# Research Papers — 2026-09 · bundle page 54 of 266 > Full-text markdown of papers #2651–2700 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/54 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/53.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/55.md) ### Prospective Interpretation Risk: Principled Communication Control Between LLMs Authors: Wanrong Yang, Rehan Deen, Julian Ma, Yuheng Fan, Yaoyu Jin, Taher Jafferjee, Ziquan Liu, Dominik Wojtczak, Yalin Zheng, David Henry Mguni Published: 2026-09-27 Categories: cs.MA, cs.LG arXiv: https://arxiv.org/abs/2609.33885v1 Large language model (LLM) agentic systems increasingly rely on models communicating with one another, yet existing uncertainty and multi-agent methods rarely estimate how a particular receiver will interpret a message before it is sent. This matters in heterogeneous systems, where capable receivers can reconstruct different tasks from the same message. We model this as a sender-receiver problem with a latent receiver type and define prospective interpretation risk (PIR): the probability that a receiver reconstructs a task other than intended. Rather than model an LLM's full input-output behaviour, we use black-box probes relating messages, intended tasks, and receiver-specific reconstructions, yielding scalable supervision while separating interpretation from downstream capability failure. Offline, heterogeneous frozen receivers provide supervision for receiver-conditioned risk and the effects of predefined mutable message features. At deployment, history induces a posterior over receiver types, guiding message revision and selection. We introduce value of interpretation information (VoII), querying for receiver information only when its expected communication benefit exceeds its cost. Our theory characterises when receiver information has decision value and bounds such queries. Empirically, interpretation-failure rates vary by 4-13x across receivers. Receiver information reduces PIR calibration error by 68% relative to a receiver-agnostic predictor, largely by correcting receiver-specific risk levels. PIR-guided revision reduces interpretation failure by 44% relative to the original message and 40% relative to a generic rewrite, mostly through a repair that helps every receiver. VoII outperforms information-gain and random querying at matched cost on the interpretation objective it optimises, lowering interpretation failure from 3.84% to 3.79% while querying 18.2% of episodes. --- ### Lost with a Map: Conversational State and Behavioral Reliability in Language Models Authors: Atahan Dokme, Larry Heck Published: 2026-09-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.33883v1 Task-oriented dialogue requires maintaining and updating information across turns, yet language models expose no explicit belief-state object. We study how conversational state is represented, updated, and used inside eight instruction-tuned language models from four families on MultiWOZ and SGD. Structure and values separate: which domains, slots, and requests are active is linearly readable just before the model acts, whereas exact values are far more readable where the user stated them. After a user changes a value, both values remain accessible at their mentions, and causal interventions show that both continue to influence the model's action. In natural closed-loop interaction, query failures separate into cases of weak structural support, incorrect value resolution, and failure to deploy otherwise-supported constraints, with targeted interventions producing systematically different repair behavior across these cases. These findings motivate a state-action controller that starts from the base model action and selectively edits it using structural readouts, without requiring a complete predicted belief state as an intermediate representation. On held-out MultiWOZ interaction across five models, it raises the base model exact-query accuracy from .318 to .621 and task success from .272 to .371 at negligible added cost. Overall, reliable interaction requires not only retaining conversational information, but resolving which available constraints currently apply and ensuring that they govern action. --- ### Neuron-Level Architecture Growth: A Controlled Evaluation for EEG Time-Series Decoding Authors: Adam Mounir, Stella Douka, Arnault H. Caillet, Bruno Aristimunha, Sylvain Chevallier Published: 2026-09-27 Categories: eess.SP, cs.HC, cs.LG, cs.NE arXiv: https://arxiv.org/abs/2609.33880v1 Convolutional EEG decoders are trained at a fixed width, usually set by their authors on other data. Growing methods add neurons during training where the loss could decrease the most, but whether they improve compared to a reference width is untested on EEG. Here, we grow three convolutional backbones on 12 motor-imagery datasets under three protocols and compare each with its reference model per subject. The growing ShallowFBCSPNet scores 2.9 points above its reference model with only half the parameters (0.57x), SCCNet changes by at most 1.2 points. Deep4Net growing models show decreased accuracy, but they require adaptation that prevent to compare faithfully the results. These differences follow the selection step, which keeps a candidate neuron relying on a dynamic threshold from singular values decomposition. Overall, these results suggest that growth helps when its criterion can rank the candidate neurons, and that the rate of skipped neuron addition tells where a decoder can be grown small from scratch. --- ### DCEmbed: Scalable Optimization over Neural Surrogates Authors: Akshay Sreekumar, Nicolas Christianson, Priya L. Donti, Ellen Vitercik, Ram Rajagopal Published: 2026-09-27 Categories: math.OC, cs.LG arXiv: https://arxiv.org/abs/2609.33879v1 Neural surrogates can accelerate large-scale optimization by replacing expensive or intractable model components with efficient learned approximations, but solving the resulting embedded problems can remain prohibitively costly. For instance, standard exact encodings of neural networks with ReLU activations allow the problem to be solved by mixed-integer solvers, but add large numbers of binary variables to accommodate the nonlinearity of the activations, which can render the problem computationally prohibitive. To address this, we propose DCEmbed, a heuristic for optimization problems with embedded neural surrogates that leverages the difference-of-convex (DC) representation of the network and avoids adding activation binaries. Exploiting shared structure within the DC representation of a ReLU network, we derive a reduced-size, exact formulation for its convex components that can be embedded in optimization problems using just two linear inequalities and one continuous auxiliary variable per hidden neuron. Using this, the problem is solved via an iterative penalty convex-concave procedure, where only the concave portions of the neural terms are approximated at each stage. The original objective, constraints, and any discrete decisions are retained, allowing standard convex or mixed-integer optimization solvers to optimize the host and surrogate jointly at each iteration. In experiments on quadratic programs, mixed-integer resource allocation, and neural two-stage stochastic programming, our method demonstrates much faster progress toward high-quality feasible solutions than approaches using exact mixed-integer embeddings. In particular, DCEmbed achieves $4\times$ lower normalized primal integral than the best exact baseline on the resource allocation problem, while in two-stage stochastic programming it reaches the global surrogate optimum $\sim 5\times$ faster than Gurobi ML. --- ### Curating Merchant-Matching Training Data with Two Confidence-Gated Local LLM Judges Authors: Donghao Huang, Jinling Pei, Zhaoxia Wang Published: 2026-09-27 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.33878v1 Merchant matching resolves a noisy payment descriptor to a retrieved merchant entity or returns no match. A key challenge in curating training labels is distinguishing teacher abstention from evidence that no acceptable entity exists: false no-match labels contaminate pseudo-labeled data, while conservative labeling reduces coverage. We investigate whether agreement between two local large language model judges improves pseudo-label reliability. A label is retained only when the judges agree, with separate ordered thresholds for selections and abstentions that guarantee disjoint positive and negative label sets. Retrospective replay on 2,000 expert-annotated queries shows that higher selection thresholds can improve positive-label purity, whereas higher abstention thresholds increase false no-match labels. At thresholds (0.86, 0.80), Muse Glimmer 30B and Gemma 4 31B jointly label 1,633 queries (81.7% coverage) at 96.88% purity; positive and negative purities are 99.47% and 93.38%. This exceeds either constituent model at the same thresholds by more than two percentage points, with lower coverage. A split-half check finds only 0.14 percentage points of threshold-selection optimism. A symmetric threshold of 0.86 adds 40 erroneous no-match labels, while 46 false abstentions persist even with no confidence threshold. Across five matched within-model comparisons, higher reasoning effort yields no clear F0.5 gain and increases median latency by 1.8-5.0 times. These results motivate separate thresholding and auditing for positive and negative pseudo-labels. The study establishes label purity, not student utility; fresh-data curation and student fine-tuning remain necessary to demonstrate downstream value. --- ### Counterfactual Rollout Replay: Forkable Environments as Free Process Rewards for Software Engineering Agents Authors: Yuanhao Li, Hongbo Wang, Xuhong Chen, Yiming Cao, Xunzhu Tang Published: 2026-09-27 Categories: cs.SE, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.33875v2 Outcome-only reinforcement learning gives software engineering (SWE) agents a terminal success signal but little direct guidance about intermediate decisions. We introduce Counterfactual Rollout Replay (CRR), a training-time procedure that uses forkable executable environments to obtain step-level return contrasts. CRR selects a small set of decision points, restores each state, samples an alternative action, and rolls the branch forward under the policy. It retains the realised training trajectory and replaces the advantage at selected steps with the difference between its terminal return and the sampled counterfactual return. The method needs no human process labels or learned process