# Research Papers — 2026-10 · bundle page 26 of 156

> Full-text markdown of papers #1251–1300 of 7799 published in 2026-10, newest first. AI-readable mirrors of the full 7799-paper 2026-10 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-10/page/26
[← Previous bundle](https://aipod.dev/papers/archive/2026-10/page/25.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-10/page/27.md)

### Target-Token Asymmetry Can Masquerade As Depth Specialization

Authors: Danlin Li, Miao Li, Xu Han, Jiecheng Lu
Published: 2026-10-14
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2610.19344v1

Layer-ablation studies often perturb early or late layers of a language model, compare the damage to prediction on two corpora, and read a larger early-layer effect on one corpus as early-layer specialization for its task. Such a comparison averages over the tokens that each corpus asks the model to predict, and these target tokens differ: math solutions, for example, ask for far more whitespace and numbers than encyclopedia prose. We show that this difference can create the appearance of depth specialization. We split a cross-corpus early--late damage interaction by target-token type and recompute it under a target mixture shared by both corpora. Between GSM8K and Wikitext, a positive interaction replicates across five model instances. In the three fully decomposed models, only whitespace has a positive interaction. Removing whitespace reverses the pooled effect in all three, and matching the full target mixture reverses it in two. The diagnosis holds under position matching, adversarial reassignment of token types, and a continuous readout. The risk can also be anticipated from the tokenizer alone. A screen computed before any model runs places every held-out pair it scored in the region its label predicts, and on the pairs it rates at high or medium risk, removing whitespace reverses every positive pooled estimate. A smaller content-word effect one token ahead follows the share of content targets that the unablated model already predicts. A cross-corpus depth effect therefore supports an interpretation of task specialization only after the target-token mixture is reported, the effect is decomposed by token type, and a mixture-matched summary keeps its sign.

---

### Counterfactual Gate-Support Decomposition: Separating Selection from Redistribution in Training-Data Filtering

Authors: Hyojung Han, Jongmin Kim, Seung-Hun Jeon
Published: 2026-10-14
Categories: cs.LG, cs.CL
arXiv: https://arxiv.org/abs/2610.19343v1

A training-data quality gate changes both which examples survive within regions of the input space and how training mass is distributed across those regions, so a gate-versus-random comparison does not isolate these mechanisms. We introduce Counterfactual Gate-Support Decomposition (CGSD), a four-arm, budget-matched design that crosses gate eligibility with support distribution and decomposes the observed gate effect into within-support selection, between-support redistribution, and their interaction. The practical constraint is finite-pool realizability: we characterize mass and token-assembly conditions under which the required arms can be constructed, together with a support-extinction case in which the ordinary design is undefined. On Korean-English and Korean-Japanese retrieval, the mean within-support selection component is the only one that exceeds its calibrated null band (+29.7 and +5.6 holdout items; 30/30 positive replicates), decisively in Korean-English and marginally in Korean-Japanese; redistribution and interaction remain unresolved at the available power. The same pattern holds for a second, unrelated gate family -- a model-confidence threshold rather than a deterministic verifier -- judged against its own null (+27.3 items, 9/9 positive), so the result is not a property of one gate's mechanism. On a mathematics gate, zero-eligibility supports leave the ordinary design undefined at every budget, while the extended design is constructible only inside narrow budget bands that an exhaustive scan locates and that the natural operating point misses -- the assembly condition separating a property of the pool from a limitation of the packer. CGSD therefore returns a decomposition, a constructible-but-underpowered design, or an explicit non-constructibility result, and says which; it prevents support redistribution from being silently interpreted as sample-level gate quality.

---

### Why Do Agent Teams Fail Across Boundaries? A Failure Taxonomy for Agent-to-Agent Systems with Private States

Authors: Sarath Shekkizhar, Prathyusha Jwalapuram, Zixuan Ke, Vaidehi Patil, Jiayu Wang, Srijan Bansal, Trong Huy Nguyen, Dingfeng Jiang, Dongqi Liu, Ruochen Zhao, Wei Lu, Giuseppe Carenini et al.
Published: 2026-10-14
Categories: cs.MA, cs.AI
arXiv: https://arxiv.org/abs/2610.19342v2

Agents are increasingly interacting with other agents that act for different principals (people or organizations on whose behalf they work), such as a buyer's personal agent negotiating with an enterprise's seller agent. In such agent-to-agent (A2A) scenarios, each agent has its own environment and holds information and authority that the other cannot see. Moreover, the interests of each agent may be aligned (cooperative), partly aligned (mixed-motive), or opposed (competitive). Existing multi-agent system failure studies assume one orchestrator, shared workspace, or single principal. We ask how A2A systems fail when these assumptions are removed. We study more than 2,500 recorded runs of A2A systems built from different agent harnesses and models, spanning 488 scenarios across eight families (legal, CRM, security, commerce, etc.). From these traces, we derive $A^2$-Failures, a taxonomy of 20 A2A failure modes in four categories: (i) discovery and delegation, (ii) grounding and record consistency, (iii) verification, and (iv) conduct and fidelity. Three findings hold across every harness, model, and scenario studied: (1) discovery and interaction are driven by the phrasing of the task rather than the availability of peer agents; (2) no agent verifies a peer's factual claim on its own initiative, so an A2A system inherits the reliability of its least reliable participant unless the protocol or mechanism performs the check; and (3) the aggregate outcome score credits the agent completing the task while staying blind to the coordination, verification, and conduct failures beneath it, so A2A systems must be evaluated per agent, against each agent's own principal. We release the run logs behind the taxonomy at https://huggingface.co/datasets/a2a-salesforce/a2-failures.

---

### MMKG-R1: Constructing Multimodal Knowledge Graphs via Reinforcement Learning

Authors: Zongyu Wu, Yilong Wang, Xiaochen Wang, Minhua Lin, Yuan Zhong, Fenglong Ma, Xiang Zhang, Suhang Wang
Published: 2026-10-14
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.19341v1

Multimodal knowledge graphs (MMKGs) organize information across different modalities to support multimodal Graph Retrieval-Augmented Generation (RAG), which aims to improve the performance of multimodal large language models (MLLMs). The effectiveness of these multimodal Graph RAG methods largely depends on the quality of the constructed MMKGs. However, most existing methods construct MMKGs without explicit guidance on graph quality, while reinforcement learning (RL)-based methods that rely on final-answer correctness may provide only indirect feedback on MMKG quality. To address these limitations, we propose MMKG-R1, a novel MMKG-quality-driven RL framework for training an MLLM as a multimodal knowledge constructor. To be specific, MMKG-R1 introduces multiple quality-aware rewards to directly evaluate both the quality of information extracted from individual knowledge sources and the quality of the assembled MMKG. These rewards are then used to guide the RL training. Experiments across multiple datasets show that MMKG-R1 constructs MMKGs that improve downstream question-answering accuracy.

---

### Can We Believe AI Peer Review?

Authors: Zongrui Yang, Haoyuan Li, Zhongsheng Wang, Gillian Dobbie, Jiamou Liu
Published: 2026-10-14
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.19339v1

As large language models become widely used in scientific review and manuscript revision, AI scores are increasingly serving as feedback for assessing paper quality and guiding edits. Yet whether a single AI score reliably reflects the reviewer's judgment, and whether revision gains based on that score are endorsed by other reviewers, remain understudied. Using 200 papers from ICLR 2025 and 2026 and approximately 14,000 AI reviews, we study score reliability, score-distribution estimation, and cross-model agreement on revision gains. Repeated reviews fluctuate even under identical conditions. Finer-grained scoring protocols reduce score variance but also systematically shift mean scores, while the gains assigned to manuscript revisions depend strongly on the reviewer.

---

### LLM-jp-4 Technical Report

Authors: LLM-jp, :, Chaoran Liu, Daisuke Kawahara, Hideyuki Tachibana, Hirokazu Kiyomaru, Jun Suzuki, Koshi Eguchi, Kouta Nakayama, Qianying Liu, Rio Yokota, Ryo Takahashi et al.
Published: 2026-10-14
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2610.19331v1

This paper presents LLM-jp-4, a family of Japanese sovereign large language models (LLMs) developed with an emphasis on transparency, reproducibility, and practical usability. LLM-jp-4 comprises an 8B dense model and a 32B-A3B mixture-of-experts model, both trained from scratch on approximately 12 trillion tokens. To support independent verification and reproduction, we establish explicit criteria for selecting training data, external models, and software, and make the development process publicly available. Evaluation results demonstrate substantial improvements over previous LLM-jp models and competitive performance against open-weight models of similar scale. By documenting the development principles, training procedures, and empirical findings of LLM-jp-4, we aim to provide a reference for future sovereign AI development.

