1 Unlearning in large language models (LLMs) is typically evaluated at the output level, where a model appears to suppress sensitive or undesirable content. In this work, we show that such evaluations can create an illusion of forgetting: even when output-level leakage is eliminated, sensitive information can remain encoded in the model's hidden representations. We first provide a theoretical analysis establishing a fundamental separation between output suppression and representational erasure. Specifically, we show that the decoder can be made arbitrarily insensitive to sensitive directions, driving output-level leakage to zero, while the hidden representations retain the underlying information. To empirically validate this phenomenon, we train generative probe decoders on hidden states across transformer layers, enabling layer-wise measurement of information leakage. Across three widely used benchmarks, TOFU, MUSE, and WMDP, and state-of-the-art unlearning methods, we find that substantial sensitive information remains recoverable from hidden representations, even when standard output-level metrics indicate successful unlearning. To address this gap, we propose Probe-Adversarial Representation Suppression (PARS), an unlearning objective that adversarially minimizes the extractable information from hidden representations. PARS directly targets representational leakage and provides significantly stronger guarantees of erasure under adversarial probing and relearning attacks, outperforming all evaluated baselines. Our results highlight a fundamental limitation of existing unlearning paradigms and suggest that true forgetting in LLMs requires controlling not only model outputs, but also the information encoded in hidden representations. Codes are available at https://github.com/OptimAI-Lab/HiddenStateUnlearning. --- ### Multi-Channel Mitigation of Source-Trust Shortcuts in Fact-Checking RL Agents Authors: Jianchang Su, Yiwei Yang, Wei Zhang Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36611v1 Retrieval-augmented fact-checkers often receive a reliability label, such as HIGH or LOW trust, for each evidence source. These labels should adjust the model's confidence and its decision to search for more evidence, while the verdict should follow the evidence content. We introduce TrustSwap, a counterfactual test that swaps, lowers, or removes source labels while keeping every evidence text fixed, and measures its three output channels (the verdict, the confidence, and the search decision) separately. Across untrained and RL-trained models at two scales, three datasets, and two prompts, confidence and search respond to the labels as intended in 49 of 50 comparisons, yet a label change alone alters 4-23% of confident verdicts for Qwen3 models and up to 50% for an existing RL-trained fact-checker. Standard GRPO fine-tuning amplifies this shortcut at 8B in all six settings. To reduce it, we propose trust-swap augmentation (TSA), which trains GRPO on each claim with both its original and its label-swapped evidence under the same gold verdict. At 4B, TSA lowers the verdict flip rate by 7-35% (relative) in four of six settings, keeps accuracy and the intended confidence and search responses, outperforms reward-based alternatives in the main setting, and carries over to an unseen label-removal perturbation. An added consistency reward helps on the trained-on swap but not on unseen perturbations. At 8B, TSA's effect is not detectable, which makes scale the main open question. --- ### Communication-Efficient Agnostic Federated Learning via Faster Convergence and Compression Authors: Haomin Bai, Junyan Sun, Sifan Yang, Bo Xue, Lijun Zhang Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.36610v1 Agnostic federated learning (AFL) seeks a model that performs reliably across $m$ heterogeneous workers, but communication remains a bottleneck. We improve communication efficiency by reducing the number of synchronization rounds via faster convergence and the communication cost per round via compression. We first propose AFL-BR, which updates the dual weights over workers using online mirror ascent with KL divergence and blockwise restarts. It achieves an $O((\log m)^{1/4}T^{-1/8})$ stationarity rate after $T$ update rounds, reducing the $m$-dependence of the synchronization rounds required for convergence from polynomial to logarithmic order. Building on AFL-BR, we develop AFL-Com by applying bidirectional compression with error feedback (EF). Instead of compressing local gradients, workers apply EF to their dual-weighted gradients, enabling direct control of the aggregated compression error under time-varying weights. We then establish an $O((δ^{-1}+(\log m)^{1/4})T^{-1/8})$ stationarity rate for AFL-Com under general $δ$-approximate compressors and improve the $δ$-dependence from $δ^{-1}$ to $δ^{-1/2}$ for additive-and-idempotent compressors with shared randomness (SR). With suitable compression levels, AFL-Com retains the same convergence rate as AFL-BR at a lower per-round communication cost, yielding reductions in total communication complexity by factors of $(\log m)^{1/4}$ with Top-$k$ and $(\log m)^{1/2}$ with Rand-$k$ and SR. Experiments validate the improved synchronization and communication efficiency of our methods. --- ### Best Practices in EEG Analysis: Preprocessing, Modeling, and Machine Learning Authors: Parsa Razmara, Woojae Jeong, Aditya Kommineni, Raymundo Cassani, Richard Leahy, Takfarinas Medani Published: 2026-09-29 Categories: eess.SP, cs.LG arXiv: https://arxiv.org/abs/2609.36609v1 Electroencephalography (EEG) analysis requires careful choices in preprocessing, statistical modeling, and machine learning because EEG signals are highly susceptible to artifacts, volume conduction, low signal-to-noise ratio, and substantial inter-subject variability. This chapter provides a practical and methodological guide to modern EEG analysis, spanning EEG preprocessing, artifact removal, filtering, bad-channel detection and interpolation, re-referencing, independent component analysis (ICA), and preprocessing of simultaneous EEG-fMRI recordings. We review major approaches for computational EEG analysis, including event-related potentials (ERPs), time-frequency analysis, functional and effective connectivity, source localization, multivariate decoding, permutation testing, and multiple-comparison correction. We then examine machine-learning methods for EEG, from feature-based classifiers to deep learning and emerging EEG foundation models, with emphasis on cross-subject generalization, limited-data regimes, data leakage, evaluation metrics, and fair benchmarking. Reproducibility is treated as a core requirement throughout, including transparent preprocessing, BIDS-EEG data organization, standardized derivatives, preservation of raw data, and FAIR data practices. The chapter is intended as a practical reference for researchers developing reliable, interpretable, and reproducible EEG analysis and machine-learning pipelines. --- ### Act First, Reason Later: Accelerating On-Policy Distillation for Multi-Turn Agents via Reference-Conditioned Inverse Dynamics Authors: Zubin Zheng, Jiahao Wu, Shaofeng Zhang, Zhirui Zhang, Yew-Soon Ong, Shengcai Liu Published: 2026-09-29 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2609.36608v1 On-policy distillation (OPD) trains multi-turn language agents with dense teacher supervision on student-generated responses. However, standard think-then-act rollouts require lengthy reasoning before each short action, delaying environment transitions and experience collection. Generating actions directly reduces this delay but can degrade rollout quality. To address this, we propose ActFirst-OPD, an act-first, reason-later training framework that decouples environment interaction from full-response generation. The student