reward model; free refers to those supervision costs, not replay compute. With a 14B policy, CRR improves pass@1 on SWE-bench Verified, SWE-bench Live, and SWE-rebench, and combines with process-reward and trajectory-search methods. On SWE-bench Verified, an equal-wall-clock comparison on the same hardware yields 41.7% versus 36.7% for extended outcome-only GRPO, a 5.0-point gain with fork overhead included. These results apply to environments with affordable, reliable state restoration; stochastic continuations and expensive or imperfect replay remain limitations. --- ### JET: Justification Evaluation in Transformer Authors: Shenghao Ding Published: 2026-09-27 Categories: cs.LG, cs.PF arXiv: https://arxiv.org/abs/2609.33874v1 JET uses pretrained language and vision-language models to select among a finite set of answers without additional training. It evaluates candidate likelihoods directly and shares computation across candidates. Experiments on desktop CPUs and consumer GPUs assess decision accuracy and execution cost. Qwen3.6-35B-A3B achieves 87.48% accuracy on the full MMLU test set and 3.69 requests per second on a separately timed MMLU subset. The accuracy-throughput comparison covers model, hardware, and reasoning choices, with Jev as an external reference. Controlled execution experiments show 2.18-2.23-fold speedups from prefix reuse and cache management, and a 30.8% reduction in process time from input preparation optimizations, with unchanged outputs. Optional reasoning has a task-dependent accuracy-throughput trade-off. These results support local decision inference from existing models. --- ### Robot-GST: geometry-aware spatial-temporal robot policy representation and evaluation Authors: Sichao Liu, Zekun Wang, Lixuan Tang, Yiming Li, Xiaohan Wang, Hanzhi Zhang, Daqiang Guo, Peng Zhou, Lihui Wang Published: 2026-09-27 Categories: cs.RO, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.33872v1 Robotic manipulation policies are advancing rapidly with increasing reliance on vision-language models for end-to-end decision making. However, reliable deployment remains challenging because many policies lack explicit mechanisms for predicting task outcomes and evaluating whether generated actions will achieve desired final states, causing execution errors to accumulate during long-horizon manipulation. We present Robot-GST, a geometry-aware spatio-temporal behaviour representation and evaluation framework that constructs a Gaussian-SAM robotic environment for real-to-sim policy verification and improves the reliability of real-world manipulation deployment. Our approach constructs a high-fidelity robotic environment from RGB-D observations using 3D Gaussian Splatting and SAM3D, enabling ``simulation and evaluation before acting''. It integrates visual observations and language instructions with spatio-temporal reasoning for long-horizon task planning using large vision-language models. To bridge high-level planning and real-world execution, we introduce Gaussian-aware final-state estimation through geometric sampling and state-based trajectory planning. Before execution, candidate action sequences are simulated and evaluated in the Gaussian-SAM environment to filter infeasible behaviours. We validate our approach on representative manipulation tasks involving rigid, soft, and deformable objects, including cube placing, toy packing, and duck rearrangement, demonstrating that geometry-aware spatio-temporal reasoning and state-aware execution improve manipulation reliability across different object categories. Our results suggest that combining geometry-aware reconstruction with high-quality rendering and simulation provides a scalable approach for evaluating robotic manipulation behaviours. Website: https://robot-gst.github.io --- ### Population Physics, Population Problems: Safety and Emergence in LLM Societies Authors: Adrian de Wynter Published: 2026-09-27 Categories: cs.MA, cs.CL arXiv: https://arxiv.org/abs/2609.33871v1 The collective behaviour of large language model (LLM) societies is not the sum of their individual outputs. It yields statistically distinct, sometimes-unpredictable phenomena, for which the tools we use to study single agents may not scale. Due to recent incidents involving autonomous agentic systems, however, understanding these systems is paramount. For that we introduce a framework for measuring self-organisation in LLM social systems and apply it to three such systems: a Schelling grid, a social network (Moltbook), and a Twitter-like misinformation simulation ('Rogue'). All three exhibit statistically significant self-organisation. Moreover, their relaxation dynamics vary with the environmental information available to the agents, with open-ended systems (Moltbook, Rogue) exhibiting sharp, phase-transition-like dynamics. Further results show that population-level pathologies can emerge even when the LLMs are safety-tuned or monitored, being primarily driven by the coordinated activity of a population subset. We also show when self-organisation does \textit{not} emerge under two additional scenarios (a commons dilemma, GovSim, and a LLM-as-a-judge deliberation scheme, ChatEval). We argue that measuring signatures of this kind offers a lightweight, agent-agnostic diagnostic layer for detecting coordinated collective behaviour in deployed multi-agent systems without relying on natural language or model versioning. --- ### When Successful Strategies Fail: Adaptation to Environmental Novelty in Terminal Agents Authors: Janvijay Singh, Vaishnavi Shrivastava, Dilek Hakkani-Tur, Ece Kamar, Asli Celikyilmaz Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33870v1 LLM agents increasingly solve long-horizon tasks by autonomously interacting with their environment. In doing so, their strategies rely on assumptions about that environment: which resources and tools exist, where they are located, and how they behave. When these assumptions no longer hold, reliable agents must detect the change and adapt while pursuing the same goal. We study this adaptation capability through environmental novelty: a change that keeps the task objective fixed while invalidating an assumption underlying an otherwise successful trajectory. We introduce AGNI, an automated pipeline that extracts trajectory-relevant assumptions, injects targeted environmental changes, and validates that the resulting novel tasks remain solvable. Across three terminal benchmarks, AGNI produces diverse novelties spanning resources, interfaces, constraints, and execution semantics. Evaluating multiple LLM agents reveals a substantial adaptation gap between base and novel tasks. Trajectory analysis suggests that agents often encounter evidence of the change but fail to diagnose its cause and revise their strategy. Finally, post-training for environmental novelty improves adaptation to held-out novel tasks while also improving performance on base tasks. Our results highlight a gap between task competence and adaptive capability and motivate environmental variation as a core dimension of agent training and evaluation. --- ### Conformal Coverage of Time Series: Validity and Inference Authors: Percy S. Zhai, Maggie Cheng, Wei Biao Wu Published: 2026-09-27 Categories: math.ST, stat.ME, stat.ML arXiv: https://arxiv.org/abs/2609.33868v1 Conformal prediction provides marginal coverage guarantees, yet practitioners may wonder if the observed coverage is truly abnormal or consistent with sampling variation. Inference for realized coverage has received comparatively little attention, especially for time series. We study split conformal prediction with adjacent calibration and test sets of temporally dependent data. Using the functional dependence measure, we derive non-asymptotic bounds on marginal coverage error without mixing assumptions, which can be difficult to verify and may fail even for simple short-memory models. We establish a Bahadur representation to derive, to our knowledge, the first central limit theorem for realized coverage of split conformal prediction under temporal dependence. A consistent block-based estimator of the standard error yields an asymptotically justified test. We further study long-memory time series, which remain understudied in conformal prediction. For Gaussian linear processes, we show how very strong temporal dependence can lead to a non-Gaussian limiting law of realized coverage and establish block-sampling inference with an estimated normalization. The resulting theory explains how temporal dependence changes coverage uncertainty. --- ### R$^2$ Flow: Recursive Self-Improvement via Recursive Skill Evolution Authors: Mingda Zhang, Qiang Huang, Yanjin Li, Zijia Wang, Qika Lin, Xiaoying Tang, Tiesunlong Shen Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33867v1 LLM-based agents can improve themselves across tasks by reusing and revising the skills they orchestrate into executable procedures. Flow-based training fits this loop: it samples procedures in proportion to reward, and the flow through each skill credits it for the next library revision. Three obstacles stand in the way of making this self-improvement reliable: flow training suffers strategy collapse over tree-structured histories; nonnegative flow-based credit rewards frequent use as if it were benefit; and library edits rest on the task reward the policy optimizes. We introduce R$^2$ Flow, a recursive self-improvement framework that alternates policy learning, independent verification, and versioned skill-library updates on a shared-state orchestration graph. The graph merges histories that differ only in the order of independent steps, allowing flow training to pool evidence across equivalent executions. A flow-share readout of the trained flow, invariant to the backward policy, and a separate signed utility rank which skills to change, verifier evidence decides whether an edit is warranted, and a residual-variance plateau sets when to update. Committed edits reshape the graph the next policy learns on, realizing recursive skill evolution. Across question answering, mathematical reasoning, interactive decision making, and code generation, R$^2$ Flow improves task accuracy and library-edit precision over heuristic orchestration, reinforcement learning, and skill-evolution baselines, and transfers across executors. Code is available at https://github.com/beita6969/r2flow. --- ### Valid and Efficient Split Conformal Regression for Time Series Authors: Percy S. Zhai, Maggie Cheng, Wei Biao Wu Published: 2026-09-27 Categories: math.ST, stat.ML arXiv: https://arxiv.org/abs/2609.33866v1 