---

### NavAble: A Large-Scale Dataset and Synthetic Data Generation Pipeline for Blind Navigation

Authors: Hochul Hwang, Jahir Sadik Monon, Soowan Yang, Shiven Umeshbhai Patel, Anh Nhat Hong Nguyen, Kien Nguyen, Keshav Garg, Dylan Gage, Duretti Hordofaa, Anshu Anjna, Eshed Ohn-Bar, Donghyun Kim
Published: 2026-10-14
Categories: cs.CV, cs.RO
arXiv: https://arxiv.org/abs/2610.19330v1

Reliable recognition of accessibility-critical objects (e.g., audible pedestrian signals, door-activation buttons, and handrails) is essential for assistive navigation technologies that support safe, independent mobility for blind and low-vision (BLV) users. Yet these categories are underrepresented in existing large-scale vision datasets, and the few datasets that include them suffer from limited class coverage, inconsistent annotations, and poor diversity. We introduce NavAble, a large-scale dataset spanning 11 accessibility-critical object classes, combining 41K curated open-source images with 8K newly collected, densely annotated real-world images. To extend scale and diversity, we develop a synthetic data generation pipeline that renders high-fidelity 3D assets across 37 environments under diverse styles and viewpoints, yielding 452K frames with rich ground-truth annotations from 565 assets. The pipeline minimizes manual effort by combining automated web image crawling, VLM-based filtering, SAM 3D-based asset generation, and convenient camera trajectory configuration, enabling large-scale data generation at a diversity unattainable through manual collection. Experiments show that fine-tuning existing segmentation models on the NavAble Dataset achieves reliable accessibility-object segmentation (>80% mIoU on our held-out test set). We also demonstrate the effectiveness of synthetic data augmentation, which yields up to +3.5 mIoU over real-world-only fine-tuning. Beyond benchmark evaluation, validation on egocentric data from two mobility-assistive robots demonstrates improved perception reliability in real-world navigation scenarios. Together, NavAble establishes a scalable foundation for BLV navigation perception, directly addressing the data scarcity and embodiment gaps. The dataset and pipeline are publicly available at https://huggingface.co/datasets/NavAble/NeurIPS_2026_BLV.

---

### SonoWorld: Towards Interactive Pose-Controllable Ultrasound World Models

Authors: Zefan Yang, Zheyuan Zhang, Boris Maihe, Yikang Liu, Hang Yu, Ankush Mukherjee, Xiao Chen, Pingkun Yan, Terrence Chen, Shanhui Sun
Published: 2026-10-14
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.19327v1

Robotic ultrasound can standardize image acquisition and reduce operator workload and variability, but policy learning remains limited by the lack of realistic, controllable simulation environments. While controllable world models have advanced rapidly in natural-image domains, ultrasound world modeling remains challenging because each frame captures only a cross-section of unobserved 3D anatomy, causing small probe motions to induce large appearance changes. Existing approaches rely on robot-tracked poses and latent-space prediction, but do not directly generate the observations required for interactive control. We propose SonoWorld, a causal ultrasound world model that learns video dynamics from unlabeled clinical videos and grounds them to 6-DoF control using a small set of freehand clinical scans tracked with an EM sensor. SonoWorld is trained in three stages: pretraining a latent-action-conditioned video diffusion model on around 6.9 million unlabeled multi-organ ultrasound frames; grounding the learned video dynamics to metric probe poses with a small EM-tracked thyroid dataset; and applying distribution matching distillation to enable causal, long-horizon autoregressive generation with selectively caching eviction tokens at inference time. We evaluate controllability through pose-following fidelity, round-trip consistency, and axis-specific control under synthetic actions on held-out tracked scans, and further validate it with a user study. In model-based action discrimination, SonoWorld correctly ranks the target-directed probe motion over an equal-magnitude opposite motion in up to 74.8% of trials. To our knowledge, SonoWorld is the first causal, pose-controllable ultrasound world model learned primarily from freehand clinical videos, providing a data-driven interactive environment for robotic ultrasound data acquisition and policy evaluation.

---

### Why don't Graph Transformers attend to the same token?

Authors: Alexandre Bismuth
Published: 2026-10-14
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.19324v1

In large language models (LLMs), attention concentrates heavily on the first token. This 'attention sink' is believed to protect deep models against over-mixing. Even though graph transformers face a related pressure, we find that a SoTA graph transformer, GRIT, rarely forms sinks on molecules, and that it sinks less when trained on larger graphs, the opposite of the trend in LLMs. In principle, a graph can prevent sinks: we prove and show empirically that its symmetry and locality limit the attention any node can receive. On molecules, however, with distinguishable atoms and global attention, neither limit prevents sinks; what stops them lies in the model. We find that the attention scoring function is the main cause. Replacing GRIT's gated logit with a dot product raises its sink rate from 0.09 to 0.68 on identical graphs. In Graph Transformers using dot-product attention, removing the sink gives about the same error as removing a random node, while GRIT's most-attended nodes matter to its predictions. Attention sinks are therefore a property of dot-product attention rather than of Transformers. In our experiments, they form when dot-product scoring has a distinguishable node to concentrate on.

---

### TopicMerge: A Divide and Conquer Approach to Topic Modeling

Authors: Matthew Raffel, Lizhong Chen, Sai Munikoti, Sameera Horawalavithana
Published: 2026-10-14
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2610.19323v1

Recent advancements in topic modeling have split into two main paths: efficient embedding-based methods that offer scalability but are often outperformed on standard clustering metrics, and Large Language Model (LLM) approaches that achieve state-of-the-art clustering performance but at the cost of prohibitive computational expense and poor scalability. This paper introduces TopicMerge, a novel multi-stage pipeline that bridges these approaches to achieve both state-of-the-art clustering performance and computational efficiency. TopicMerge first employs an efficient clustering pipeline to deliberately over-segment a document corpus into a large set of fine-grained, semantically pure micro-clusters, which are then given initial semantic labels. It then introduces a novel, iterative merging algorithm where an LLM, guided by local neighborhood context, arbitrates on consolidating candidate pairs. These approved merges are then selectively executed and semantically relabeled by the LLM, hierarchically building broader, more coherent topics. Extensive experiments on the Bills and Wiki datasets show that TopicMerge achieves state-of-the-art clustering performance while reducing required LLM calls by over 92% compared to document-level methods, offering a practical and scalable solution.

---

### Generalization Analysis of Biased Stochastic Gradient Methods for Minimax Problems

Authors: Shuang Zeng, Yunwen Lei, Yiming Ying
Published: 2026-10-14
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.19322v1

Minimax optimization refers to a class of optimization problems with some variables to minimize and other variables to maximize. Biased stochastic gradient methods have shown practical success in solving minimax problems to either improve robustness, enhance communication efficiency, or decrease computational costs. These successes motivate a lot of theoretical works to study the convergence of biased stochastic gradient methods, while their generalization analysis remains untouched. In this paper, we present the first framework to study the stability and generalization of biased stochastic gradient methods for minimax problems. We establish a connection between stability and generalization for minimax problems by relaxing the existing bounded gradient assumption to a bounded second moment condition. We then introduce a generalized Lipschitz-type condition on bias and gradient estimators, and derive a general stability bound to clarify the connection among bias, gradient estimators, and stability. We apply our general analysis to Zeroth-order and Clipped stochastic gradient descent ascent (SGDA), and derive stability bounds that match those of SGDA under appropriate smoothing/clipping parameters. We combine stability and convergence analyses together, and derive optimal excess risk bounds of order $1/\sqrt{n}$, where $n$ is the sample size.