infers and executes actions through reference-conditioned inverse dynamics using its current interaction context and a reference next observation, and switches to autonomous next-action prediction when the resulting transition deviates from the reference trajectory. From the collected interaction contexts, the student asynchronously generates full think-then-act responses for token-level teacher supervision. Experiments across 0.6B-, 1.7B-, and 4B-parameter Qwen3 students show that ActFirst-OPD achieves average wall-clock training speedups of $2.3\times$ on ALFWorld, $1.8\times$ on WebShop, and $4.9\times$ on ScienceWorld over Vanilla OPD. It matches or exceeds all compared OPD baselines in mean task success rate across eight of nine benchmark-model settings. These results demonstrate that reasoning need not block acting during multi-turn agent distillation. --- ### Pixel-wise Exposure for Highly Robust In-Vehicle Remote-PPG Authors: Jieying Wang, Xinqi Cai, Caifeng Shan, Wenjin Wang Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.36607v1 Remote photoplethysmography (rPPG) offers a promising non-contact solution for heart rate monitoring, yet its real-world robustness is fundamentally limited by an inherent hardware limitation: existing camera exposure control paradigms, whether fixed or auto-exposure, impose a uniform exposure time across all pixels within a frame. In high-dynamic-range scenes such as automotive cabins with strong directional sunlight, this spatially invariant exposure constraint inevitably leads to localized facial overexposure or underexposure, irreversibly corrupting the subtle pulsatile signals essential for rPPG at the point of capture, a physical degradation that no downstream algorithm can recover. To overcome this bottleneck, we propose PixExpo (Pixel-wise Exposure), a "temporal-for-spatial" framework that sequentially captures frames under a predefined cyclic exposure schedule and performs non-iterative pixel-wise fusion. At each pixel location, PixExpo selects the observation closest to an rPPG-motivated target intensity. This criterion seeks to reduce local saturation and severe underexposure rather than optimize perceptual appearance. PixExpo requires no sensor modification but assumes programmable frame-level exposure control. We validate the proposed PixExpo framework using our newly introduced MEX-Drive dataset, comprising 48 participants under real-world driving conditions. Experimental results demonstrate that PixExpo outperforms manufacture-default auto-exposure methods, reducing the mean absolute error (MAE) by 7.21 bpm (from 13.94 to 6.73 bpm) and increasing the success rate by 37.29 percentage points (from 25.95% to 63.24%) across challenging driving scenarios. --- ### SAKI: Maximal-Coupling-Routed Teacher Supervision for On-Policy Distillation Authors: Miteto Wei, Xiaohan Wang, Zehao Chen, Jiajun Chai, Sichao Liu, Li Wang, Haoyuan Xu, Zhaoyu Hu, Wei Lin, Guojun Yin Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36601v1 On-policy distillation (OPD) reduces train-test state mismatch by training a student on its own generated trajectories, but weak students may visit teacher-misaligned prefixes where supervision is less representative. We introduce SAKI (Supervision Allocation with KL-constrained Interpolation), which combines a KL-constrained teacher-guided rollout with maximal coupling and reuses realized accept/correction events to route token-level supervision. Accepted positions retain sampled-token reverse-KL supervision, while correction positions receive direct supervision on the teacher's highest-probability token. Under maximal coupling, the correction probability is exactly TV(p_t, q_t), so the same trust-region radius controls rollout deviation and upper-bounds intervention and specialized-supervision frequency. We further implement an engine-resident speculative verifier that preserves the exact-q trajectory distribution and coupling semantics while improving matched-workload rollout throughput by 4.22x. Across seven mathematical reasoning benchmarks, SAKI improves the matched teacher-guided baseline in Mean@8 and Pass@8 for both 1.7B and 0.6B students. Placement controls and fixed-prefix analysis further support correction-triggered routing as a conflict-adaptive supervision signal. --- ### Second-Moment Stochastic Approximation Methods Authors: Tao Jiang, Lin Xiao Published: 2026-09-29 Categories: math.OC, cs.AI, stat.ML arXiv: https://arxiv.org/abs/2609.36600v1 Classical stochastic approximation methods rely on estimators of the first moment (mean) of a random regression function. We study methods that employ estimators of both the first and the second moments, which include modern deep-learning optimizers such as Adam and Muon as special cases. We derive second-moment stochastic approximation methods through the lens of optimal preconditioning for solving matrix equations, and develop a two-stage framework for their convergence analysis. The first stage focuses on the analysis of conceptual (impractical) methods that rely on the exact first and second moments. In the second stage, we replace the exact moments with their respective estimators, and invoke Dvoretzky's theorem to show that the resulting practical methods converge almost surely to a neighborhood of the target solution. The size of the neighborhood depends on the biases and variances of the first- and second-moment estimators. We derive concrete bounds for Muon and a spectral variant of Adam that determine the radius of their neighborhood of convergence. --- ### Scaling Video Generation for Reasoning: At What Cost? Authors: Weihang Guo, Xiaoyu Wu, Yifei Wang, Niloofar Mireshghallah, Lydia E. Kavraki Published: 2026-09-29 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.36599v1 We study whether scaling video generation enables models to reason about hidden information from the past frames, and at what computational cost. Our controlled benchmark requires predicting nine prescribed moves of an initially solved 2x2x2 Rubik's Cube from a fixed view of three faces. Correct predictions require inferring how actions change hidden states, and the simulator provides exact ground truth for evaluation. Models learn plausible cube geometry early, while correct sticker configurations require substantially more training. Although validation MSE follows approximate power-law scaling, lower MSE loss does not reliably indicate downstream reasoning capabilities. Smaller autoregressive models achieve higher state accuracy with limited compute, while larger models reach higher accuracy after more training. At roughly 0.1 PF-days, the 70M-parameter model correctly predicts the visible sticker configuration in 44.6% of post-action frames, compared with 0.3% for the 1B model, which reaches 83.7% at 3.14 PF-days. Symbolic state supervision raises the 20M model's frame accuracy from 31.1% to 67.3% at the same training-data budget, suggesting that learning representations of state changes can complement scaling. --- ### Beyond Legibility: Benchmarking Visual Text Rendering and In-Place Editing in Unified Video Generation Authors: Ziying Zhang, Litao Li, Junchao Liao, Tianyi Zeng, Siyu Zhu, Long Qin, Zhenghao Zhang Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.36598v1 A video can exhibit convincing motion and photorealism yet fail immediately when visual text collapses. Unlike generic scene content, visual text is unforgiving in video generation: minor stroke corruption, temporal instability, or editing errors instantly break legibility and realism. Existing benchmarks overlook this challenge by treating text as incidental or using static OCR metrics that ignore temporal dynamics. We introduce VidScribe, a unified diagnostic benchmark spanning four generation regimes: writing from language (T2V), transferring text identity from a