We study conformalized quantile regression and conformalized median regression that fit a model on one block of a time series and calibrate the conformal interval on the adjacent block. The existing theory of conformal prediction for time series rests largely on mixing conditions, which are hard to verify from a time-series model and fail for many standard processes, including simple ones with short memory. We replace this theoretical toolbox with the functional dependence measure, which in principle accommodates long-memory observations. The accuracy of the conformal interval length for time series has been understudied. To the best of our knowledge, this paper is the first work that establishes non-asymptotic coverage guarantees and accuracy of interval length simultaneously for split conformal regression on time series. Furthermore, for Gaussian linear processes with long memory, where both the estimation of the center and its calibration converge slowly, we establish a sharper rate for the length error. We show that the calibrated length converges faster than the estimated center itself, and provide a matching lower bound for the usual centers when the calibration block is sufficiently large relative to the training block. To our knowledge, this is the first theoretical analysis of conformal interval length dedicated to long memory. --- ### In-Context Adaptation of Encoder-Decoder Models in Speech Recognition Authors: Yen Meng, Sharon Goldwater, Hao Tang Published: 2026-09-27 Categories: cs.CL, cs.SD, eess.AS arXiv: https://arxiv.org/abs/2609.33865v1 In-context learning offers an appealing approach to adapt automatic speech recognition (ASR) models to new speakers, accents, and domains by providing speech-text pairs as demonstrations at inference time. Recent work shows that some LLM-based speech models are capable of ASR in-context adaptation, when providing interleaved speech-text demonstrations. In this work, we ask whether in-context adaptation is an inherent ability for all encoder-decoder models. We study two forms of demonstration, collated and interleaved demonstration, across six encoder-decoder models, spanning conventional cross-attention-based and LLM-based architectures. We find that all tested models are able to perform in-context adaptation out of the box, achieving up to 30% relative improvement in the oracle experiments and up to 23% using first-pass hypotheses. Through controlled experiments on three English datasets, we show that lexical and speaker information both contribute to successful adaptation. While interleaved demonstration is effective in certain cases, collated demonstration brings consistent adaptation across the board. Our results suggest that in-context adaptation for ASR is not unique to specific architectures, training, or demonstration approaches. --- ### PI-NOMT: Physics-Informed Neural Optimal Mass Transport for Brain Fluid Dynamics Authors: Mehmet Emin Acar, Vahit Bugra Yesilkaynak, Helene Benveniste, Gozde Unal Published: 2026-09-27 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.33857v1 Recovering hidden transport mechanisms from sparse spatiotemporal observations is a fundamental inverse problem in scientific machine learning. In brain tracer imaging, dynamic contrast-enhanced MRI (DCE-MRI) provides time-resolved measurements of tracer concentration, while the underlying velocity and source mechanisms governing tracer propagation remain unobserved. We formulate this problem as physics-informed latent-state inference, in which the transport field itself is the primary object of inference rather than an auxiliary variable used only to reconstruct observed densities. We propose Physics-Informed Neural Optimal Mass Transport (PI-NOMT), a framework that represents density, velocity, and source as continuous neural fields and combines a continuous neural density teacher, recursive differentiable advection--diffusion--source rollout, unbalanced optimal-transport regularization, and governing-equation supervision. Physical laws act as structural priors that constrain the space of admissible transport mechanisms, while observed tracer dynamics provide evidence for estimating the latent transport state. We evaluate PI-NOMT on a synthetic benchmark with known ground-truth transport and on DCE-MRI sequences from nine control rats. On the synthetic benchmark, PI-NOMT accurately recovers the prescribed velocity field, including its magnitude, direction, and integrated trajectories, rather than merely reconstructing endpoint densities. Across the nine rat datasets, the framework yields sub-percent local endpoint error, consistent physical speed scales, and low post-training PDE and incompressibility residuals. These results support physics-informed latent-state inference as a general framework for recovering hidden transport mechanisms from observed dynamic scalar fields. --- ### Program-Verified Self-Evolution for Vision-Language Models Authors: Ahmed Heakl, Sungik Choi, Moontae Lee, Salman Khan Published: 2026-09-27 Categories: cs.CV, cs.AI, cs.CL, cs.LG, cs.NE arXiv: https://arxiv.org/abs/2609.33855v1 Self-evolving vision-language models train on questions they generate from unlabeled images. Since these questions have no gold answers, prior methods label them by majority vote over sampled answers or by a model judge. In a human evaluation, we find that 24\% of majority-vote labels and 18\% of model-judge labels produced during self-evolution are wrong. To address this problem, we present Verifiable QA Generation for Self-Evolving Models (VQS), which changes how the model judges answers. Instead of voting on an answer, the model parses each image into a structured record, such as a scene graph, a chart table, or a diagram graph. Fixed programs then write a question from the record and compute its answer. The model still acts as a visual checker, but it only confirms the individual facts the program reads, one short claim at a time. These claim-level checks select the parser's training targets, so the parser also improves without labels. Human raters find 94\% of VQS answers correct, against 76\% for majority voting. Across ten benchmarks, VQS improves Qwen3-VL by up to 3.18 points at the 2B, 4B, and 8B scales and outperforms the strongest self-evolving baseline at each. Gains keep growing over three training rounds, reaching 3.84 points at 2B. Code is released at https://github.com/ahmedheakl/VQS --- ### ReDrive: Shaping Representations with World Modeling for End-to-End Driving Authors: Yueting Zhu, Shaoyu Chen, Yuehao Song, Hui Sun, Qian Zhang, Wenyu Liu, Xinggang Wang Published: 2026-09-27 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.33854v1 Driving policies require capabilities of scene understanding and future evolution prediction. To achieve this goal, current end-to-end models typically construct complex perception-planning pipelines or introduce world models that explicitly predict future states, resulting in a complex system architecture. Inspired by the transferability of general-purpose visual representations, we argue that combining sufficiently strong visual representations with representation world modeling can support effective planning without relying on complex inference-time auxiliary modules. Based on this insight, we present ReDrive, an end-to-end driving framework that strengthens planning-oriented visual features via future representation prediction. To achieve this, ReDrive adopts a three-stage training pipeline consisting of driving video pretraining, joint world-modeling and planning training, and planner adaptation. This yields a strong planning-oriented representation and a high-performance planner, while requiring neither auxiliary perception modules nor future prediction at inference time. Experiments on NAVSIM demonstrate strong performance, achieving 91.0 PDMS on NAVSIM v1 and 90.8 EPDMS on NAVSIM v2. These results show that shaping representations with world modeling is sufficient to enable high-performance end-to-end planning while retaining a simple encoder-planner inference pipeline. --- ### Rethinking Contextualization by Reinterpreting Attention Head Channels Authors: Hakaze Cho, Haolin Yang, Zhun Sun, Naoya Inoue, Benjamin Heinzerling, Kentaro Inui Published: 2026-09-27 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2609.33851v1 Contextualization, the core operation of language modeling, transmits information across words to build sentence-specific word representations. Prior works mainly study contextualization, focusing on individual words and attention heads as a growing discrete dictionary, lacking a global view of their general behavior. Therefore, we propose a general principle: Globally, we find and estimate that different words carry different amounts of information, and less-informative words tend to absorb more contextual information. Specifically, these low-information words do not absorb contextual words uniformly, and finer-grained selectivity enables more precise routing to promote information transmission between matched words. Moreover, to find what mechanism causes such processing, we reinterpret attention heads as channels gated by their singular vectors and find that: (1) these singular vectors point to the hidden states of more informative words, allowing such words to write their information to others more strongly to act as information sources, and vice versa; and (2) these singular vectors can be viewed equally as hidden state features, enabling automated interpretation of attention heads beyond prior heuristic head discovery, also embedding heads into a continuous space rather than treating them as discrete, independent dictionary entries. --- ### QwenGyre: An Elastic Reinforcement Learning Framework for Training xLong-Horizon Agents Authors: Weiqi Wang, Yuxin Zhou, Mouxiang Chen, Siyuan Zhang, Yi Zhang, Yuyan Luo, Zhiyu Yin, Chencan Wu, Jiemin Jiang, Wentao Yao, Chujie Zheng, JianWei Zhang Published: 2026-09-27 Categories: cs.LG, cs.DC arXiv: https://arxiv.org/abs/2609.33848v1 Large language model (LLM) agents increasingly undertake extreme-long (xlong) horizon tasks, where a single execution can span hours, hundreds of model--environment interactions, and nearly 1M tokens per rollout. Applying online reinforcement learning (RL) to such executions poses two fundamental challenges: (1) severe execution variance and prolonged rollout delays cause massive GPU idling; and (2) complex non-linear branching generates massive trajectory redundancy, crippling training efficiency. To address these, we presents QwenGyre, an end-to-end