---

### Global-in-time Spectral Duhamel-Picard Method for Post-Exposure Bake Simulation and Physics-Informed Neural Operators for EUV Lithography

Authors: Doyun Kim, Mircea Dusa, Werner Gilijns, Philippe Leray
Published: 2026-10-13
Categories: physics.comp-ph, cs.LG, physics.app-ph
arXiv: https://arxiv.org/abs/2610.20655v1

We present a global-in-time spectral Duhamel-Picard method with polynomial expansions for post-exposure bake (PEB) simulation. The coupled acid-quencher reaction-diffusion equations are reformulated as nonlinear Volterra integral equations using the Duhamel principle. Full trajectories are represented by finite polynomial expansions, with temporal coefficient fields determined over the bake interval by Anderson-accelerated Picard iteration, thereby avoiding sequential time marching. In 3-D PEB simulations, we employ Chebyshev and Legendre temporal bases while evaluating the Duhamel integrals using the corresponding Gaussian quadrature rules. We demonstrate that the proposed method simulates a large domain of $1200\times1200\times100$ voxels on the order of one minute. The representation is further incorporated into a physics-informed neural operator (PINO) with a mixed-transform variant of a factorized Fourier neural operator, which is trained without precomputed solution trajectories. Together, the integral formulation and its PINO extension enable accelerated, time-marching-free prediction of concentration fields at arbitrary times within the bake interval.

---

### Truthful and Expected-Payment-Optimal Model Routing in Graph-Based Agentic Workflows

Authors: Jiayuan Liu, Mingyu Guo, Jiarui Gan, Vincent Conitzer
Published: 2026-10-13
Categories: cs.GT, cs.CL, cs.MA
arXiv: https://arxiv.org/abs/2610.19316v1

Assigning routine tasks to cheaper large language model (LLM) providers can reduce generation cost, but switching providers may require processing accumulated context again. Keeping dependent tasks with the same provider can instead reuse eligible context. Graph-based workflows therefore couple routing decisions across tasks. Independent providers may also privately know their real-time unit costs and strategically misreport them. We formulate this routing problem as a single-parameter procurement auction and propose STAR (Strategy-proof Topology-aware Agent Routing), which minimizes virtual-cost-weighted graph workloads. Its allocation rule and payments, computed by integrating workload over possible cost reports, make truthful reporting utility-maximizing regardless of other providers' reports and cover each truthful provider's execution cost. With independent cost distributions and non-decreasing virtual costs, STAR additionally minimizes the user's expected total payment to providers. Controlled synthetic experiments separate the execution-cost benefit of topology awareness from the payment benefit of virtual-cost optimization: the former grows with fan-out and transferred payload, while the latter increases gradually along the tested prior-shape heterogeneity path. A paired sensitivity study quantifies how payment savings vary with prior misspecification while strategy-proofness is preserved.

---

### Historical-Feature Residual Adaptation for Sequential Screening

Authors: Yixiang Hong, Yuchen Zhang
Published: 2026-10-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.19315v1

Sequential screening policies must use acquired feedback effectively under a fixed experiment budget. We study acquisition-time residual adaptation: a small ranking correction built from fixed historical features and anchored to base scores recorded before feedback is revealed. The interface accepts a history-conditioned scorer and a candidate feature table; its primary instantiation augments AssayFormer's ten-dimensional representation with 64 historical directions while retaining the pretrained model. In retrospective evaluation on 20 official-test screens from 15 recorded sources, a comparison fixed before target-outcome access increased source-macro fraction of hits recovered from 22.58% to 28.61%, exceeding all six prespecified comparators. A same-feature ordinary ranking head achieved 25.73%. Supplementary strong logistic, rank-fusion, and unprojected controls achieved 26.42%, 26.02%, and 26.78%, respectively. A separate model-independent pilot on 13 reused development screens reconstructs the features without AssayFormer assets. Across three seeds, residual adaptation improves random-forest hit recovery from 28.87% to 31.27%, while logistic recovery changes from 30.22% to 29.78%. The results show effective pretrained-policy adaptation and useful transfer to an independent nonlinear base, with gains that depend on the base model.

---

### PersEvalBench: Can LLM Evaluators Judge Contextually Appropriate Use of Persona Information?

Authors: Xinyi Liu, Hooshang Nayyeri, Emine Yilmaz, Dilek Hakkani-Tur, Charith Peris, Hari Thadakamalla
Published: 2026-10-13
Categories: cs.CL
arXiv: https://arxiv.org/abs/2610.19314v1

Large language model (LLM) evaluators are increasingly used to assess personalized assistants, yet whether they can reliably judge personalization has rarely been tested. We introduce PersEvalBench, a benchmark testing whether an evaluator can recognize appropriate and inappropriate uses of user history. It contains 403 multi-episode scenarios where user information should persist, transfer, or cease to apply. The benchmark covers nine controlled error subtypes across three families: under-personalization, over-personalization, and misapplied-state errors. Human audits confirm that generated errors match their intended types and that the prescribed uses of user information are reasonable. Across five LLM evaluators, recognition of controlled personalization errors remains limited even with the full dialogue history. On the human-evaluated subset, under-personalization shows the largest gap: humans recognize 94.4% of these errors, compared with 63.9% for the strongest reported evaluator. Evaluators often recognize that a response is problematic but reverse the error direction, mistaking failure to follow an applicable preference (UNDER) for use of an inapplicable preference (OVER). Easier evidence access offers little help: highlighting the relevant history turns raises error recognition only slightly, suggesting that the remaining difficulty lies in judging what role that information should play in the response.

---

### Did the LLM Answer the Question? A Semantic Reconstruction Game for Jailbreak Evaluation

Authors: Liam Davis, Romain Emil-Popescu, Suman Jana
Published: 2026-10-13
Categories: cs.LG, cs.CR
arXiv: https://arxiv.org/abs/2610.19306v1

Jailbreak attacks against large language models (LLMs) are evaluated using attack success rate (ASR), which depends on correctly identifying responses that fulfill prohibited requests. Existing evaluators can mistake refusals or topically related responses for successful attacks. We show that changing only the judge LLM can alter reported ASR by up to $82$ percentage points. We introduce the semantic reconstruction game, a training-free evaluator that decomposes compliance judgments into constrained semantic comparisons. Given only a response, an LLM reconstructs the request answered by its substantive content, excluding information merely repeated or refused. The game compares this reconstruction against similar decoy requests to determine whether it matches the original request, allowing abstention when neither candidate matches. Across three off-the-shelf LLM judges, the game achieves the lowest mean error against human labels among evaluated methods and reduces variation in reported ASR. It prioritizes avoiding false positives, even at the cost of missing some successful jailbreaks.

---

### Data-free Multistep Diffusion Distillation

Authors: Ngoc Duy Tran, Kien Do, Duc Thanh Nguyen, Truyen Tran
Published: 2026-10-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.19305v1

One-step diffusion distillation achieves strong performance on small datasets but remains insufficient for high-resolution or complex text-to-image settings. Extending to multistep distillation often requires training data, suffers from early-step degradation, inconsistent outputs across sampling budgets, and incurs high computational cost. We identify that these limitations stem from simultaneous training, where a single adapted network approximates multiple score functions of different student distributions at once, causing biased distribution matching and unstable optimization. Therefore, we propose Progressive Multistep Data-free Distillation (PMDD), which assigns a dedicated adapted network per student distribution while avoiding the memory overhead of maintaining multiple adapted networks. Combined with a stage-wise unforget loss to preserve earlier-step generations, PMDD reduces optimization bias in the student, provides faster convergence, and a single unified model that supports multiple inference budgets. Extensive experiments show that PMDD matches or outperforms data-free few-step baselines at every multistep budget, reaches FID 1.99 on ImageNet 64x64 in two steps, below its 512-NFE teacher (2.32), and achieves FID 7.76 on zero-shot COCO 512x512, while requiring 4-5x less training compute than data-free DMDv2 to support three inference budgets.

---

### DTDD: Divergence-Triggered Dynamic Distillation for Reliable On-Policy Supervision

Authors: Qijia He, Zequn Li, Spencer Luo, Tianle Chen, Pradyumna Narayana, Zixian Ma, Ranjay Krishna, Lan Nie, Sugato Basu
Published: 2026-10-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2610.19302v1

On-policy distillation (OPD) provides dense token-level supervision on student rollouts, but this supervision can become unreliable when the student and teacher diverge, either because of a large initial policy gap or compounding errors in long-horizon reasoning and tool use. We introduce Divergence-Triggered Dynamic Distillation (DTDD), an adaptive OPD framework that uses segment-level student-teacher divergence to trigger teacher takeover during training. When divergence triggers an intervention during reasoning, the student prefix continues to receive standard OPD supervision. The teacher then generates a recovery suffix, which the student learns through behavior cloning, using cross-entropy on the teacher-generated tokens. We evaluate DTDD in three regimes: long-horizon reasoning, distillation from a base initialization without offline supervised fine-tuning (SFT), and multi-turn agents. We also compare alternative trigger designs and teacher-intervention rates. DTDD reduces the post-warmup peak gradient norm by $8.6\times$ and, from a base initialization, reaches $21.3\%$ held-out math accuracy in $2.6\times$ fewer optimizer updates than OPD; both methods finish near $22\%$. It improves best-observed performance by $4.8$ and $4.1$ percentage points on ALFWorld with LoRA and agentic code, respectively. Ablations show that segment-averaged triggers outperform token-spike triggers. Behavior cloning is also more effective than the tested OPD surrogate for training teacher-generated recoveries. Across 23 experimental arms, takeover tends to help when the teacher substantially outperforms the OPD-trained student, provided that the teacher can recover from student-visited states, the student can learn those recoveries, and teacher intervention is neither excessive nor insufficient.