reference (R2V), sustaining text under dynamics (I2V), and localized text editing (V2V). VidScribe contains 803 human-verified samples across a 12-axis conditionally orthogonal factor space covering Intrinsic Text Properties, Physical Imaging Conditions, and Temporal Behavior. For reliable evaluation, we build a track-grounded, gated suite with 11 shared metrics and 2 task-specific probes under strict measurability conditions. Benchmarking 11 commercial and open-source systems shows that video text capability is non-monolithic, with content recognition decoupled from stroke-level glyph correctness. Performance is highly task-asymmetric: I2V sustains text most reliably, whereas V2V editing is the primary bottleneck. Counter-intuitively, degradation concentrates on a small subset of text-centric structural and temporal factors rather than adverse imaging conditions. Further probes show that visual references improve glyph and typographic fidelity rather than content accuracy, while localized editing fails to isolate target text without corrupting undeclared source text. Beyond evaluation, VidScribe also provides an actionable training signal, where benchmark-aligned preference optimization measurably improves visual text generation. https://huggingface.co/datasets/Vicky0720/VidScribe. --- ### Simple Agentic Memory for Generalist Robot Policies Authors: Yuyou Zhang, Yunbei Zhang, Miao Li, Janet Wang, Zijian Jin, Shilong Liu, Ding Zhao Published: 2026-09-29 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.36595v1 Visual-memory systems commonly retain or compress past observations. Robot control additionally requires interaction-derived state that no individual frame may explicitly represent, such as persistent identity relations, accumulated progress, or ordered procedures. We introduce Simple Agentic Robot Memory (SimpleARM), a training-free memory layer for frozen generalist robot policies. From the task instruction, SimpleARM specifies what to monitor; frozen perceptual tools maintain compact typed state online; structured access retrieves that state only when a proposed subgoal depends on history; and current-view grounding resolves recalled entities before execution. We evaluate SimpleARM on RoboMME, a benchmark of memory-dependent robot manipulation tasks that require history information no longer available in the current observation. Across all 16 tasks and three policy seeds, SimpleARM achieves 67.17% mean success, compared with 44.51% for the strongest non-oracle baseline. Matched ablations show mechanism specificity: removing relation, reference, progress, or route state produces large losses where the affected state is retrieved for control, while largely sparing other tasks. These results support a state-based view of robot memory: effective memory for control is not simply retained visual history, but compact task-relevant state derived from the interaction history. --- ### Optimal detection of general moment changes: Simultaneous mean and covariance change detection and beyond Authors: Xiaokai Luo, Chenghao Xu, Haotian Xu, Carlos Misael Madrid Padilla, Daren Wang Published: 2026-09-29 Categories: stat.ME, math.ST, stat.ML arXiv: https://arxiv.org/abs/2609.36594v1 We study multiple change-point detection in multivariate time series whose distributions change in a piecewise constant manner. Distributional changes can manifest across different moment orders, from shifts in the mean and covariance to changes in higher-order moments. Higher-order moments capture increasingly rich distributional features but become difficult to estimate in high dimensions. Our tensor representation unifies moments of different orders within a common linear algebraic framework, enabling a new method to detect changes in moments of all orders up to a prescribed fixed order $p$. The resulting procedure accommodates temporal dependence and allows the dimension of the time series to grow with the sample size. Under suitable regularity conditions, the proposed procedure achieves a localization error rate that matches a newly developed minimax lower bound. We further derive limiting distributions under both nonvanishing and vanishing moment jumps and construct asymptotically valid confidence intervals in the vanishing-jump regime. Numerical experiments and real-data analyses demonstrate the method's effectiveness in detecting moment changes and a range of distributional shifts. --- ### Text2Sim: Agentic Physics-Based Simulation Generation with Distilled Expertise Authors: Xiaoyu Xiong, Tsun-Hsuan Wang, Yi-Ling Qiao, Tao Du, Minchen Li Published: 2026-09-29 Categories: cs.GR, cs.AI arXiv: https://arxiv.org/abs/2609.36593v1 Creating diverse physical simulations remains labor-intensive because assets, layout, physical parameters, motion, control, and rendering must be designed and debugged jointly. We present Text2Sim, a simulation-specialized agentic pipeline that converts a text-only request into an executable, editable dynamic case. Built on Genesis, Text2Sim uses a hierarchical agentic structure that combines a Planner with specialized Writers, asset-generation tools, and an independent Critic. Compact skills (Debug Cards) distilled from graphics demonstrations provide role-specific physical guidance for execution-based repair. We evaluate physical quality, visual quality, and human preference on 42 held-out prompts spanning rigid, articulated, deformable, and cloth phenomena, with a paper-level split between experience construction and evaluation. We design automatic physical and visual scorers to evaluate the quality of the results, and Text2Sim achieves higher scores than all four state-of-the-art baselines on both metrics. In blinded user studies with these baselines, significantly more participants prefer Text2Sim than prefer the baselines, which is consistent with the results from our automatic scorers. The pipeline also supports a broad range of downstream applications; we select dataset construction and extension to multimodal input as two representative examples. We will release the code, the Debug Card library, and a dataset of generated cases, each pairing the text prompt and rendered video with the executable program, assets, physical parameters, controls, and recorded states. --- ### SEED: Self-Speculative Decoding via Implicit Encoder-Decoder Authors: Hankun Lin, Patrick Pynadath, Ruqi Zhang Published: 2026-09-29 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.36590v1 Self-speculative decoding accelerates large language model (LLM) inference by drafting tokens from the target model itself, but faces a sharp tradeoff between the quality and cost of the draft. Early-exit methods produce drafts cheaply by terminating computation at intermediate layers, but forgo the deeper representations that later layers provide and thus suffer in draft quality. Multi-token prediction preserves draft quality by emitting from the model's final hidden states, but pays for a full forward pass to produce those states at every drafting step. We propose self-speculative encoder-decoder (SEED), a self-speculative method that obtains high-quality drafts cheaply by reusing the deep contextual representations already computed during verification. We reinterpret the standard decoder-only transformer as an implicit encoder-decoder: the first layers (encoder) build deep contextual representations, and the last few layers (decoder) emit tokens from them. Encoding and verification are merged into a single step: verification is performed by the full encoder-decoder, and the contextual representations of the verified prefix are cached for reuse during drafting. Drafting is therefore very fast: between verifications, the lightweight decoder drafts multiple tokens autoregressively, each conditioned on the cached representations and on preceding drafts. Experiments across multiple