framework for xlong-horizon online RL. QwenGyre elastically reallocates GPUs between rollout and training without interrupting live executions, while its trajectory processor reconstructs branching histories, scores partial progress, and deduplicates redundant paths to bound training costs. Scaled to our flagship model, Qwen~3.8 2.4T, with 700K tokens per rollout, QwenGyre yields a 6.0% absolute gain on NL2RepoBench (52.5% $\to$ 58.5%) in 48 steps. Across our evaluations on diverse domains of training datasets, QwenGyre delivers up to $1.85\times$ and $1.78\times$ speedups over Colocate and Async, respectively. --- ### How code helps different tasks? A decompositional lens on LLM post-training Authors: Zheng Yu, Yiwei Li, Yishen Chen, Xiang Li, Jiale Han, Benyou Wang, Jingbang Chen Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33845v1 Evaluating code data as a single corpus can obscure which types of code data benefit which models and downstream tasks. Effective data selection requires understanding both the benefits of individual categories and whether these benefits persist when categories are combined. We introduce a decompositional lens for studying these effects in LLM post-training. We first decompose an execution-verified code corpus into interpretable categories based on the computational patterns of its solutions. Through controlled fine-tuning experiments, we compare individual categories with a balanced mixture across instruction-tuned models on question answering, mathematics, and code generation. The resulting response maps reveal recurring gains in average question-answering performance, while the same category can improve one model or task and degrade another. The best-performing category also varies with the starting model and target task. We then compose compact mixtures guided by these results and examine whether benefits observed in individual categories persist under joint training. On selected model--task pairs, mixtures whose constituents each improve the target task outperform both their best constituent and full-corpus training while using roughly 10--15\% of the full corpus. These exploratory findings illustrate a \emph{less is more} pattern and highlight how the value of code data in post training depends on which categories are combined for which model and task. --- ### ViBR-WM: Visual Bayesian Regression for World Modeling Authors: Jifan Li, Ning Ning Published: 2026-09-27 Categories: stat.ME, cs.CV, cs.LG arXiv: https://arxiv.org/abs/2609.33844v1 Modeling temporal dependence and uncertainty is central to forecasting with world models. The Visual Bayesian Regression World Model combines visual features, physical histories and known covariates through interpretable regression, within a modular architecture supporting trend, seasonal and cycle dynamics. Visual compression reduces representation dimension, while Bayesian variable selection reduces active regression dimension. Posterior prediction combines forecasts across predictor subsets using their posterior probabilities as weights and accounts for parameter uncertainty and future disturbances. The model forecasts joint visual--physical states recursively and physical targets directly. Across four forecasting tasks spanning object motion, vegetation greenness and solar power, ViBR-WM achieves lower mean overall physical-target error than Temporal Straightening, ConvLSTM, PredRNN and SimVP on every task. Repeated fitting and resampling support these overall gains. --- ### Laya as a Typed Probabilistic Assessor: An Independent Reproduction and a Preregistered Study of Calibration and Selective Escalation Authors: Gowthamkumar Nandakishore Published: 2026-09-27 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.33843v1 The shipped Laya Typed-Decisions checkpoint, a 421M-parameter ModernBERT-large assessor that answers typed choice/noul/score questions over workflow state, is uniformly under-confident. The signed confidence-accuracy gap is $-0.214$, every occupied reliability bin's accuracy exceeds its confidence, and that sign uniformity collapses every binned ECE variant to the same value, $0.214$. The card frames the risk as over-confidence; the measured direction is the opposite, and the direction decides which way a confidence-gated cascade fails. A single disjointly fitted temperature ($T=0.469$, sharpening) removes most of the miscalibration (held-out ECE $0.204$ to $0.037$) and outperforms the shipped per-option-count table. The frozen selection rule instead chose isotonic regression, which overfit and failed its held-out NLL contrast on both tracks, so hypothesis H2 is not supported. Re-running the released checkpoint on its full official test split reproduces the card's headline accuracy ($0.767$ vs. $0.766$). The retrospective E1 reproduction preceded the analysis freeze; E2-E8 were prospectively preregistered, and 20 of 22 executed confirmatory tests reject under Benjamini-Hochberg FDR at $q=0.05$ (two descoped). The frozen gate beats random escalation but misses its 10% accepted-set error target on both tracks, an exploratory out-of-distribution probe finds no zero-shot transfer (accuracy $0.617$), and every score measures agreement with a synthetic teacher whose self-agreement ceiling ($0.735$) the specialist exceeds. Per-decision predictions, run manifests, and the frozen preregistration are in the ancillary files. The author has no affiliation with the model's publisher, the dataset's publisher, or TypeSafe. --- ### ChemOPD: Multi-Teacher On-Policy Distillation for Multi-Task Chemical Reasoning Authors: Yaoyao Xu, Xinjian Zhao, Xiaozhuang Song, Xuemin Chen, Tianshu Yu Published: 2026-09-27 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2609.33838v1 Large language models are increasingly expected to support diverse chemical reasoning capabilities within a unified model. One approach is to develop specialized capabilities separately and consolidate them through multi-teacher on-policy distillation, but this raises two questions: how should specialization be organized, and how should specialist guidance be integrated? We introduce ChemOPD, which addresses both. We estimate task affinities from supervised fine-tuning gradients and solve a constrained mixed-integer program(MIP) to construct partially overlapping specialist groups. During distillation, we retain a generalist teacher trained on all tasks so that specialist guidance supplements rather than replaces its supervision. Our anchor-residual objective gradually increases the routed specialist's contribution on student-generated responses. On ChemCoTBench, affinity-guided specialization produces task-dependent gains over the generalist teacher and improves several capabilities beyond semantic task grouping. Yet stronger teacher-side performance does not automatically yield stronger students: with the same specialists and routes, anchor-residual OPD improves most reported metrics over specialist-only distillation and realizes a larger share of the available teacher gains. These results highlight specialization and capability integration as connected but distinct design problems in chemical reasoning. --- ### DeltaSeek: Toward Active Perception in Evolving Construction Environments Authors: Sanjay Acharjee, Md Nazmus Sakib Published: 2026-09-27 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2609.33836v1 Construction environments evolve continuously, causing large geometric changes that degrade static mapping and registration performance. This necessitates active perception, where robots deliberately select sensing configurations to resolve the environment's current state. We present DeltaSeek, an initial framework toward active perception in evolving built environments. While our broader objective is a system that reasons about where, how, and when to observe, this paper addresses a critical prerequisite: how a robot's sensing embodiment constrains the observations it can acquire. We formalize an embodiment's permissible observation set and evaluate with a Husky A300 equipped with a UR5e on an IFC-derived benchmark under chassis-mounted and wrist-mounted RGB-D configurations, scoring observations by geometric visibility and effort by drivable distance. In a room-scale scene with eight controlled changes spanning four observability conditions, exhaustive evaluation over 240 permissible base poses and five arm postures shows that two changes admit no chassis viewpoint whatsoever, while the wrist camera resolves both. For changes observed by both embodiments, the median base travel is $6.0$~m for the wrist camera and $15.2$~m for the chassis camera. These results distinguish sensing limitations from acquisition costs, clarifying whether an observation is impossible or simply requires more travel. --- ### An Active-Bottleneck Mechanism for Weak-to-Strong Generalization Authors: Mohammad Zeinalpour, Amir Najafi Published: 2026-09-27 Categories: stat.ML, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.33835v1 Weak-to-strong generalization (W2SG) occurs when a student trained on a teacher's predictions outperforms that teacher. We study when this happens under fully converged, ridgeless two-stage learning, with no early stopping, no explicit regularization, and no assumption that the student is more expressive than the teacher. In two-stage linear regression, a teacher is fit from $n$ labeled examples and a student is trained solely on the teacher's predictions on $m$ fresh, unlabeled inputs. Although both stages share the same hypothesis class and the same training rule, we show that the student outperforms the teacher exactly when $m$ lies in an explicit intermediate range: too few pseudo-labels leave the student without enough signal, too many let it inherit the teacher's noise. Under power-law covariance, we derive this range in closed form as a function of the spectral decay and noise level, including regimes where the improving region splits into two disjoint intervals of $m$. We then study a random-feature model in which the student has strictly more features than the teacher, and identify two regimes, again given by explicit thresholds: one where improvement occurs only for $m$ in a bounded interval, and one where it occurs only once the student width $N_S$ exceeds an explicit threshold. Both regimes are governed by a single "active-bottleneck" principle: whichever of $m$ or $N_S$ is scarcer controls how much teacher error is filtered out, while increasing the other resource only reduces estimation noise. Together, these results show that finite data and finite width can themselves