---

### Beyond ICD Codes: Semantic IDs for Generative Medical Coding

Authors: Yihang Cheng, Qian Zhang, Yonghui, Huang, Baoshuo Kan, Junyi Shen
Published: 2026-10-13
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.19301v1

Generative approaches to automatic International Classification of Diseases (ICD) coding use large language models (LLMs) to predict ICD codes from clinical documents. However, original ICD code strings are not ideal generation targets because the ICD hierarchy does not always reflect clinical relationships, and LLMs struggle to interpret and generate these codes. We introduce SemICD, a framework that trains LLMs to generate Semantic IDs (SIDs) instead. SemICD constructs SIDs by hierarchically clustering embeddings of textual code profiles, allowing semantically related codes to share prefixes while each complete SID uniquely identifies one ICD code. To align the SID tokens with clinical semantics, we employ two auxiliary tasks for bidirectional generation between textual code profiles and SIDs. Experiments on MIMIC-III and MIMIC-IV with four LLMs show that SemICD outperforms direct code generation, improving Micro-F1 by 1.34-8.00 points. Our code is available at https://github.com/youwillbethebest/SemICD.

---

### Verifier-Grounded Agentic RL for Provably Correct Code

Authors: George Ma, Yuhao Zhang, Bing He, Zeren Shui, Ferhat Erata, Murali Krishna Ramanathan, Somayeh Sojoudi, Hao Zhou
Published: 2026-10-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.19298v1

Reinforcement learning for code usually rewards passing tests, which a model can do while remaining wrong on the inputs the tests do not cover. A program verifier instead decides whether a program satisfies a formal specification for all inputs. Language models are almost always paired with verifiers at inference time only, and turning a verifier into a reward raises two difficulties. It reports only acceptance or rejection, which gives no gradient when every attempt fails, and a policy can cheat it, most simply by telling it to assume what the proof was supposed to establish. We build an agentic environment and reward that address both. The reward reads the verifier's diagnostics rather than its verdict, scoring a candidate by the weighted fraction of proof obligations it discharges, extended with hidden probes from a reference proof's annotations. An integrity audit outranks the verifier, rejecting any candidate that alters the specification or assumes what it was meant to prove. The model works as an agent against a live verifier over many turns, with a separately trained proof specialist as a frozen collaborator. Across Qwen3 models from 1.7B to 14B and a 35B Qwen3.6 mixture-of-experts model, held-out verification on code generation rises from 66.1% to 93.5% on DafnyBench and from 56.3% to 100% on SWE-Proof. This reward leads every other training signal we compare, including a binary pass-or-fail reward trained under otherwise identical settings, with the widest margin on code generation. When the agent must also deliver the repository patch, graded by each project's withheld tests, the fraction of held-out issues it resolves rises from 59.6% to 72.5%. Evaluated unchanged on three verification suites withheld from training, the checkpoints improve on all of them, including one whose language and verifier are absent from training.

---

### Qwenfant: Train Big Character-level BabyLMs on Concise Parallel Text

Authors: Lukas Edman, Alexander Fraser
Published: 2026-10-13
Categories: cs.CL
arXiv: https://arxiv.org/abs/2610.19294v1

This work concerns our submission for the Multilingual track of the 2026 BabyLM Challenge. We improve on our baseline using data that has been summarized and translated by an LLM, showing that higher-quality data can come from synthetic sources. We further improve with 2 simple changes: using a larger model, and using a character-level vocabulary. Both of these changes require LoRA-style fine-tuning for realizing their full potential. We dub our final Qwen3-based model Qwenfant, which, at the time of submission, ranks 1st in the Multilingual track.

---

### Whose turn is it? Context-guided model routing for stateful agents

Authors: Anikesh Parashar, Avik Dutta, Vashu Chauhan, Sumit Gulwani, Priyanshu Gupta, Hosein Hasanbeig, Vu Le, Ashish Tiwari, Jack Williams
Published: 2026-10-13
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.19292v1

Model routing can reduce the cost of LLM applications by assigning simpler requests to cheaper models without sacrificing the quality of the response. However, most routers treat requests as independent. Stateful agents do not operate under this assumption: a user turn may require multiple model and tool calls, its difficulty depends on the state produced by earlier turns, and changing models can sacrifice reuse of the conversational prompt cache. We study zero-shot, cost-aware routing at the user-turn boundary, systematically varying the routing rule, access to session history, candidate models that differ in size, reasoning effort, or model provider, and prompt-cache behavior. We evaluate on multi-turn versions of SheetCopilot and SpreadsheetBench 2, where persistent workbooks and cross-turn dependencies make isolated routing decisions especially brittle. At a matched offloading rate, giving the router session history improves task pass rate over the same router without history, with gains for every model pair; history-aware routing remains within one point of always using the expensive model. We also found that cost depends on execution dynamics rather than token prices alone: the cheaper model's per-task cost advantage narrows from 6.8x to 3.2x on longer sessions. A closed form cache analysis further shows that downward switches (expensive to cheap) recover their cache cost within two LLM calls, whereas upward switches do not break even under the evaluated conditions. Together, these results show that agent routing is a sequential decision problem: effective policies must account for what happened earlier in the session and for the asymmetric cost of switching models.

---

### Can LLM Agents Efficiently Identify Appliance Faults? Insights From Diagnosing a Simulated Appliance

Authors: Lucas H. McCabe, Sarah Lukens, H. Howie Huang
Published: 2026-10-13
Categories: cs.CL
arXiv: https://arxiv.org/abs/2610.19291v1

In so-called "Industry 4.0," integrated sensors, analytics tools, and other digital technologies supplement industrial workflows, potentiating enhanced visibility and automation. Prognostics and health management (PHM) and natural language processing (NLP) practitioners alike have explored the application of large language models (LLMs) to industrial asset management, but many prior works' focus on exam-like benchmarks has limited their assessment of LLMs' ability to automate practical operations. Herein, we introduce CoffeeBench, a fault diagnostics task for LLM agents, using a simulated appliance. The LLMs we evaluate struggle to consistently identify faults from the sensor information alone, but the best-performing ones can iteratively refine the fault-option space to eventually obtain a successful diagnosis. Our subsequent analyses, therefore, focus on: differentiating models by diagnostic efficiency (RQ1); examining the lagging performance of the three worst models (RQ2); and identifying viable heuristics for selective prediction (RQ3), including that initial diagnostic uncertainty and LLM-judged soundness of reasoning are high-signal predictors of diagnostic success. In addition to our benchmark results (Section 3.1), we provide practical insights for improving the performance of the open-weight models on CoffeeBench, such as by explicitly tracking episode state on their behalf to reduce timeouts (Section 3.2) or applying a deliberation loop to refine diagnostic hypotheses (Section 3.3).