benchmarks show that SEED achieves up to 2.7$\times$ average speedup on 4B-scale models, outperforming both early-exit and MTP-style self-speculative baselines and running 28% faster than the state-of-the-art EAGLE-3, while preserving or even improving the generation quality of standard autoregressive fine-tuning. Code is available at https://github.com/lhk2004/SEED. --- ### Cooperative Multi-Agent Vision-Language-Action Models via Reinforced Fine Tuning Authors: Ruixiao Xu, Wong Lik Hang Kenny, Zhiqian Liu, Jianing Guo, Hanxiao Li, Kejian Shi, Shuning Zhang, Pu Feng, Yongjia Ma, Yuqing Ma, Kai Chen, Qi Dou et al. Published: 2026-09-29 Categories: cs.RO, cs.AI, cs.MA arXiv: https://arxiv.org/abs/2609.36588v1 We study reinforcement learning (RL) methods for cooperative multi-agent Vision-Language-Action (VLA) models. This problem is challenging because VLAs are pretrained on large-scale single-agent data and therefore lack the fine-grained coordination skills required for inter-robot collaboration. Supervised fine-tuning (SFT) on multi-robot demonstrations partially bridges this gap, but its performance is bounded by the demonstration data and cannot improve from its own experience. We present a three-stage reinforced fine-tuning (RFT) pipeline for multi-agent VLAs. First, initialization-aware data collection sweeps over initial configurations and invokes human demonstrations only when the pretrained VLA repeatedly fails, yielding robustness to initialization shift with reduced human cost. Second, offline credit-filtered tuning assigns credit to individual agents and fine-tunes on per-agent trajectories with positive advantage rather than on entire joint rollouts. Third, we find existing online RL for VLAs are less effective for hard multi-agent tasks, which we attribute to noisy co-exploration and unstable updates. We instead use online latent-space fine tuning, which freeze the VLA and perform RL in its latent noise space. We evaluate our multi-agent VLA with both $π_0$ and $π_{0.5}$ backbones across 11 tasks in RoboTwin, RoboFactory and real-world manipulation with two Franka robots. Our multi-agent VLA improves the average success rate by $+23.1\%$, $+16.4\%$, and $+44\%$ on RoboTwin, RoboFactory, and real-world tasks, respectively. Code available at https://anonymous.4open.science/r/mavla_rft-2BC0/. --- ### Learned Reporting Preferences in RLVR Can Conflict with the Current Request Authors: Yupeng Chang, Wenxuan Zhang, Yuan Wu Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.36587v1 Reinforcement learning with verifiable rewards (RLVR) has become a prominent approach for improving language-model performance on reasoning tasks using automatically checked answers. Yet convention-matched evaluation cannot reveal whether reinforcing one reporting convention reduces adherence to a different request that the initial policy already follows. To test this, we train matched policies under two reporting conventions and evaluate each policy under both current requests, using the same initial policy as a shared reference. We complement this crossed design with controlled interventions and independent human calibration. On GSM8K, boxed-format RLVR reduces the fraction of Qwen2.5-7B responses containing the requested hash-format payload by 35.33--74.37 percentage points relative to a 95.45% initial baseline in four of five training seeds; the fifth improves by 2.50 points. In the four deteriorating runs, almost every response that omits the requested payload instead retains the trained boxed convention, and the same four seeds deteriorate under two fixed paraphrases. Changing only the final-answer marker in supervised targets reverses which reporting convention the model prefers across three seeds, providing controlled evidence that this preference is learnable. Across three settings with independent human calibration, gains under a convention-sensitive scorer exceed the corresponding gains in committed-answer correctness, i.e., the correctness of the answer the model actually commits to. Together, these results separate three distinct post-training outcomes: learned reporting preference, current-request adherence, and committed-answer correctness. They show that convention-matched accuracy alone does not fully characterize post-training behavior and motivate evaluating current-request adherence alongside convention-matched task accuracy. --- ### Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It Authors: Zehao Jin, Ruixuan Deng, Junran Wang Published: 2026-09-29 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.36585v1 Pretrained transformers use little of their depth to follow references in context. Thirteen base models reliably follow only 1.4-3.6 lines, and extra pretrained loops add little. A task-trained rank-8 LoRA at one early layer extends this computation with all model weights frozen. Qwen3-8B improves from 15.5% to 99% exact accuracy on 24-line chains; a longer-trained LoRA reaches 50 lines. Ouro-1.4B reaches 60 lines after four loops and at least 160 after eight. The LoRA starts a relay: program lines pass on their chain identity through a short range of middle layers. Frozen heads read progressively further up the chain, and removing parent-line attention stops the relay. A frozen-model measurement locates the last useful intervention layer within tolerance in three of four held-out models. Task-specific LoRAs also improve MuSiQue. Default answers therefore understate the computation accessible through a tiny edit. Code and an interactive demo are available at https://lunamos.github.io/stop-thinking-too-early/ --- ### Making Analog Training Scale: Co-Designing Mapping, Optimizer, and Converters Authors: Zhaoxian Wu, Tayfun Gokmen, Omobayode Fagbohungbe, T. Patrick Xiao, Tianyi Chen Published: 2026-09-29 Categories: cs.LG, cs.AR arXiv: https://arxiv.org/abs/2609.36584v1 Analog in-memory computing (AIMC) offers an alternative for model training by executing matrix operations directly where weights are stored. However, scaling AIMC to train modern deep models remains an open challenge due to severe hardware non-idealities, including physical weights with finite dynamic range and write granularity, analog-digital converters with finite resolution, and noisy and asymmetric updates. Guided by the insight that gradient accumulation is sensitive to precision and rounding errors, we adopt a mixed-precision training paradigm: executing forward and backward matrix multiplications in the analog domain while computing weight gradients in the digital domain. To enable scalable training, we present a holistic system-algorithm co-design that co-optimizes weight mapping to ensure well-conditioned physical and logical weight profiles, couples a preconditioned optimizer with threshold-triggered open-loop pulsing to stabilize training trajectories, and aligns converter dynamic ranges to suppress quantization errors. Evaluated via hardware-calibrated architectural simulations calibrated with electrochemical RAM measurements, our framework scales Transformer training up to $123\text{M}$ parameters with validation loss scaling as $L\propto N^{-0.231}$, where $N$ is the parameter count, comparable to $L\propto N^{-0.238}$ for digital training. --- ### MemEvo: Automatic Discovery of Streaming Video Memory Mechanisms Authors: Guohong Liu, Jialei Ye, Shanhui Zhao, Yunxin Liu, Yuanchun Li Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36581v1 Query-agnostic streaming video understanding requires vision-language models to continuously compress an indefinitely growing visual stream into a bounded memory before future queries are known. The performance depends critically on the memory mechanism--what observations to preserve, how to represent and consolidate them, and what information to retrieve when a query eventually arrives. Rather than designing