regularize a two-stage learner, with no explicit mechanism doing so. --- ### CLIMB-flow: Coupled Linear Inverse posterior sampling via Multiscale-Based flow Authors: Zeqiu Yu, Ruizhi Yuan, Mathews Jacob Published: 2026-09-27 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2609.33834v1 Diffusion models are now widely used in Bayesian inverse problems in imaging as priors, where latent diffusion models are often used for larger scale problems to keep the computational complexity and model-size manageable. Unfortunately, the auto-encoder based compression results in loss of spatial detail. In addition, the optimization is converted to a non-linear problem. In this paper, we introduce a posterior sampling algorithm customized for the pyramidal/cascaded architecture, which relies on a coarse to fine hierarchical strategy to generate images in the pixel domain. We present CLIMB-Flow which alternates between three steps: an end-point estimation from the current coarse and noisy image, data-consistent update of the clean image, and re-noising it back to the level the network expects. Together these steps sample the posterior at that scale using an approximate Gibbs sampling from two conditional distributions. Experiments on ImageNet, CelebA, AFHQ and fastMRI span inpainting, deblurring, super-resolution and accelerated MRI, with PSNR gains of 1.37-7.66 dB over the strongest competing method on CelebA and pixel-domain reconstruction up to 512x512. --- ### One Attack to Fool Them All: Highly Transferable Black-Box Adversarial Attacks on Frontier MLLMs Authors: Sen Nie, Jie Zhang, Zhongqi Wang, Shiguang Shan, Xilin Chen Published: 2026-09-27 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2609.33833v1 Adversarial attacks have long posed a fundamental threat to machine learning systems. As multimodal large language models (MLLMs) rapidly evolve and become widely deployed, assessing their vulnerability to such attacks is essential for their safe use. In this work, we investigate whether a single adversarial image can consistently mislead diverse frontier MLLMs in black-box settings. We propose O-Attack, a highly transferable black-box attack framework. This framework builds on our insight that surrogate models contain a broad, high-level, cross-modally aligned semantic space. This space extends beyond final-layer outputs and provides multiple semantically consistent representations that remain underexploited by existing attacks. Within this space, O-Attack anchors aligned representations, progressively broadens semantic conditions, and optimizes perturbations through semantic consensus to promote consistent target alignment. By fully exploiting this space with the same surrogate models as M-Attack, O-Attack raises attack success rates on GPT-5.4 (29.1% to 77.2%), Claude-4.6 (42.8% to 81.6%), and Gemini-3.1 (38.2% to 80.9%). Extensive experiments across 24 MLLMs show that O-Attack outperforms six state-of-the-art methods in black-box transferability, with consistent effectiveness across prompts and improved efficiency and imperceptibility. This work exposes the practical safety risks posed by black-box adversarial attacks against frontier MLLMs, underscoring the need for more rigorous robustness evaluation and more effective defenses. --- ### Achieve What You Imagined: Learning to Align Actions with Visual Plans Authors: Yuheng Qiao, Ziran Wei, Xiaohan Wang, Daqiang Guo, Yichen Luo, Zhibo Pang, Peng Zhou, Sichao Liu Published: 2026-09-27 Categories: cs.RO, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.33832v1 World-action models can jointly predict future visual observations and robot actions. However, discrepancies may exist between their visual predictions and the consequences implied by generated actions. We observe that WAMs can often generate visually plausible task-completion outcomes before producing action sequences that reliably achieve them. Consequently, we treat the WAM-generated visual prediction as a goal-conditioned visual proposal rather than a directly executable plan. We use a frozen action-conditioned world model to predict action-conditioned consequences and construct feedback based on consistency between the two future predictions and alignment with the terminal goal. Leveraging this feedback, we employ Flow Policy Optimization (FPO) to optimize the action head of the WAM. This framework avoids online robot interaction and additional training of task-specific reward models. Across four real-world UR5 manipulation tasks, our method increases the mean success rate from 43.4% to 75.1%, compared with 61.4% for $π_{0.5}$. These results show that cross-model prediction discrepancy can provide useful feedback for improving robot policies under the evaluated manipulation tasks. Website: https://imagine-to-achieve.github.io/ --- ### dOPT: Differentiating Conic Optimization via Geometric Reduction Authors: Fengyu Yang, Connor W. Magoon, Tyler Watts, Shahar Z. Kovalsky Published: 2026-09-27 Categories: cs.LG, math.OC arXiv: https://arxiv.org/abs/2609.33828v1 Optimization layers enable the incorporation of structured constraints and decision problems into learning systems. Training such systems requires differentiating through the embedded optimization problem, which can be challenging for general conic programs. We introduce dOPT, a solver-agnostic framework that, rather than differentiating the full conic formulation, reduces it at a computed primal-dual solution to an equality-constrained quadratic program that preserves the reference solution and its first-order sensitivity. The reduction captures the local first- and second-order conic geometry relevant to differentiation and remains well defined at singular configurations. Computing solution derivatives then requires a single symmetric linear solve, independently of the forward solver. We derive explicit reductions for convex NLPs, QPs, SOCPs, and SDPs. Numerical experiments validate the computed gradients and show favorable backward-pass scalability, with substantial speedups over existing differentiable conic optimization methods as problem size increases. --- ### Vestrum: Improving Agent Harnesses by Adapting Their Verification, Structure and Memory Authors: Jayant Parashar, Eugene F. Douglass, William C. Bastian, Suchendra M. Bhandarkar Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33822v1 An agent harness controls how a language model accesses information, uses tools, preserves memory, and checks its work. Improving this software is costly when each evaluation requires a long interaction with an environment. We introduce Vestrum, a framework that turns failures in execution traces into scoped harness changes without training the task model. Its organizing overhypothesis is that tasks of a shared kind may exhibit recurring failures whose remedies transfer within that kind. Vestrum expresses failures as recognizable classes, proposes changes across verification, retrieval, decomposition, and knowledge synthesis, and screens their scope before evaluating them as a bundle. A persistent lessons file informs subsequent proposals. Across five settings and two baseline harnesses, the frozen harnesses improve held-out performance: UltraHorizon rises from 47.6 to 59.8 over GAM, Terminal-Bench 4 Hard from 63.7% to 70.3% of checks passed over Claude Code on eight held-out tasks at 1.03x test cost, and cell-type annotation agreement from 67.5% to 77.8% on held-out sections of one slide, alongside gains on LoCoMo and AMA-Bench. Across our searches, verification grounded in evidence helped both intermediate steps and final answers, at lower cost at intermediate steps, while critics asked to rebuild finished answers broke more than they repaired. On the three memory benchmarks, Vestrum also scores above the evaluated GEPA configurations in every paired evaluation. --- ### Augmenting Visual Anomaly Detection with Automated Interpretability Authors: Antonio De Santis, Arsenio Leo, Marco Brambilla Published: 2026-09-27 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.33818v1 Visual anomaly detectors identify deviations from known-normal data, but their anomaly signals may mix evidence of actual anomalies with benign visual variation. We investigate whether automated interpretability can augment visual anomaly detectors by identifying and intervening on different components of this signal. We decompose PatchCore nearest-normal residuals into sparse features using Sparse Autoencoders (SAEs), and provide high-activation and contrastive non-active examples to a Multimodal LLM, which describes each feature and labels it as anomaly, distractor, or uncertain. These labels guide interventions in the SAE hidden representation, where distractor features are suppressed and anomaly features amplified. The edited representation is then used to reconstruct patch embeddings, which are rescored with PatchCore. Across 40 categories from four benchmarks, applying both interventions jointly improves macro-average image-level AUROC from 0.8724 to 0.8857 on source data and from 0.8066 to 0.8210 under synthetic corruptions. On three additional RobustAD categories with real acquisition shifts, the same interventions improve AUROC from 0.8745 to 0.9056 on source data and from 0.6069 to 0.6599 under real acquisition shifts. Finally, individual feature interventions across all 43 categories show that the MLLM labels are aligned in aggregate with how features differently affect normal and anomalous images. --- ### Dual-Vocabulary Language Model for Cross-Tokenizer Distillation Authors: Kedi Chen, Chen Lin, Yutao Sun, Wei Zhang Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33816v1 On-policy distillation (OPD) bridges teacher supervision and student behavior, but different teacher-student tokenizers introduce misalignment in both input tokenization (#1) and output logits (#2). Existing approaches address the former by matching same-text spans or converting tokens to bytes, often losing fine-grained token information or disrupting the native-token paradigm, while for the latter, strategies such as ranking, padding, or key-token selection retain only shared logit dimensions, resulting in much distribution loss. In this paper, we propose Dual-Vocabulary Language Model (DVLM), which replaces the teacher's LM head with a new student-vocabulary projection head and obtains full-dimensional student logits (for #2). To support student tokens (for #1), it takes a Parallel-Tokenized Sequence (PTS) as input, which concatenates the original teacher-tokenized sequence and a re-tokenized sequence formed by independently