---

### Kandinsky S2V 1.0: Audio-Driven Avatar Video Generation

Authors: Nikolay Semenov, Artem Riabinin, Sergey Bassenko, Anna Averchenkova, Julia Agafonova, Anastasia Aliaskina, Kirill Kozlov, Lev Novitskiy, Nikita Samsonov, Nikolai Gerasimenko, Denis Dimitrov
Published: 2026-10-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.19290v1

We present Kandinsky S2V 1.0, an open family of speech-to-video models for controllable audio-driven human animation, comprising Kandinsky S2V 1.0 Lite (2B parameters) and Kandinsky S2V 1.0 Pro (21B parameters). From a single reference image, a speech waveform, and a text prompt, the models generate long-form 25 FPS video in which lip motion, expression, pose, and gesture follow the speech. The architecture draws text and audio conditioning from one Qwen2.5-Omni model and adds a reference-aware three-scale visual memory. Training starts from the Kandinsky 5.0 Video foundation models (1) and combines mixed reference-memory layouts with an area-adaptive face loss. The training corpus contains 1.17M single-speaker clips selected by active-speaker detection, with captions generated directly in English and Russian. A side-by-side human evaluation against the strongest open-source baseline and two proprietary commercial models shows that Kandinsky S2V 1.0 Pro performs comparably to or better than these models on average and is also preferred for lip synchronization. On common speech to video benchmarks, Kandinsky S2V 1.0 Lite and Kandinsky S2V 1.0 Pro rank among the top performing models according to mean rank over five reported metrics. Beyond the base models, we provide π-Flow distilled variants of both Kandinsky S2V 1.0 Lite and Kandinsky S2V 1.0 Pro. The distilled Kandinsky S2V 1.0 Pro model generates video in 16 sampling steps without classifier-free guidance, achieving a 5.4 times end-to-end speedup over the base model, while matching or exceeding it on nearly all reported benchmark metrics. To accelerate open research and deployment in speech-to-video generation, we release the code, model checkpoints, and diffusers integration under the MIT license.

---

### From Efficiency to Sufficiency: The Cost-Aware Measurement Design Framework for Automated Essay Scoring

Authors: Yi Gui
Published: 2026-10-13
Categories: cs.CY, cs.CL
arXiv: https://arxiv.org/abs/2610.19287v1

Compute cost in automated essay scoring (AES) is usually treated as a technical efficiency issue rather than a measurement design concern. Yet deployment decisions require more than reporting latency, model size, or inference expense. In high-stakes settings, total burden includes validation, monitoring, adjudication, compliance, and governance; in instructional settings, pricing, licensing, and integration may dominate cost. This paper reframes AES as a cost-constrained measurement system and introduces the Cost-Aware Measurement Design (CAMD) Framework, a four-layer structure consisting of metering, allocation, unit economics, and decision, for translating computational and workflow resource records into decision-relevant cost constructs. Direct cost is specified as monetary outlay plus attributable labor and time, and two criteria, design attribution and auditability, determine what the cost boundary excludes. Burdens that satisfy neither criterion, such as institutional capability and governance posture, are represented as constraints on which designs are admissible rather than priced. The paper defines cost per defensible score (CPDS) and argues that AES systems should be compared by quality-cost sufficiency: minimizing cost among defensible scoring designs that satisfy measurement requirements. Scoring volume conditions all four layers, so cost per defensible score is treated as a function of scale rather than as a single transferable number. CAMD provides a portable basis for representing, comparing, and evaluating cost across AES studies and deployment contexts.

---

### Exact Scaling Laws for a Random-Features Bigram Model

Authors: Julian Enoch, Elvis Dohmatob
Published: 2026-10-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.19281v1

Understanding neural scaling laws is increasingly important for theory and practice, yet first-principles explanations remain limited even in simple settings. In this work, we study capacity-limited scaling in a bottlenecked bigram model with power-law (Zipf) token statistics, viewed as an analytically tractable toy model of a large language model. Fixed random embeddings serve as a proxy for representational capacity and induce an explicit bottleneck of width, allowing us to derive scaling laws for prediction regret as a function of model width, vocabulary size, and the Zipf exponent. Our results complement recent theoretical accounts based on optimization dynamics by isolating scaling due purely to representational constraints. In the appropriate regime, the theory recovers the functional form of classical neural scaling laws and suggests that heavy-tailed language statistics may underlie both optimization- and capacity-driven scaling. Experiments natural language datasets (WikiText2, OpenWebText) closely match the predicted curves.

---

### Dimension-Free Scaling Laws for Size Generalization: An Exponential Separation

Authors: Behrooz Tahmasebi, Melanie Weber
Published: 2026-10-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2610.19279v1

Modern machine learning models increasingly operate on variable-size structured data, such as sets and graphs. A central requirement is \emph{size generalization}: the ability of a model trained on inputs of one size to generalize to inputs of a different, potentially much larger, size. Classical statistical learning theory, however, primarily considers fixed input domains, and its guarantees often deteriorate with input dimension. Although extensive work has established when models on sets and graphs can be transferred across input sizes, such results do not characterize the statistical cost of doing so: the sample complexity of size generalization, and its relationship to classical in-domain learning, remain largely unknown. We develop a statistical learning theory of size generalization and establish a sharp exponential separation in sample complexity. Letting $n$ denote the number of training samples, we show that polynomial source decay is insufficient for efficient size generalization: while it yields error $n^{-Θ(1)}$ and hence polynomial sample complexity for in-domain learning, size-generalization error decays only as $(\log n)^{-Θ(1)}$, yielding an exponential separation. Thus, polynomial in-domain sample complexity can become exponential across sizes. We then identify an exponential source condition under which size generalization achieves error $n^{-Θ(1)}$, recovering polynomial sample complexity. These guarantees are independent of input dimension and depend only on the number of training samples, yielding dimension-free scaling laws. We also provide matching minimax lower bounds, establishing optimality. To our knowledge, this work provides the first dimension-free minimax theory of size generalization and a rigorous characterization of when, and at what statistical cost, learning can transfer across input sizes.

---

### Deep Learning for Atmospheric Turbulence Mitigation: A Comparison Between GANs and Transformers

Authors: Juan S. Carrillo, Jérôme Gilles
Published: 2026-10-13
Categories: eess.IV, cs.CV
arXiv: https://arxiv.org/abs/2610.19274v1

Atmospheric turbulence creates complex spatial and temporal distortions that significantly reduce image quality. This work compares two deep learning methods for turbulence reduction: Generative Adversarial Networks (GANs) and Transformer-based models. Both methods are trained and tested on the SOTIS dataset, which includes multi-frame sequences under different turbulence conditions with ground truth references. Our results show an important finding: GAN-based models perform better in severely degraded, low-quality situations, while Transformer-based models achieve better results in moderate to high-quality settings, especially in correcting geometric distortions. We also enhance the Turbulence Mitigation Transformer (TMT) by introducing a refined training strategy, an augmented input design that includes a temporally averaged frame as prior structural information, and a modified loss function that incorporates the Structural Similarity Index (SSIM). These improvements lead to consistent performance gains across various turbulence conditions. Overall, this study provides an in-depth evaluation of deep learning methods for turbulence reduction and offers practical guidance for selecting models in different imaging environments.

---

### Outcome-Routed Distillation and Reinforcement Learning for Diffusion Language Models

Authors: Wentao Lu, Tianyu Zhu, Jesse Clark
Published: 2026-10-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.19271v1

On-policy distillation provides dense supervision on the student's own samples, but its targets still come from the teacher. Verifier reinforcement learning (RL) instead uses an outcome signal, but it provides no direct constraint on capabilities outside the verifier. Combining the two is difficult for diffusion language models (dLLMs), because bidirectional denoising makes the policy likelihoods required by ratio-based RL unreliable. We develop a likelihood-free pipeline for a mixture-of-experts dLLM. Each rollout group is routed by verifier outcome. Mixed groups use ratio-free, group-centred REINFORCE. If every student rollout fails but a teacher succeeds, the model receives dense KL supervision on the student's denoising canvases. All other groups are skipped. A KL term anchored to the distilled checkpoint limits changes outside the verifier reward. The tool-calling teacher succeeds on only 18.6% of the training prompts. At 24 denoising steps, distillation exceeds the teacher by 6.77 percentage points, and routing improves the distilled checkpoint by a further 18.31 points. Routing lowers reasoning-trace accuracy by 0.71 points but raises direct-answer accuracy on the same items by 13.21 points. The anchored arm restores trace accuracy. Its tool-calling score differs by -0.16 points, with a confidence interval that includes zero, although this comparison also removes a small fallback distillation route. A domain-conditional recipe retains a 16.37-point tool-calling gain; a second seed gives 18.23 points and reproduces both preservation results, but the math result is seed-dependent. We pre-registered every arm and evaluated each comparison on paired items in a single sweep on one machine.