a single memory architecture by hand, we formulate memory design as a search problem over executable memory programs. We introduce a lightweight domain-specific language that expresses memory mechanisms through structured primitives for representation, admission, retention, consolidation, budgeting, and retrieval, while enforcing causal and bounded-memory constraints. Although structured, the derived program space remains large and contains heterogeneous, conditionally dependent design choices whose effects can only be assessed via downstream execution. We therefore propose MemEvo, an LLM-driven auto-research framework that uses pretrained LLM as a semantics-aware proposal model to iteratively generate and refine candidate memory programs based on accumulated experimental feedback. At runtime, a deterministic evaluation pipeline validates and evaluates each candidate, while the underlying vision-language model remains frozen throughout discovery. We finally produce a training-free, bounded-memory mechanism. Extensive experiments on StreamingBench and OVO-Bench demonstrate strong streaming video understanding performance together with substantial context and inference efficiency. --- ### SafeCoEvo: Co-Evolving Safety Harnesses and Guards for LLM Agents at Test-Time Authors: Yu Cheng, Yongkang Hu, Shuaijie Ma, Zhihang Lin, Weicheng Meng, Jingyang Qiao, Jiuan Zhou, Yushuo Zhang, Yihang Chen, Weilin Luo, Kun Shao, Dong Li et al. Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36580v1 LLM agents deployed in real-world environments continually encounter new tasks and safety risks, while execution feedback typically becomes available only after each task is completed. However, existing self-evolving approaches commonly rely on multiple rounds of optimization over fixed and repeatedly accessible task distributions, fundamentally differing from test-time adaptation in real-world deployment, where only experience accumulated from past tasks can be used to improve safety decisions on future unseen tasks. To address this limitation, we propose SafeCoEvo, a test-time Harness-Guard co-evolution framework for LLM agent safety that enables the external safety system to continually adapt from accumulated runtime experience. SafeCoEvo jointly improves two complementary safety capabilities at different timescales: S-Harness rapidly externalizes recent runtime experience into updatable explicit safety knowledge that can promptly influence subsequent tasks, while GuardVPO internalizes accumulated runtime safety experience over a longer timescale into parametric risk-judgment capabilities. By combining short-term rapid adaptation with long-term capability consolidation, SafeCoEvo continually improves the agent's safety capabilities, reducing the unsafe outcome rate by 10.05% while improving the task success rate by 12.15% over the strongest baseline, thereby achieving simultaneous gains in safety and task utility. --- ### Factorized Scheduling Principle: Learning Interpretable and Transferable Policies via Structured Additive Functions Authors: Hong Je-Gal, Hyun-Suk Lee Published: 2026-09-29 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.36578v1 Scheduling problems arise from repeatedly selecting one item from a set of candidates based on their states. These problems often reduce to assigning priority scores and choosing the highest-ranked item. In this work, we propose a factorized scheduling principle (FSP) framework to learn interpretable and transferable scheduling rules. The FSP framework represents system states as condition distributions and decomposes a global scheduling principle into additive univariate and pairwise components with identifiability constraints. The scheduling principle enables the framework to maintain a simple priority-based structure during deployment. This principle is learned by using a policy-based objective combined with a temporal-difference signal defined on the condition distribution. Experiments on synthetic and realistic scheduling tasks demonstrate the FSP framework's strong performance, interpretability, and zero-shot generalization across different system scales. --- ### Long-Term Memory-Guided Enhancement for Target Perception in Audio-Language Models Authors: Zhenhong Zhou, Xuanyue Zhao, Youji Liu, Yuanhe Zhang, Xiaoyu Ma, Lianyu Hu, Yang Liu Published: 2026-09-29 Categories: cs.SD, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.36577v1 Audio large language models (ALLMs) can reason about the content of audio recordings to perform complex tasks. However, these capabilities usually collapse in real-world environments when background noise and competing sources mix the target sound. Inspired by long-term memory in human listening, we propose Long-Term Memory-Guided Audio Enhancement (LTM-AE) to improve selective target perception by refining the audio representations of ALLMs without training. LTM-AE extracts representations in hidden states from separate clean reference recordings as long-term memory for each category, guiding enhancement toward a user-specified listening target. We reconstruct incoming audio tokens in the selected category long-term memory and interpolate the reconstructions with the original tokens before language backbone decoding. This interpolation controls the influence of stored auditory experience while keeping all ALLM parameters fixed. Diagnostic readouts across twenty sound categories and three ALLMs show that LTM-AE strengthens responses to a specified target amid three interfering sources. Averaged over constrained and free-form classification, accuracy gains over raw mixtures range from 29.53 to 46.15 percentage points across multiple open source models. For speech content recovery, LTM-AE with an additional learned token-level gate reduces Qwen2-Audio's word error rate from 23.07% to 14.77%. This work takes an initial step toward using principles of human long-term memory to enhance ALLMs for real-world listening. Our code is available at https://github.com/aynlp/ltm-audio-code --- ### Divide and Inject: Can Agents Reconstruct an Indirect Prompt Injection from Fragments? Authors: Michael Lee, Zhipeng Wei, Yue Dong, N. Benjamin Erichson Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36576v1 Agentic systems are now being widely used to orchestrate tools and reason over long contexts. However, the improving capabilities of the large language models powering these agents also create new attack surfaces for indirect prompt injection. In particular, an attacker may not need to place a complete malicious instruction in retrieved content if the agent can reconstruct the objective from incomplete fragments distributed across a long context. In this work, we introduce adaptive long-context prompt injection (AdaLCPI), which combines long-context fragmentation with adaptive search. AdaLCPI splits an attack objective into incomplete fragments, embeds them in external content retrieved through the agent's tools, and uses a reconstruction cue to prompt the agent to combine them. It then iteratively refines the fragments and cue with OpenEvolve using graded scoring and natural-language execution feedback from the target agent. Empirically, AdaLCPI achieves higher attack success than strong adaptive baselines, reaching 61.4\% macro-average ASR compared with 32.8\% for Trojan Hippo-style and 30.0\% for AgentVigil. Safety evaluations should therefore test whether agents remain robust when harmful objectives must be reconstructed from incomplete fragments. --- ### Visual sensitivity is not claim retractability: persistence-aware credit assignment for multimodal reinforcement learning Authors: Zhongan Bi, Kepeng Lin, Xuanang Gao, Yuhan Sun, Lianrun Zhang Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36572v1 Reinforcement Learning with Verifiable Rewards (RLVR) has been extended to