converting each student token into a teacher-token group. To avoid inference inconsistency with the original teacher tokens, the Hybrid-Prefix Attention (HPA) further restricts re-tokenized groups to their corresponding teacher prefix and uses its last state as the aggregation of the original student-token representation for projection into the student vocabulary space. Similarly, via the combined use of PTS and HPA, the DVLM teacher can provide distribution-aligned supervision with the student's input-tokenization and output-logit during OPD. Experimental results demonstrate that our DVLM teacher has a similar converged loss as the original teacher model and enables student models to improve performance across six reasoning tasks. --- ### Annealed Sinkhorn with Momentum: Certified Unregularized Optimal Transport in Linear Memory Authors: Samuel J. K. Chin, Maximilian Schiffer Published: 2026-09-27 Categories: math.OC, cs.LG arXiv: https://arxiv.org/abs/2609.33814v1 We characterize Bregman Douglas-Rachford splitting (BDRS) for unregularized discrete optimal transport and develop an anytime primal-dual certificate in linear memory. We first establish that BDRS coincides with warm-started Inexact Proximal point method for exact Optimal Transport (IPOT) using a single inner Sinkhorn iteration. By eliminating the primal transport plan from the updates, we derive an equivalent dual formulation that reveals BDRS as annealed Sinkhorn under an implicit inverse-linear temperature schedule, with an additional log-scaling momentum term and a cooler kernel. While this explains the role of the temperature parameter in BDRS as an initial temperature, it also reduces the solver's memory requirement from quadratic to linear. Utilizing this annealing perspective, we introduce overrelaxed BDRS, which combines annealing and overrelaxed scaling within a single recursion. We derive a primal-dual certificate for both methods that can be evaluated in linear memory without transport plan construction, thus providing a computable stopping rule. On the DOTmark benchmark, combining momentum with the cooler kernel produces substantially smaller optimality gaps than annealed Sinkhorn under the same schedule. For pixel-level color transfer between $1024\times1024$ images, BDRS attains a lower repaired transport cost than MDOT-TNT with a 9$\times$ speed up, reaching a relative duality gap of $1.59\%$ in 24 minutes. We further demonstrate a color transfer with $4238\times2365$ images, yielding 10 million pixels per image and approximately one hundred trillion implicit transport entries, reaching a best relative duality gap of $2.41\%$ and $2.80\%$ within 35 hours in each direction on a single NVIDIA L40S GPU. --- ### Identical Runs, Different Results: Benchmarking AI Coding Agents on Open-Weight Models Authors: Eduardo Ariño de la Rubia, Szilard Pafka Published: 2026-09-27 Categories: cs.SE, cs.LG arXiv: https://arxiv.org/abs/2609.33812v1 Repeated runs of the same coding agent are known to give different benchmark scores. We ask what that variation means for a team running an agent on its own task, by intensive replication on one machine-learning task: an agent improves the training code of an XGBoost classifier for airline delays, and a holdout it never sees scores the result. Across 584 runs, we compare six agents on six open-weight model endpoints, run six agent-model pairings 52 times each under fixed settings, and repeat three of them on a larger model from the same family. Identical runs of one pairing varied more than the pairings differed from one another, so comparisons of a few runs ranked them unreliably; resolving the agent differences we observed would take tens to more than a hundred runs of each. Runs on the larger model scored clearly higher, but by less than one run-to-run standard deviation, and the gap was more than twice as large with one agent as with the others. Fewer than one run in twenty broke the task's data rules, but those runs held the highest scores. Rejecting those runs first and keeping the best compliant result among a few attempts reliably improved the delivered model, even though a few runs could not rank the agents. On flights from a later year, the delivered models kept only a third of their gain over the starting code. At list prices, cost differed more than twentyfold between two agents on the same model, mostly through the prompt cache. Agents and models should be evaluated as pairings, over repeated attempts, with compliance reported beside quality. Data, code and every delivered program: https://github.com/earino/identical-runs-different-results --- ### Eyes on the Road: A Naturalistic Comparison of MTW Rider Gaze in Urban Indian Traffic Authors: Prerak Srivastava, Bhaiya Vaibhaw Kumar, Kavita Vemuri Published: 2026-09-27 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.33811v1 Motorized two-wheelers (MTW) dominate Indian roads but remain underrepresented in driver behavior research. This study presents the first large-scale analysis of MTW driver gaze behavior in naturalistic, heterogeneous urban traffic, using the \textit{myEye2Wheeler} dataset. A semantic segmentation pipeline (YOLOv11 + SAM2) was used to extract object-level gaze metrics under two attention modes: direct gaze (foveal overlap) and central vision (parafoveal monitoring). Results reveal a functional division: central vision supports broad monitoring, while direct gaze enables brief, selective sampling. Novice riders exhibit road-anchored scanning, returning to the road between object fixations, while experienced riders form longer chains of attention across multiple objects. The findings suggest that experience primarily refines temporal rhythm rather than altering allocation strategy and reduces object-class effects in gaze patterns. These findings offer new insight into MTW attention structures and inform future work on behavior modeling and safety systems. --- ### Controlling Speaking Rate in Autoregressive TTS via Activation Steering Authors: Francesco Verdini, Antonis Asonitis, Aref Farhadipour, Marzieh Razavi, Pierre-Edouard Honnet, Vijeta Avijeet, Juan Pablo Zuluaga Gomez Published: 2026-09-27 Categories: cs.SD, cs.CL, cs.LG, eess.AS arXiv: https://arxiv.org/abs/2609.33810v1 Autoregressive text-to-speech (TTS) systems synthesize natural speech but, once trained, offer little control over speaking rate. We show that speaking rate can be steered at inference time, without retraining, by clamping a single decoder block's activation along a discovered speed axis. A decoder-block analysis recovers the rate axis, a neutral operating point, and a per-step intensity scale; at inference, the activation's projection onto this axis is set to a fixed scalar. Learning this direction from synthetically time-stretched and time-compressed speech yields rate control that largely preserves speaker identity, generalizes across model architectures, and maintains high naturalness in objective and human evaluations. Unlike standard additive steering, which breaks at the slow extreme, clamping remains stable on all three systems tested; at moderate targets, the better rule depends on the model. Finally, we show that rate information is decodable across layers but causally steerable only within a mid-depth window, and demonstrate the effectiveness of our approach on the public Seed-TTS-Eval benchmark. --- ### CodeActionBench: Evaluating Agentic Code-as-Policy for Embodied Manipulation Authors: Yiheng Lyu, Xueying Jiang, Wenhao Li, Shijian Lu, Gongjie Zhang Published: 2026-09-27 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.33807v1 How well can general-purpose multimodal models turn visual understanding and reasoning into embodied manipulation via executable code? We introduce CodeActionBench, a benchmark of 25 manipulation tasks that evaluates this capability through agentic Code-as-Policy. Without task-specific fine-tuning, demonstrations, external specialist perception or grasp modules, privileged scene state, or predefined task policies, agents should select visual evidence, form task-relevant 3D estimates, construct manipulation targets, and iteratively execute and revise their policies. A shared robot API provides RGB observations, calibrated geometric operations, robot feedback, and bounded motion, leaving task-dependent decisions to the evaluated agent. Fixed task instances, resource budgets, and a hidden physical-outcome verifier support controlled comparisons across models and harness configurations. Extensive evaluations across nine configurations and 675 attempts achieve success rates ranging from 2.7% to 73.3%. The strongest configuration, GPT-6 Astra with Codex CLI, solves 22 of 25 tasks at least once in three attempts, demonstrating the best performance while still leaving substantial room for improvement. Trajectory analyses reveal difficulties in spatial alignment, object retention, and completion judgment, including task failures despite successfully completed motions. CodeActionBench provides a controlled testbed for measuring how general-purpose models translate their capabilities into manipulation behavior and for examining typical failure scenarios in that process. --- ### MinkowskiPE: Minkowski Positional Encoding for Spatiotemporal Perception Authors: Yuhao Li, Louie Hong Yao, Tianyi Shi, Hanqun Cao, Hongxia Hao, Zhen Zhao, Shengchao Liu Published: 2026-09-27 Categories: cs.LG, cs.AI, cs.CV, physics.comp-ph arXiv: https://arxiv.org/abs/2609.33804v1 Modeling spatiotemporal coupling is a key challenge in building physical intelligence across scales, from microscopic to macroscopic. Existing models capture such structure broadly through physics-motivated dynamical formulations or learning-motivated architectures. The former provide stronger priors but may constrain flexibility, whereas the latter are more flexible but leave the spatiotemporal coupling largely implicit. We therefore seek an approach that combines flexible learning with an explicit geometric bias for jointly modeling time and space. To this end, we propose Minkowski Positional Encoding (MinkowskiPE), which uses joint temporal and spatial coordinates to parameterize Lorentz transformations applied to query and key features. With MinkowskiPE, the query-key attention score depends on position only through the relative spacetime displacement between the two tokens and is therefore invariant to global translation of the coordinates. This paradigm retains the standard dot-product attention interface and remains compatible with efficient attention implementations. We evaluate MinkowskiPE on