---

### Exploring Interpretability in Deep Learning-Based Forest Classification - Separating Similar Forest Classes using Sentinel-2 Time Series

Authors: Maximilian Kirsch, Christine Preisach
Published: 2026-10-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2610.19270v1

The ecological insurance theory posits that diverse forests are more resilient and better able to adapt to changing environmental conditions. Increasing drought and heat stress in Central Europe promotes mixed forests for maintaining ecosystem services. However, a diverse species distribution complicates remote sensing-based species mapping, as distinguishing pure from mixed species compositions at the pixel level remains challenging. We evaluated the potential of recurrent neural networks (RNNs) for tree-species classification from Sentinel-2 time series and investigated the spectral-temporal features underlying class discrimination, with particular emphasis on separating pure from mixed Norway spruce pixels. We compared RNN classification models with Random Forest (RF) and Extreme Gradient Boosting (XGBoost) and used Integrated Gradients (IG) to quantify feature attributions across spectral bands and time. The best-performing RNN achieved a Macro-F1 score of 0.851 on the test data, outperforming XGBoost by 0.012 and Random Forest by 0.103. Beyond classification performance, the attribution analysis revealed distinct spectral-temporal patterns associated with the separation of pure and mixed Norway spruce pixels. Attribution values increased particularly after approximately day of year (DOY) 125, while B11 showed the strongest contribution during periods of increasing spectral divergence between the two classes. These findings show that recurrent models can provide competitive classification performance while also revealing the phenological periods and spectral regions that are most relevant for distinguishing closely related forest compositions. Combining time-series classification with explainability therefore provides additional insight into the spectral-temporal mechanisms underlying tree-species discrimination and may support more detailed mapping of mixed forest structures.

---

### Whose Data, Whose Benchmark? A User-Centric Framework for Auditing Evaluation Resources

Authors: Mir Tafseer Nayeem, Davood Rafiei
Published: 2026-10-13
Categories: cs.CL, cs.AI, cs.CY, cs.ET, cs.HC
arXiv: https://arxiv.org/abs/2610.19268v1

Benchmarks shape how model capabilities are measured and interpreted, yet their scores can obscure whose conventions, contexts, and judgments define successful performance. We introduce a user-centric audit framework that distinguishes users who contribute benchmark content from those whose needs evaluation results are intended to represent. The framework operationalizes seven dimensions across benchmark content, metadata, documentation, and scoring procedures. Across 16 resources, we find that American-centered conventions account for 62.4-94.4% of matched signals in 11 text benchmarks, while geographic visibility in five image-text resources is concentrated and strongly associated with national GDP share (mean Pearson r = 0.780; Spearman p = 0.778). Documentation reveals a systematic source-target gap: all 16 resources operationalize source-user context, but none operationalize target-user context. Paired scoring experiments further show that meaning-preserving answer variants reduce DROP exact-match accuracy from 100.0% to 33.3%, while TruthfulQA exhibits metric-dependent reductions in reference similarity. Our framework and documentation checklist make user assumptions auditable, clarifying what benchmark scores represent and where evaluation claims are supported.

---

### MemeTrace: A Multimodal LLM-Based Agent for Safety-Aware Meme Provenance Analysis

Authors: Bo Shao, Yihan Ma, Yiting Qu, Michael Backes, Yang Zhang
Published: 2026-10-13
Categories: cs.AI, cs.CL, cs.CY
arXiv: https://arxiv.org/abs/2610.19267v1

Online memes are often carriers of unsafe content such as disinformation and propaganda. Their rapid evolution in response to news events and strong dependence on contextual information make it difficult for platforms and moderators to detect and moderate them effectively. Moreover, existing moderation studies typically classify memes as safe or unsafe based on individual images, lacking the ability to perform broader provenance analysis, i.e., where a meme originates, how it spreads, what cultural symbolism it represents, whether it poses potential risks to society, etc. To bridge this gap, we introduce MemeTrace, a multimodal LLM-based agent that automatically traces the provenance and potential risks of emerging memes. We evaluate MemeTrace on Meme100, a dataset of 100 real-world memes, using both human annotation and automatic metrics. The results demonstrate that MemeTrace substantially outperforms the VLM baseline. More importantly, MemeTrace achieves higher accuracy in detecting unsafe memes, especially those involving disinformation and sensitive content, and remains robust under visual perturbations.

---

### LAMP: 3D Source-Aware Lighting Estimation at Multiple Points

Authors: Yanqiao Wang, Jin Sun
Published: 2026-10-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.19266v1

Existing illumination estimation methods primarily recover lighting distributions for rendering, but typically leave the contributions of individual physical light sources at specific scene locations implicit. We address this gap by leveraging the scene understanding capabilities of vision-language models (VLMs) to connect local illumination with its physical sources. We introduce LAMP, a framework for point-conditioned, source-aware illumination reasoning from a single RGB image with marked query points. LAMP constructs a shared inventory of active light sources, ranks contributing sources for each query, and predicts their 3D query-to-source directions. We supervise these relationships using procedurally controlled indoor scenes, reusing independently rendered source responses to derive contribution targets across lighting configurations and obtaining exact directions from source and query positions. On the Infinigen test set, fine-tuning Qwen3-VL-8B increases dominant-source 3D PCK@0.5m from 0% to 70.6%, with 94.5% of directions within 12.5 degrees of ground truth among correctly localized sources. Combined with a lightweight network predicting dense RGB irradiance, LAMP achieves probe-rendering quality comparable to prior lighting-estimation methods while additionally localizing individual physical sources in 3D and explicitly attributing local illumination to them.

---

### Beyond Maliciousness: Static and Runtime Attack-Surface Profiling of Agent Skills

Authors: Ratnesh Pandey, Japneet Singh, Harsh Akshit, Sonali Tyagi, Cassie Crossley
Published: 2026-10-13
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2610.19264v1

Agent skills are Markdown documents that a tool-using agent reads as instructions and then acts on with the privileges of the session that loaded them. Installing one grants authority (the standing to direct the agent, not new privilege), and the size of that grant is what we measure. We present a two-phase attack-surface profiling methodology that combines whole-bundle static analysis and dependency provenance checking with instrumented sandbox execution and separate model-based assessment of the resulting traces. We apply the methodology to 6,309 listings from two marketplaces, selecting listings for execution without conditioning that decision on a prior threat or risk signal. Half the corpus carries evidence of shell-execution capability; the corresponding shares are 35% for filesystem read, 31% each for filesystem write and network egress, and 13% for credential access; seven listings in ten carry at least one elevated capability. That is the surface a user accepts at install time, and almost none of it is visible then. A permission model exists but is close to inert. Authors may list the tools a skill pre-approves in its own frontmatter, which removes confirmation prompts rather than restricting what the skill can reach; of the 6,061 listings whose frontmatter parses, 770 declare anything at all, the vocabulary cannot express credential or environment access, and nothing checks what it does express against the bundle. Declaring is also not declaring accurately: 59% of declaring listings understate what their bundle can do. ClawHub declares roughly twice as often as skillsmp, yet under-declares on 88% of its declarations against 31%. Triage at the marketplace boundary therefore cannot rest on what a listing declares. It must rest on what the bundle can do. These results position capability-oriented attack-surface profiling as a necessary complement to maliciousness detection.

---

### MIND: Instruction-Conditioned Incremental Grounding for Robots

Authors: Mohamed Abouagour, Junhyun Nam, Wonse Jo, Byung-Cheol Min
Published: 2026-10-13
Categories: cs.RO, cs.AI
arXiv: https://arxiv.org/abs/2610.19262v1

Robots following persistent language instructions must decide which cached room judgments remain valid as their environment changes. We present MIND (Maintenance via INstruction-conditioned Dependencies), a framework for maintaining target-room decisions without re-evaluating every room after each event. A language model compiles each instruction into dependencies for semantic qualification and target selection. The controller intersects semantic dependencies with event footprints, refreshes affected judgments, and updates reachability and selection separately. For fixed deterministic evaluation, planning, and selection, conservative dependency and footprint coverage preserves the decisions of full recomputation. On 240 five-event ProcTHOR sequences with an exact benchmark predicate, MIND reduces mean semantic evaluations from 30.6 to 11.2 while preserving all 1,200 post-event decisions. Frozen-evaluator experiments with Phi-4 and Qwen3-14B also preserve all 1,200 decisions, without improving the evaluators' underlying task accuracy. In 32 closed-loop AI2-THOR episodes using exact-predicate qualification and metadata-derived footprints, MIND matches the final target decision of full recomputation in every episode. Unitree G1 experiments demonstrate target reuse and reselection across room transitions without additional semantic evaluations under MIND, and abstention after a barrier and an object removal at 4 evaluations against 15 to 18. Additional materials are available at https://mind-icra27.github.io/.