Large Vision-Language Models (LVLMs), and perception-aware methods further encourage policies to rely on visual evidence. Yet relying on the image does not guarantee that visual claims are supported by it. Before RL training, 27.81% of the correctly answered responses of Qwen2.5-VL-7B on four multimodal reasoning benchmarks contain at least one direct visual claim that the image does not support. Since outcome-level RL rewards each response as a whole, these claims inherit the positive credit of the correct answer. We introduce a fixed-rollout counterfactual diagnostic that re-scores the same response under an intervened image to separate Evidence-Function Sensitivity (EFS), how strongly the model's predictions change, from claim persistence, whether the model keeps supporting the same claim rather than retracting it. The diagnostic reveals Sensitivity-Persistence Decoupling (SPD): under DAPO and VPPO, EFS increases and claims become more retractable overall, yet unsupported claims become significantly more persistent, whereas GRPO raises EFS without this deterioration. We therefore propose Persistence-Aware Credit Gating (PACG), which attenuates positive credit for unusually persistent visual claims and leaves all other credit unchanged. It requires no supported/unsupported labels and adds no inference cost. On Qwen2.5-VL-7B, PACG raises the nine-benchmark average over three seeds from 58.1% to 59.9% with DAPO and from 59.8% to 60.9% with VPPO, while making unsupported claims more retractable. The gains extend to a larger model, a newer backbone, and the accuracy of HallusionBench also improves consistently. These results suggest that visual sensitivity and claim retractability are complementary dimensions of multimodal credit assignment. --- ### CounterSteer: Suppressing Indirect Prompt Injection with Activation Steering Authors: Mark Russinovich Published: 2026-09-29 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.36570v1 Indirect prompt injection makes an LLM agent treat untrusted retrieved text as instructions. We present CounterSteer, an inference-time defense that suppresses this behavior inside the model. Per model, a five-step recipe fits a residual-stream direction from paired episodes differing only in whether an embedded instruction is followed, and retains it only if it passes pre-specified causal and capability gates. At deployment, the direction is subtracted from every tool-result token during prefill. The edit is always on--there is no detection decision to evade--and requires no fine-tuning, auxiliary model, or added tokens, only white-box serving and tool-result span boundaries. Across five open-weights models (8B-106B, five vendor lineages), held-out attack success falls from 0.21-1.00 undefended to 0.00-0.17 defended, and AgentDojo compromise rate from 0.10-0.49 to 0.006-0.079, at 93-100% typography-normalized benign utility, with larger task-dependent costs when reasoning over steered content. A benchmark-level adaptive attacker reaching 0.67-0.73 undefended is held to roughly a quarter of that on the two most deeply evaluated models. Among the defenses we measured on capable models, those achieving lower compromise rates either lost 22-89% of benign utility or fine-tuned the served weights. White-box gradient attacks through the deployed vector compromise at most 2 of 52 episodes, and none of 2,052 replayed human red-team attacks succeeds. CounterSteer largely neutralizes instructional takeover: a black-box framing search cracks 3 of 18 development samples. Parameter manipulation--attacker-chosen arguments in otherwise legitimate calls--is only partially resisted (13 of 18); the decision becomes linearly readable at argument emission but not at the examined pre-generation sites, and is not removed by the tested prefill- or decode-time steering, motivating argument-provenance controls. --- ### From Checkpoint Variation to Selection Gains in Supervised Fine-Tuning Authors: Yupeng Chang, Wenxuan Zhang, Yuan Wu Published: 2026-09-29 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.36569v1 Checkpoint selection is a routine decision in supervised fine-tuning (SFT): training produces multiple checkpoints, but only one is retained. Yet fixed-budget comparisons do not by themselves distinguish three empirical claims: whether more validation data improve checkpoint selection, whether a selection rule outperforms validation-loss selection, and whether it improves over simply retaining the final checkpoint. We therefore treat checkpoint selection as a finite-information decision problem. Holding completed training trajectories, candidate checkpoints, and independent test items fixed, we vary the validation budget and separately measure improvement from additional validation data, gain over negative log-likelihood (NLL) selection, and gain over the final checkpoint. Across 60 mathematical SFT trajectories and 19 configurations, increasing the validation budget from 32 to 305-313 examples raises independent-test accuracy by 0.32 percentage points (pp) for generated-accuracy selection and 0.29 pp for checkpoint agreement, with 95% configuration-bootstrap CIs of (0.10, 0.56) and (0.11, 0.50), respectively. At the full validation budget, the two generation-based rules outperform matched NLL selection by 0.71 and 0.85 pp, respectively, while their gains over the final checkpoint remain unresolved. A cross-domain replication on 12 newly trained Commonsense trajectories shows the same qualitative separation: increasing the validation budget from 32 to 1,024 questions improves generated-accuracy and checkpoint-agreement selection by 0.87 and 0.27 pp, while gains over the final checkpoint again remain unresolved. Together, these results show that benefiting from more validation data, outperforming NLL selection, and outperforming the final checkpoint are distinct empirical claims that require separate evidence. --- ### Sharp Convergence and Sampling Trade-offs for Riemannian Diffusion under Nonnegative Ricci Curvature Authors: Yuhao Liu, Longbo Huang Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.36568v1 Diffusion models have emerged as state-of-the-art generative models, with recent extensions from Euclidean spaces to Riemannian manifolds. However, existing convergence guarantees for Riemannian diffusion models typically require $\tilde{O}(\mathrm{poly}(d,T)/ε^2)$ score evaluations, with potentially unfavorable dependence on the dimension. In this work, we develop a general framework that separates score discretization from Brownian-motion simulation and allows multiple geodesic random-walk steps per score evaluation. Under nonnegative Ricci curvature assumption and an exact Brownian-motion simulation oracle, we show that $\tilde{O}(d/ε^2)$ score evaluations suffice to achieve an $ε^2$ KL divergence from the target distribution, matching the existing convergence rate of Euclidean diffusion models. We further show that $\tilde{O}(d^4T/ε^2)$ geodesic random-walk steps suffice to approximate the required drifted Brownian motion to $ε$ total variation error. Combining these results yields a sampling scheme with $\tilde{O}(d/ε^2)$ score evaluations and $\tilde{O}(d^4T/ε^2)$ geodesic random-walk steps, motivating multiple random-walk steps between consecutive score evaluations. Our results provide a sharper characterization of the convergence and sampling complexity of Riemannian diffusion models. --- ### AffectReveal: Event-Grounded Emotion Recognition Beyond Visual Appearances Authors: Yihao Qian, Runhao Zeng, Sicheng Zhao, Feng Liang, Hongmin Cai, Mingkui Tan Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.36563v1 Visual emotion recognition commonly assumes that all evidence required for prediction is contained in the observed image or video. Yet the same visible reaction can convey different emotions depending on events