microscopic molecular dynamics and macroscopic video prediction tasks, achieving the best results on all nine multi-trajectory molecular evaluations and reducing KTH video-prediction MSE by 9.9% relative to the best baseline while using roughly one-tenth as many parameters. --- ### Diffusion Reward Models Authors: Xiangyang Wang, Bingxiang He, Zeyuan Liu, Jiaze Wang, Ziqing Qiao, Yuxin Zuo, Huan-ang Gao, Cheng Qian, Wenbin Zhang, Ran Li, Youbang Sun, Ning Ding et al. Published: 2026-09-27 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.33803v2 Reward models underpin the alignment of large language models, yet the dominant designs reduce each prompt--response pair to a point estimate or to a distribution from a fixed parametric family. This is at odds with human preference, which is inherently multimodal: the same response can be reasonably judged in many ways, and no single family covers all of them. To better fit this structure, we introduce DRM, a Diffusion Reward Model that recasts reward modeling as conditional density estimation over $p(\mathbf{r}\mid x,y)$. Conditioned on a frozen LLM encoder, a lightweight Diffusion Transformer denoises Gaussian noise into a reward vector, placing no parametric assumption on the output distribution and naturally representing its multimodal structure. A single architecture handles both multi-attribute regression and pairwise preference data, and at inference $N$ samples form an empirical reward distribution that can be aggregated into a scalar, a variance, or quantiles. Across five benchmarks, DRM matches or surpasses baselines under matched data and backbone, stays competitive with much larger discriminative, distributional, and generative RMs despite its modest training scale, and recovers multimodal reward structure where conventional heads collapse to a point. Uncertainty-aware rejection and lower-confidence-bound (LCB) aggregation further demonstrate that DRM can exploit distributional information beyond a scalar reward to improve reward-model decisions. Downstream RLHF experiments additionally show that using DRM as the training-time reward leads to improved policy performance, directly validating the practical benefit of diffusion-based reward modeling for RLHF training. --- ### Autonomous phase discovery Authors: Shiyu Zhou, Yuxuan Zhang, Sebastian Wetzel, Roger Melko, Xiu-Zhe Luo Published: 2026-09-27 Categories: quant-ph, cond-mat.str-el, cs.LG arXiv: https://arxiv.org/abs/2609.33802v1 Understanding quantum phases of matter has long relied on physicists' intuition and mathematical tools such as symmetry and topology. Remarkably successful as these approaches have been, they provide no universal way to explore a Hamiltonian space whose organizing principle is not known in advance. In this work, we introduce a fully autonomous system combining differentiable programming and unsupervised learning for quantum phase discovery. The search evaluates ground-state data along an adaptive trajectory rather than on a predetermined parameter grid. We demonstrate the system with three different solvers and benchmark it against random sampling at equal ground-state-evaluation budgets. On a generalized cluster chain hosting up to $200$ distinct phases, the search finds up to $25$ more phases at the same budget, and matches random sampling given thirty times its budget. On a $50$-parameter Chern insulator, it reaches sectors not obtained by the simple harmonic constructions considered here, in a family whose inverse problem remains open, while recovering all sectors found by sampling. Our results establish autonomous, gradient-driven exploration of Hamiltonian space as a practical route to discovering quantum phases without phase labels or a prescribed target phase. --- ### Binding Multiple Modalities via Multimodal Wasserstein Barycenter Authors: Xiaole Tang, Jiayi Xu, Xiang Gu, Yan Yang, Jian Sun Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33800v1 Multimodal learning beyond two modalities commonly leverages a specific modality (e.g., text) to bind other modalities. However, how to establish a more balanced representation space that approximates shared semantics while respecting the holistic geometry of $n$-modal data remains challenging. In this work, we present BaryBind, which aims to transport the specific modality towards the Wasserstein barycenter (WB) optimized across all modalities and introduces a volumetric alignment objective to establish a unified semantic space around the WB embedding. Specifically, we project specific modalities to the WB, which minimizes the average Wasserstein distances to multimodal distributions and serves as the anchor for subsequent alignment. We then construct a barycenter simplex, whose volume is taken as a similarity metric for global alignment centered at the WB. Experiments show that BaryBind achieves competitive performance in text-video-audio retrieval, classification, videoQA, and cross-modal generation tasks, along with robustness under modality absence and scalability to more than three modalities. Code is released at https://github.com/xl-tang3/BaryBind. --- ### RAISE: Reinforcing Access Control Policy Synthesis in LLMs via Symbolic Evaluation Authors: Yingming Zhou, Adarsh Vatsa, William Eiers Published: 2026-09-27 Categories: cs.SE, cs.CR, cs.LG arXiv: https://arxiv.org/abs/2609.33796v1 Translating natural-language access-control requirements into policies requires careful reasoning about permissions, constraints, and exceptions, and even frontier LLMs often produce policies that violate the intended authorization semantics. We construct CedarInstruct, to our knowledge the first dataset that supports both training and semantic evaluation for formally verifiable Cedar policy synthesis. It contains 5,800 scenarios across 44 domains and 1,408 representing a single synthetic organization, each with a verified target policy and an executable verification plan. On this data we introduce RAISE, which trains policy synthesizers from formal verification in two stages, verified supervised fine-tuning (SFT) followed by a reinforcement learning (RL) stage that learns from verifier signal. We find that SFT succeeds largely by letting models express authorization logic they already have, since untrained models rarely write valid Cedar but often reason correctly when they do. After SFT, how the verifier's information is used matters more than how much of it is used. Of six RL instantiations that consume progressively richer verifier signal, only RAISE-OC improves meaningfully on SFT; it turns failed checks and symbolic counterexamples into guided exploration and learns from the result with off-context GRPO. With about 5.4K verified scenarios and LoRA fine-tuning, RAISE-OC trains Qwen3.5-9B to surpass zero-shot GPT-6 Astra and Claude Opus 5 by 13.33 and 16.26 percentage points in semantic success on held-out scenarios, and training transfers to the independently constructed CedarBench. --- ### Do We Really Need KL Divergence for On-Policy Distillation of Large Language Models? Authors: Wenze Lin, Jiyuan Long, Jiale Zhao, Shenzhi Wang, Xitai Jiang, Ce Luo, Rui Lan, Qianli Ma, Fukang Wen, Hui Wu, Liyuan Chen, Shuoling Liu et al. Published: 2026-09-27 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.33791v1 Since the advent of knowledge distillation, KL divergence has been the standard loss in distillation. Recently, on-policy distillation (OPD) has emerged as an efficient post-training paradigm for LLMs. As a distillation method, OPD naturally inherits KL divergence as its standard loss. However, in this work, we find that KL divergence may not be necessary for OPD. We show that simply preserving the update direction is sufficient for effective OPD. As long as the update direction is toward the teacher, OPD works. More precisely, it is not the direction of every token, but the direction of a small subset of tokens where the teacher and student disagree strongly. We first show that simply assigning a reward of (+1) to tokens where the teacher probability is higher than the student probability and (-1) where it is lower, which merely encourages updates toward the teacher, reproduces almost the same training mode as OPD with reverse KL. We further show that only the direction of a small subset of tokens with large teacher-student disagreement is critical, and training works as long as their update direction is toward the teacher, even if other tokens are pulled away from the teacher. And as an application of these findings, we introduce Consensus Multi-Teacher On-Policy Distillation (C-MOPD) to improve Multi-Teacher On-Policy Distillation (MOPD). Unlike MOPD, which routes each sample to a single teacher and may cause capability conflicts across domains, C-MOPD lets every sample be supervised by all teachers. Experiments show that C-MOPD consistently outperforms MOPD on both math and code benchmarks. Our code is available at https://github.com/LeapLabTHU/KL-Free-OPD. --- ### Multi-Aperture PPG with MAPIS: Spatial Optical and Temporal Coherence Fields Authors: Shuguang Wang, Yuanjing Wang Published: 2026-09-27 Categories: physics.optics, cs.CV arXiv: https://arxiv.org/abs/2609.33790v1 Conventional reflectance photoplethysmography (PPG) typically reduces tissue optical responses to one or a few detector channels. Matrix Pinhole Image Sensing (MAPIS) instead simultaneously samples multiple aperture-dependent optical signals. We analyzed 44 recordings of 30 seconds each from a single MAPIS device to characterize the spatial relationships among DC intensity, pulsatile AC, perfusion index (PI), and temporal coherence across Red and infrared channels. AC increased sublinearly with DC at both wavelengths, resulting in higher PI toward lower-DC apertures, consistent with a first-order relative path-sensitivity interpretation. A second counterintuitive spatial trend was observed in temporal coherence: lower-DC and lower-AC apertures exhibited higher autocorrelation and greater cardiac-periodic spectral concentration. IR signals also showed consistently higher temporal coherence than corresponding Red signals. These observations demonstrate that multi-aperture PPG resolves reproducible spatial relationships among optical intensity, fractional pulsatile sensitivity, and waveform coherence that are largely averaged together in conventional single-channel PPG. The present study establishes within-device relationships and motivates controlled studies of the underlying photon-path mechanisms. --- ### LA-CPD: Local-Evidence-Aware Change-Point Detection for Human-LLM Authorship Segmentation