---

### Long-horizon Video Generation with Slot Memory Addressed By Region Tokens

Authors: Eren Cetin, Julius Erbach, Vukasin Bozic, Hyojun Go, Zixiang Zhao, Nikolai Kalischek, Federico Tombari, Konrad Schindler, Dominik Narnhofer
Published: 2026-10-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.19257v1

Memory for long-horizon video generation must satisfy two competing requirements: it must remain bounded so generation can continue indefinitely, while still recovering what was observed at a place when the camera returns, regardless of how long ago it was seen. Reconstruction-based memories provide spatial recall by caching and reprojecting explicit geometry, but require growing storage and depend on estimated 3D structure. Recurrent memories keep a constant-size state, but retention is governed by temporal recurrence, causing old observations to fade with time. We introduce SMART, a slot memory addressed by region tokens, which combines bounded storage with spatially addressed recall. SMART stores content in a constant-size slot recurrence whose writes are addressed by the observing camera and whose reads are queried by the rendering camera, without constructing an explicit 3D representation or maintaining a growing cache. Consequently, recall is conditioned on inter-view geometry rather than on recency. Each read additionally provides a confidence estimate that conditions a chunk-causal diffusion transformer, allowing it to selectively trust retrieved content while relying on its generative prior elsewhere. On long trajectories in Memory Maze and Minecraft, SMART maintains high-fidelity recall far beyond its training horizon while using constant memory and per-step cost.

---

### Beyond Time Induction: Structured Conditional Neural Solvers for Parametric Stochastic-Volatility Pricing

Authors: Jose Miguel Lara Rangel, Mixtli Vazquez Mendoza
Published: 2026-10-13
Categories: cs.LG, cs.CE, q-fin.CP
arXiv: https://arxiv.org/abs/2610.19254v1

Physics-informed neural networks provide a flexible framework for learning parametric PDE solution families without labeled data, yet preserving accurate derivatives across heterogeneous regimes remains challenging. Structured neural solvers offer compact representations, but restrictive conditional parameterizations can limit adaptation across the solution family. We introduce BeTINN, a structured conditional architecture extending time-induced modulation through multi-directional time-parameter conditioning and a joint residual pathway. We evaluate BeTINN on the Heston stochastic-volatility PDE, a demanding parametric problem with mixed derivatives and economically relevant first- and second-order sensitivities. BeTINN improves joint accuracy across normalized option value and key sensitivity metrics relative to our baselines. Controlled ablations characterize the roles of its architectural components. We further find that validation physics loss is informative but does not fully determine differential fidelity, and that accurate solution values can coexist with localized derivative errors. These results support structured conditional adaptation and motivate derivative-aware evaluation when sensitivities are downstream quantities of interest.

---

### Multi-Sensitive Distributional Fairness: Joint Wasserstein Audits and Shapley Allocation

Authors: Agathe Fernandes Machado, Charlotte Jamotton, Arthur Charpentier
Published: 2026-10-13
Categories: stat.ML, cs.LG, stat.ME
arXiv: https://arxiv.org/abs/2610.19252v1

Fairness audits with several sensitive attributes are often performed one coordinate at a time. This can miss purely intersectional disparities: a score may be independent of each sensitive coordinate while depending on their joint profile. We study distributional demographic parity for a real-valued score using one-dimensional Wasserstein geometry. We distinguish two complementary constructions. Conditional and sequential audits characterize joint parity through their zero sets, but their positive values do not generally decompose a fixed joint Wasserstein disparity. By contrast, full-cell conditional quantile functions define a Hilbert-valued random variable whose variance is the joint disparity. Conditional expectations of this quantile process induce a monotone information game: every reveal order yields an exact martingale decomposition of the same joint disparity, while Shapley values provide an order-free attribution across sensitive attributes. Under independent sensitive attributes, the game reduces to a Hoeffding-ANOVA decomposition and Shapley splits each interaction energy equally among its members; under dependence, orthogonality is lost and explicit formulas reveal suppression and redundancy effects that marginal audits can hide. We give fixed-grid estimators and permutation calibration, and illustrate the framework on synthetic examples and ACSIncome. The resulting framework separates three questions that are often conflated: whether joint distributional parity holds, how a fixed joint disparity decomposes along an information path, and how that disparity can be attributed across sensitive coordinates without privileging an ordering.

---

### Copy-Paste Language Models

Authors: Nathan Godey, Yoav Artzi
Published: 2026-10-13
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2610.19247v1

Transformers are intuitively considered to be efficient at copying from their inputs. We actually show that the ability of LLMs to copy in context can be improved through explicit modeling. We propose Copy-Paste Language Models (CPLMs), which add a lightweight Copy-Paste head to the conventional language modeling (LM) head. CPLMs improve training efficiency, downstream performance, and offer new operating points in long-context and retrieval setups. We validate our approach by pretraining models from 135M to 1.7B parameters, showing improvements across all scales and backbones with negligible training and inference overheads. We explain the success of CPLMs by showing that they partially mitigate the LM head gradient bottleneck through an alternative gradient path.

---

### Refined Topology-Pass Scheduling: Damped Trust-Region Compensation for Dynamic k-NN Graphs in SE(3)-Equivariant Structure Prediction

Authors: Jiankai Huang, Hanwen Shen
Published: 2026-10-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2610.19246v1

Equivariant structure-prediction pipelines alternate coordinate updates with the construction of a k-nearest-neighbour (kNN) graph whose membership is a discrete, piecewise-constant function of the coordinates. Rebuilding that graph at every block boundary (PB) is the expensive part of an epoch, whereas reusing one frozen graph (F) is fast but leaves a topology gap. We study a two-pass design in which the cheap pass records a small amount of state (margins, kNN radii, measured per-boundary displacements) and a second pass compensates conservatively. We formulate the choice of the graph-construction basis as a damped fixed-point iteration whose unique fixed point is exactly PB, and derive (i) exact non-flip certificates with the two-sided constant $m_i/4$ and one-sided constant $m_i/2$, (ii) the minimum-variance (BLUE) replay gain $γ^*$ and the resulting feasibility crossover, (iii) an $\ell_\infty$ trust-region projection that preserves convergence, and (iv) an analytic near-tie spacing law $E(m) \approx r_k/(3k)$ that matches measurements to 1\%. Three compensation levels are implemented in C++ (L1: frozen set with live geometry; L2: margin-gated local repair; L3: damped, projected basis) and validated under a matched-conditions protocol (identical seed, MSA depth and trajectory family, only the topology mode differs). L2 reproduces the PB edge set exactly ($D = 0.0000$ at every boundary), L1 stays within 0.6% and L3 interpolates monotonically: L2 reproduces the PB edge set exactly when the trust region is removed, and L3 interpolates between the frozen and the PB topology. The recorded trajectory itself yields a measured $1.5\times$ Pass-1 and $1.76\times$ epoch speed-up with no measurable loss gap. An open-source implementation of the runtime and of all the experiments reported here is available at https://github.com/AlbertHuang96/ProteinPredictMachineLearning.

---

### FinGREAT: Evaluating LLMs on Financial Graph-guided Retrieval And Typing

Authors: Ashman Mehra, Bryan Routledge, Jane JaeYeon Pyo, Pierre Jinghong Liang
Published: 2026-10-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.19240v1

Financial disclosures express the underlying relationships of double-entry accounting through unstructured text. While financial LLM benchmarks predominantly emphasize question answering and classification, schema-guided evidence discovery remains relatively under-explored. We investigate whether open-weight LLMs can extract and classify evidence corresponding to predefined double-entry accounting relationships underlying the text. Using a financial accounting graph comprising 11 nodes (accounts) and 11 edges (transaction types), we evaluate mid-sized models from 5 open-weight families across 1,300 SEC 10-K Item 7 filings spanning 26 years, comparing their extraction performance against three proprietary models. Results reveal substantial differences in extraction volume, edge distributions, and cross-model agreement. Building on this study, we introduce a human-annotated relation classification benchmark comprising 483 passages across three financial accounting categories, evaluating 34 open-weight and 10 proprietary models. The strongest open-weight and proprietary models achieve 84.1\% and 92.8\% accuracy, respectively, with considerable variation across accounting categories. Our findings highlight the importance of evaluating evidence discovery and relation understanding separately, while investigating locally deployable models for scalable, privacy-sensitive financial analysis.