beyond the input: tears, for example, may indicate grief or joy. We formulate Event-Grounded Emotion Recognition (EGER), where emotion recognition requires recovering the affect-determining event. We construct EGER-Bench, comprising 10,052 videos and 10,734 images across 11 emotions, two source domains, and four visual settings. A study with six annotators shows that event context raises human recognition accuracy from 33.96% to 72.08%, confirming that visual evidence alone is often insufficient. Semantic relevance alone does not solve EGER: a plausible event may imply the wrong emotion if its identity, focal-person role, relationship, or outcome is misinterpreted. We therefore propose AffectReveal, a tuning-free framework that first constructs and independently verifies evidence-grounded alternatives over these affect-critical factors. It then cross-checks the recovered event against face-masked in-media facts through bidirectional atomic evidence support, while retaining the original unmasked input for final prediction. Across three downstream models and four input settings, AffectReveal yields average UAR gains of 5.26--10.53 points. For three fine-tunable models, it also enables untuned models to outperform their fine-tuned visual-only counterparts in all 12 accuracy comparisons, without updating downstream parameters. --- ### ThinkingGuard: Decoding Implicit Hazards via Step-by-Step Risk Attribution in Multimodal Large Language Models Authors: Ruochen Zhang, Yao Huang, Yitong Sun, Jiahe Xie, Jin Yan, Jifan Ma, Yuanfang Guo, Xingxing Wei Published: 2026-09-29 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.36562v1 While Multimodal Large Language Models (MLLMs) are increasingly deployed in safety-critical domains, their reliability is threatened by multimodal implicit risks. Unlike explicit threats, these hazards emerge when individually benign text and neutral visual entities logically converge to induce unsafe outputs. Current detection methods fail to address this because they overlook the underlying risk activation mechanisms that govern cross-modal risk activation, leading to single-modality shortcut learning and hallucinated rationalizations. To bridge this gap, we first construct TriggerBench, the first dataset explicitly modeling risk compositionality (5,600 instances). By formally isolating Key Elements and Trigger Elements to build counterfactual contrastive pairs, TriggerBench eliminates risk residues and forces models to perform genuine logical deduction rather than superficial pattern matching, which provides a rigorous foundation for both large-scale training and fine-grained evaluation. Building on this, we propose a Step-Supervised Structured Reasoning training framework and employ it to train ThinkingGuard, a specialized guard model. Inspired by Situation Awareness theory, we decouple implicit risk identification into progressive cognitive stages, and utilize a step-reward Monte Carlo Tree Search algorithm to explore optimal reasoning trajectories, which are then distilled into the model through Dual-Constraint Preference Alignment. Extensive experiments across both standard and implicit safety benchmarks demonstrate that ThinkingGuard achieves strong performance. Project resources are available at https://github.com/FroggyChen/ThinkingGuard. --- ### FM-ReID: Selective Competitive Token Routing for Object Re-Identification Authors: Zhiqi Li, Xiaowei Zhou, Zeyuan Sun, Feng Gao, Junyu Dong Published: 2026-09-29 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.36560v1 Object re-identification (ReID) faces a recurring challenge: different identities can share highly similar global appearances, while the cues that distinguish them are localized, heterogeneous, and visible only under particular viewpoints. This challenge arises in animal ReID through markings, contours, and scars, in person ReID through subtle clothing and accessory cues, and in vehicle ReID through localized appearance details. Although visual foundation models encode such information in dense tokens, a single holistic descriptor can obscure discriminative local signals. We propose FM-ReID, an end-to-end framework that formulates local representation learning as selective competitive token routing. Its Competitive Fine-grained Mining module uses multiple mining queries and a residual query to compete for dense DINOv3 tokens. Above-prior selection retains tokens preferentially allocated to each mining query, while the residual slot receives tokens excluded from the retrieval descriptors. The resulting multi-query descriptors are jointly trained with a holistic representation for retrieval, without fixed spatial partitions or equal-area constraints. FM-ReID achieves strong results on animal, person, and vehicle ReID benchmarks, supporting competitive token routing as an effective way to augment holistic foundation-model representations. --- ### HiTS-CL: A Continual Learning Framework for Long-Horizon Temporal Knowledge Graph Extrapolation Authors: Yansong Liu, Rui Liu, Yuan Zuo, Hongwei Zhao, Da Fu, Fuwei Zhang, Fuzhen Zhuang, Yong Chen, Zhe Li Published: 2026-09-29 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.36559v1 Extrapolative temporal knowledge graph reasoning (TKGR) predicts future facts from historical snapshots. Most existing methods train once on an early prefix of the timeline and then use a frozen model for all future timestamps. We argue that this fixed-prefix protocol is misaligned with extrapolation. It learns from a static prefix, whereas the target stream is non-stationary: new entities and facts emerge, temporal dependencies shift across regimes, and recurring historical signals must be refreshed online. As a result, models trained only on early snapshots become outdated and degrade over long horizons. We address this mismatch by formulating extrapolative TKGR as continual learning over streaming snapshots. Under this view, effective extrapolation must jointly handle current dynamics, stable knowledge, and recurring historical evidence. Based on these requirements, we propose History-enhanced Two-Step Continual Learning (HiTS-CL), a backbone-agnostic continual learning framework for extrapolative TKGR. HiTS-CL tracks current dynamics via continual fine-tuning, preserves stable knowledge via multi-teacher adaptive distillation, and retains recurring historical evidence via a selective memory of recent and frequent facts. We integrate HiTS-CL into five representative TKGR backbones and evaluate it on four benchmark datasets. HiTS-CL consistently improves extrapolation accuracy, reduces long-horizon degradation, and outperforms strong continual-learning baselines, including a recent method for temporal knowledge graphs. Source code and data are available at https://github.com/liuyansong98/HiTS-CL. --- ### How Medical VLMs Underutilize Their Vision Encoders: A Dermatology Perspective Authors: Janet Wang, Yunbei Zhang, Xiao Wang, Jihun Hamm Published: 2026-09-29 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.36557v1 Medical Vision-Language Models (VLMs) show significant promise for clinical image understanding, offering accurate diagnosis with interpretable reasoning. However, a critical performance gap exists between their strong vision encoders and the full multimodal model: in dermatology, the MedSigLIP encoder outperforms MedGemma by an average of 10.26 percentage points even when both use zero target-task labels; few-shot linear probing provides further evidence of strong visual representations. This gap motivates an investigation of how visual information is used in end-to-end diagnosis and why plausible-sounding predictions can lack grounding in image evidence. Using dermatology as our primary testbed, we systematically investigate three hypotheses for this phenomenon. We further provide a mechanistic analysis of the