Authors: Qing Yang, Zhenyu Mao, Zixiang Luo, Zezheng Wu, Xinghe Cheng, Qinggang Zhang, Jingwei Zhang, Jiapu Wang Published: 2026-09-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.33787v1 As LLM-generated text becomes increasingly human-like, accurately localizing LLM-authored spans in human-LLM co-authored documents is important for attribution and accountability in cases involving copyright infringement, fraud, and other harmful uses of AI-generated content. Sentence-level detectors provide local authorship evidence, but content variation can cause score fluctuations even among sentences from the same source, creating spurious boundaries. Recovering a coherent document partition therefore remains challenging when both the number and locations of authorship transitions are unknown. We propose Local-Evidence-Aware Change-Point Detection (LA-CPD), a structured method that transforms noisy sentence-level score sequences into coherent authorship segments. Given scores from a frozen local detector, LA-CPD combines a length-weighted within-segment residual with a windowed two-mean contrast to capture segment consistency and sustained changes around candidate cut points. Dynamic programming optimizes cut locations for each candidate count, while an AIC-style criterion selects the final partition, yielding sentence labels, authorship boundaries, and maximal LLM-authored spans. On a held-out human-LLM co-authored test set, LA-CPD outperforms WCP+AIC, increasing sentence-level accuracy from 0.747 to 0.796 while improving boundary localization and LLM-span delineation. --- ### Is your uncertainty map wrong, or is its target? Exact diagnostics for the Tweedie diagonal, and a gradient-free alternative Authors: Vicent Ribas, Anna Oliveras Tous Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33786v1 A diffusion model can predict a follow-up medical scan from a baseline, but a clinician needs a per-voxel map of where that prediction can be trusted. Many such maps approximate the diagonal of the Tweedie posterior covariance, and are evaluated against another approximation of it, so whether the estimator or the target limits them is unclear. We compute the exact diagonal on six checkpoints across fourteen model-corpus conditions. Hutchinson at M=200 tracks it at rank agreement of at least 0.92 everywhere, yet in four of the fourteen the exact diagonal is anti-correlated with the denoising error, reaching -0.13, so a faithful estimator reproduces that reversal. All four are real-image conditions; on the models' own samples the reversal does not appear, so evaluating on generated samples flatters this family. What limits these maps is the target, not the estimator. We then introduce Tweedie Probe-Tangent (T-PT), a gradient-free residual probe that corrupts one model-supported prediction repeatedly and measures the voxel-wise variance of the denoiser's response. T-PT reads a different functional of the same Jacobian, and its exact second-order form ranks with the diagonal wherever the diagonal reverses; at thirty probes it returns a map too unstable to reproduce that ranking, while Hutchinson at M=5 already reproduces it, so T-PT there is not evidence against the reversal. We offer it as an instrument, not a better approximation. On brain MRI at full resolution, where every Jacobian-based estimator we test runs out of memory, T-PT leads a twenty-chain Monte-Carlo ensemble on five of eight endpoints inside tissue and trails it on none, at 16x fewer network evaluations; over the whole volume the ensemble leads, and fifty chains close the tissue gap. On lung CT the ensemble is ahead throughout. Both lose most of their discrimination where the change is, which remains open. --- ### Surprising Success, Repeated Failure: Entropy-Guided Credit Assignment for Exploration in LLM Reasoning Authors: Woongyeong Yeo, Minki Kang, Chanuk Lee, Sangwoo Park, Jinheon Baek, Sung Ju Hwang Published: 2026-09-27 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.33781v1 Reinforcement learning with verifiable rewards (RLVR) enhances reasoning in large language models (LLMs) through outcome-level feedback, yet recent approaches to finer-grained credit assignment often require auxiliary models, additional sampling, or privileged information. Although policy entropy provides a readily available signal, prioritizing uncertain positions under both reinforcement and penalization concentrates penalties where failed responses still retain alternatives for recovery, which can suppress opportunities for exploration. To address this, we introduce Entropic Advantage Policy Optimization (EAPO), an entropy-guided credit assignment method that treats success and failure asymmetrically. Specifically, motivated by the observation that success under uncertainty is less repeatable while confident failures tend to recur, EAPO couples normalized policy entropy with the sign of the response advantage to reinforce surprising success and correct repeated failure. It assigns stronger reinforcement to high-entropy decisions in successful responses and stronger penalties to low-entropy decisions in failed responses, while attenuating penalties at uncertain positions to preserve opportunities for recovery. By redistributing the response advantage across tokens, EAPO derives token-level credit directly from existing rollout signals without additional supervision. We validate EAPO on a range of reasoning tasks across both base and reasoning backbones, demonstrating that it achieves the best overall performance. We further show that EAPO promotes more effective exploration, broadening problem coverage and generating more diverse candidate answers. --- ### Selecting Diverse SFT Traces Improves Post-RL Generalization Authors: Dylan Zhang, Mingyuan Wu, Jinning Li Published: 2026-09-27 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.33780v1 Verified solutions are not equally useful for preparing reasoning models for reinforcement learning (RL). We present a comprehensive study of route diversity, the variation in the sequences of reasoning steps in supervised fine-tuning (SFT) data, and propose a lightweight, rule-based fingerprint to select for it. From one pool at one budget, with matched training recipes and checkpoints, selecting diverse rather than similar routes improves post-RL problem coverage across puzzles and mathematics, including on problems harder than those seen in either training stage. In synthetic experiments, route-diverse SFT improves OLMo3-7B's pass@8 by 16.9 points on environments held out from SFT. In a single-model condition, where one model writes every candidate, diverse selection gains up to 6.2 points of mean pass@8 across 10 mathematics benchmarks. Pre-RL diagnostics suggest why: diverse SFT can produce both successful and failed attempts on more prompts despite slightly lower mean accuracy, giving group-relative RL more prompts with a learning signal. On 3 open-source corpora, our CPU-only selector, without model calls, outperforms more expensive alternatives in every comparison of mean post-RL performance. These results identify reasoning-route diversity as a practical criterion for selecting SFT data that better prepares models for RL. --- ### Evidence-Inference Reconstruction: When The Evidence Is Recalled But The Reasoning Goes Wrong Authors: Megan Diehl, Ser-Nam Lim Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33778v1 Modern multi-hop LLM agents are equipped with built-in mechanisms to detect errors in intermediate reasoning steps. Such errors trigger corrective actions from these agents, which mostly follow the paradigm of retrying the steps or the reasoning trajectories. Not only are these retries expensive, we present in this paper that they are also potentially unnecessary. To this end, we introduce Evidence-Inference Reconstruction (EIR), which uses structured state to guide one retrieval trajectory, accumulating source evidence in the process. We show that as long as the relevant evidence has been collected, EIR is capable of generating the correct answer in a single final model call even if erroneous evidence has been mixed in due to incorrect intermediate reasoning steps. In one evaluation, using Haiku 4.5 and GPT-4.1 Mini, we evaluate EIR on matched 1,000-question subsets of HotpotQA, 2WikiMultiHopQA, and MuSiQue, showing that EIR improves Answer F1, the overlap between the model's and the correct answer, over the baseline by 8.3--32.8 points, Agentic SSR by 10.6--29.1 points, and Reflexion by 1.1--15.9 points. Additionally, we show that EIR averages 4.85 total model calls per question, compared with 35.29 for Agentic SSR and 12.41 for Reflexion. Together, these results corroborate EIR's central premise: separating evidence retrieval from the final answer model call can improve answer accuracy while utilizing substantially less computation. --- ### Tokens Change, Structure Endures: Spectral Watermarking for Generated Speech Authors: Kanghwi Lee, Kyeongseok Jeong, Jeongmin Liu Published: 2026-09-27 Categories: cs.SD, cs.CL, cs.CR, cs.LG, eess.AS arXiv: https://arxiv.org/abs/2609.33774v1 Watermarking is a promising tool for establishing the provenance of AI-generated speech. While many neural audio watermarking methods rely on a separately trained watermark generator, token-level watermarking is a training-free alternative that operates directly during generation. Its main weakness is retokenization: decoding generated speech to a waveform and encoding it again can change token identities and erode the watermark. To make the watermark robust to these changes, we propose Redwing, REtokenization-Durable Watermarking IN Generation. It builds a graph from the token substitutions observed under retokenization, whose Laplacian yields a basis that assigns similar values to tokens likely to substitute for one another. Over this basis, embedding and detection functions are jointly optimized to preserve watermark signal through retokenization while limiting embedding distortion and detector variability on unwatermarked speech. On the Moshi full-duplex system, after eight consecutive passes of Mimi resynthesis, Redwing achieves 80.7% TPR at a calibrated 1% FPR, compared with 8.3% for KGW and at most 7.3% for WMAR. It also has the highest TPR after eight passes through three other neural codecs (77.5-93.0%), and the gains generalize to TTS models at a speech-quality cost close to that of KGW. These results show that retokenization is not merely a source of noise: its transition structure can be exploited as a design principle for robust token-level watermarking. ---