model's internal attention patterns, showing that a simple describe-then-decide prompting strategy increases vision attention by 30-40% during generation. Task-specific fine-tuning improves dermatology classification but reduces cross-domain medical question-answering performance in our evaluation. To address these challenges, we combine label-free prompting with low-label encoder-assisted reranking while keeping the VLM frozen. We validate the interventions across five VLM backbones in dermatology and provide supporting representation and attention analyses across additional medical modalities. --- ### MAADBench: The Refreshable Paradigm for Anomaly Detection in Multi-Agent Systems Authors: Lei Ma, Dennis Hofmann, Haowen Xu, Joshua DeOliveira, Peter VanNostrand, Lei Cao, Elke Rundensteiner Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.36556v1 Recent studies report that LLM-based multi-agent systems (MAS) fail at rates of 41%-87%, yet to our knowledge, no benchmark to date supports systematic anomaly detection (AD) for them. Building MAS AD benchmarks is hard because they must remain fresh as LLM systems evolve: tasks may leak into training data and thus be memorized by LLMs, traces and anomaly patterns expire as backbones evolve, and labels must be provided reliably for each refresh. To address these challenges, we present MAADBench (MA: multi-agent; AD: anomaly detection), the first refreshable MAS AD benchmark designed for diverse, evolving LLM backbones underlying the agents. MAADBench combines (1) sampled-and-coupled generative tasks over an approximately 10^37-task space to mitigate task leakage, (2) refreshable trace generation under configurable LLM backbones, and (3) automated provision of cost-free, deterministic step-level labels for fine-grained AD evaluation. Beyond offering the paradigm itself, we run MAADBench with five state-of-the-art LLM backbones and release the MAADBench-Full dataset with 5,200 step-labeled traces. Benchmarking 25 AD methods on the MAADBench dataset reveals substantial limitations in current approaches: they rely heavily on supervision, struggle with subtle MAS-specific anomalies, and lack robustness across LLM backbones. These gaps point to a rich research agenda for MAS-specific anomaly detection, with MAADBench providing a systematic and refreshable testbed for method development and evaluation. We open-source MAADBench-Full at https://huggingface.co/datasets/hww123/MAADBench-full. --- ### SERA: Scale-Equalized Rollout Allocation for Maximum Likelihood Reinforcement Learning Authors: Zihao Chen, Fanxiang Xiong, Hongran Ren, Xuefeng Bai, Zhongxiang Dai, Kehai Chen, Zhiguo Zhang, Zhiyong Wang, Yu Cheng Published: 2026-09-29 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.36552v1 Maximum Likelihood Reinforcement Learning (MaxRL) targets prompt-wise log-success and has shown strong performance on reasoning tasks. Under finite rollout budgets, however, the estimator used by MaxRL attenuates each prompt's likelihood gradient by a factor that depends on its success probability and rollout count. Under uniform rollout allocation, the common rollout count fails to compensate for success-dependent attenuation, leaving low-success prompts more strongly attenuated and distorting their relative contributions to the expected aggregate gradient. We introduce SERA (Scale-Equalized Rollout Allocation), which redistributes a fixed rollout budget to approximately equalize these finite-rollout scaling factors. Building on our theoretical analysis of how finite rollouts distort prompt-wise likelihood gradients, we formulate the allocation as a fixed-budget max--min problem, derive a waterline solution to its continuous relaxation, and introduce a multiplicity correction to remove the additional prompt weighting induced by heterogeneous rollout counts. Experiments show stronger alignment with exact likelihood gradients in a controlled ImageNet setting and improved multi-sample solution coverage over MaxRL on maze navigation and mathematical reasoning under matched training rollout budgets. --- ### Grounded Revision vs. Prior Injection: Probing Retrieval-Augmented Patent Claim Amendment Authors: Josepha Michiko Leo, Hyun-seok Min, Yehoon Jang, Irvan Zidny, Jin-Woo Chung, Sungchul Choi Published: 2026-09-29 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.36550v1 Retrieval-augmented generation is widely used in professional writing, yet whether retrieval grounds revision or merely injects templates is rarely tested where "correct" has a definable meaning. Patent claim amendment supplies that signal: the examiner names the attacked limitation and cites prior art, providing per-case ground truth. We release three artifacts: (i) a corpus of 7,385 USPTO prosecution cases with XML-aligned pre/post claims, rejection, and cited prior art; (ii) a seven-probe battery comparing random and structural-match retrieval as two policies under a fixed prompt scaffold; (iii) a deterministic five-channel metric (C1-C3 and C5 in main, C4 supplementary) requiring no LLM evaluation. Across 9,600 pre-registered calls on four frontier LLMs (Claude Sonnet 4, Claude Haiku 4.5, GPT-5.4, GPT-4o-mini), no tested model exhibits detectable classical prior-injection behavior; retrieval effects are small and direction-inconsistent between random and structural retrieval, and the null is unchanged under a dense (semantic) retriever, across retrieval depths k in {1,3,5,10}, and under a paraphrase-sensitive grounding metric. Revision locality reveals a model-specific difference that the template channel misses. The four-cell taxonomy, which we treat as exploratory, leaves the prior-injector cell unoccupied. --- ### Interactive-Policy Distillation with Bidirectional Propose-and-Verify Authors: Shutong Wu, Xiwen Chen, Brendan Rappazzo, Daiheng Zhang, Anderson Schneider, Yuriy Nevmyvaka, Jiawei Zhang Published: 2026-09-29 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.36546v1 On-policy distillation (OPD) trains a student model on its self-generated trajectories with dense token-level teacher feedback. However, naive OPD may suffer from teacher unanchoring, where the student's reasoning trajectory drifts far from the teacher, causing the teacher to be queried on states it would hardly visit and thus provide unreliable supervision. We propose Interactive-Policy Distillation (IPD), which applies adaptive teacher intervention to the student rollout. Under a bidirectional propose-and-verify state machine, the student and teacher alternately exchange their roles as proposer and verifier, and collaboratively generate mixed-source trajectories. Then different supervisions are applied according to the source of each token. This bidirectional propose-and-verify mechanism and the source-split loss make IPD not only a more performant distillation method, but also a unified bridge between on-policy and off-policy paradigms. To make the interleaved dual-model rollouts more efficient, we also design a dedicated fused inference engine that co-hosts both models in one serving instance with separate KV caches and instantiates the state machine model to distribute, collect, and process requests. On math reasoning tasks and across multiple teacher-student model pairs, student models trained with IPD not only outperform those trained with OPD, but also demonstrate higher data efficiency. Specifically, when distilling Qwen3-30B-A3B into Qwen3-1.7B-Base, IPD brings a +3.28 mean@8 and a +3.28 best@8 benchmark-averaged accuracy improvement compared with OPD. Besides, IPD only consumes about 1/4 of the training examples and steps to outperform OPD trained on the whole training dataset for one epoch. We also investigate the impact of different loss variants and takeover / handback configurations, and demonstrate the robustness of IPD on different training data. ---