---

### LLMs program using Reusable and Composable Abstract Representations

Authors: Casper L. Christensen, Kyle J. Ray, Dmitry Manning-Coe, Adam Shai, Paul Riechers
Published: 2026-10-13
Categories: cs.LG, cs.PL
arXiv: https://arxiv.org/abs/2610.19238v1

When writing code, the same computation can be expressed in different programming languages and through different implementation strategies. We show that the activations of LLMs capture this shared structure in reusable and composable abstract representations when they generate code. These representations emerge in the middle layers, are reusable across programming languages, and afford the model an impressive amount of generalization ability when generating code under various interventions. Using activation steering, we show that function, programming language, and implementation strategy can be independently manipulated. The independent interventions on language and function can be flexibly composed to control code generation, even in cases when the prompt contains no information about language and function. We additionally find that finetuning on examples of arbitrarily defined functions in one programming language transfers to other languages, consistent with the models reusing abstract features of code generation that are shared across different examples. Finally, to study how such abstract representations come to be, we implement a minimal model of shared, reusable, and composable structure in a synthetic setting, and show that transformers trained on such data learn representations that reuse these parts flexibly across contexts.

---

### From Detection To Anticipation: Continuous Onset Forecasting For Stuttered Speech

Authors: Tampu Ravi Kumar, Ayoub Daliri, Vivek Gupta
Published: 2026-10-13
Categories: cs.SD, cs.LG
arXiv: https://arxiv.org/abs/2610.19235v1

Automatic stuttering systems primarily detect dysfluencies after they have occurred, limiting their use for anticipatory intervention. We address this gap by formulating stuttering prediction as continuous causal time-to-event forecasting, where a model estimates the risk of an upcoming stuttering onset every 100 ms using only preceding speech. To support this task, we augment FluencyBank Timestamped for 36 speakers with manually reviewed repetitions, prolongations, and blocks, yielding 2,872 forecasting events. We then model the temporal evolution of frozen Whisper-large-v3 speech representations using a compact causal Conformer with 3.67M trainable parameters. The proposed approach achieves an AUROC of 0.63. Temporal analysis further shows that explicitly modeling preceding speech states outperforms simply increasing the static acoustic window, while a discrete-time survival formulation provides coherent forecasts from 300 ms to 1.5 s using a single model. These results provide initial evidence that stuttering onset is preceded by measurable temporal information in speech and establish a framework for future anticipatory stuttering systems.

---

### Write Once, Plan Many: Building Reusable Visual Planning Systems from One Scene

Authors: Ruixiao Yang, Yilun Hao, Mingxin Yu, Yang Zhang, Chuchu Fan
Published: 2026-10-13
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.19234v1

Vision-language models (VLMs) can reason about visual planning tasks, but querying them for every instance incurs repeated inference cost and latency. Generating reusable planning programs offers an alternative, but constructing reliable planning systems for new problem domains without extensive supervision remains challenging. We present a training-free framework that constructs reusable perception and planning programs from a domain specification and a single practice scene without a paired solution. The central challenge is obtaining useful feedback from the limited input. Our framework addresses it by leveraging VLM's ability to provide a renderer and a dynamic model based on its scene understanding. Rendering detected states enables comparison with the practice scene, while checking simulated action outcomes against the domain specification helps expose errors in rule implementation. Together, these checks guide refinement of the grounding program and planner without requiring solution labels. After refinement and component selection, the programs are frozen and solve new instances without further VLM calls. The same framework constructs systems across domains without manual pipeline redesign. Across fifteen domain variants spanning 2D and 3D scenes with discrete and continuous state, our systems achieve 99.3% average planning success, compared with 80.9% for the strongest domain-level program-generation baseline and 90.7% for the strongest instance-level baseline. Cost analyses show a favorable success-cost trade-off, and ablations confirm the contributions of the executable domain model, execution feedback, and component selection.

---

### Spectral-MoE: Heterogeneous Per-Layer Expert Utilization in Dense-to-MoE Conversion

Authors: Zachary Broveak, Christian Kubicka, Thierry Tambe
Published: 2026-10-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.19232v1

Large language models impose substantial memory demands and arithmetic costs during inference. Dense-to-mixture-of-experts (MoE) conversion is a model optimization technique that aims to reduce these arithmetic costs. Dense-to-MoE conversion partitions pretrained feed-forward network (FFN) neurons into experts and selects a subset for each token. Leading methods fall into two categories: analytic methods produce a converted model quickly using heuristics, while learning methods produce a model much more slowly by optimizing via gradients. Both approaches, however, fix the same active expert count, or "utilization," at every layer. We introduce Spectral-MoE, which learns to allocate a utilization budget heterogeneously across layers. Its name draws a metaphor between separating a dense FFN into experts and separating a light beam into its spectrum. Spectral-MoE consists of two distinct approaches: a static approach that uses a single per-layer scalar which is shared across tokens, and a dynamic approach which emits a per-token schedule. We demonstrate that both allocators improve performance over uniform on both the analytic-based and learning-based SOTA dense-to-MoE methods. At 25% FFN utilization, dynamic allocation reduces WikiText-2 perplexity relative to uniform by 1.04, 2.96, and 2.71 points on the learned base for Llama-2-7B, Llama-3-8B, and Qwen2.5-7B, respectively. The corresponding reductions on the analytic base are 2.45, 6.26, and 2.17 points. On the learned base, dynamic allocation improves six-task mean accuracy over uniform by an average of 1.5, 4.2, and 3.6 percentage points across the three models at FFN utilizations of 12.5%, 25%, and 50%, respectively. These results establish per-layer expert utilization as a novel optimization axis in dense-to-MoE conversion.

---

### Do Syntax Heads Influence Model Behavior?

Authors: William Timkey, Tal Linzen
Published: 2026-10-13
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2610.19231v1

Transformer language models develop attention heads that specialize in syntactic relations, for example, attending from verbs to their subjects. Do these heads play a causal role in the model's syntactic predictions? Evidence for such a role is mixed to negative: even when we force these heads to attend to a distractor noun other than the subject, the model's syntactic predictions are still more consistent with the subject than with the distractor. Interventions on specialized syntax heads had a significantly stronger effect on syntactic predictions than interventions on random heads, but only in one of the two measures we propose.

---

### Agentic Self-Awareness: Aligning Knowing, Acting, and Holding

Authors: Qinglin Chen, Hiroaki Hayashi, Jiaxin Zhang, Chien-Sheng Wu
Published: 2026-10-13
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.19226v1

Frontier AI agents increasingly act in environments where seemingly solvable tasks may lack prerequisites, contain conflicting requirements, or reveal blockers only during execution. We formulate agentic self-awareness (ASA) as the alignment of task-grounded feasibility judgments (know), warranted behavior (act), and responses to reconsideration (hold), distinguishing what a task permits from what an agent can solve. We build paired feasible and blocked tasks in Terminal-Bench 2.1 and retail $τ^2$-bench and introduce an authorship-aware M0-M3 rubric that separates recognizing a blocker from taking the certified deferral action for the correct reason. Separate probes assess stated feasibility beliefs and responses to a single challenge. In a Terminal-Bench protocol baseline, blocker recognition reaches 84.4%, but strict self-awareness (SA) reaches only 40.6%. Harness search without model-weight updates increases strict SA from 45.3% to 93.8% on reused Terminal-Bench search tasks; selection for abstention alone, however, can reduce success on feasible tasks. Independently initialized joint-objective searches improve both blocked-task SA and feasible-task solve on their own search sets. Retail held-out gains remain limited, and Terminal-Bench blocked-task transfer is not measured. This evaluation pipeline makes the recognition-action gap measurable and sets a two-sided target for improving agent decisions under changing task evidence.

---

### Definition Grounding in Theory-Level Auto-formalization

Authors: Leyan Pan, Christian Szegedy, Frederick Rajasekaran, Robi Bhattacharjee, Gregory Constantine, Wenke Lee
Published: 2026-10-13
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.19223v1

LLM-based auto-formalization systems can now generate formally verified Lean code at unprecedented rates. However, a formalization advances mathematical research only when further mathematical results can be built on it, especially when its definitions and helper lemmas are general enough to be imported and used idiomatically. We propose definition grounding, where definition formalizations are continuously judged and revised, as a method to ensure formalization quality. We find that agents formalizing definitions from textbook text often produce definitions that later statements cannot use, and only 25%-50% of their definition sets are compatible with all later statements. However, clarifying the informal definitions can improve this rate up to 77.5% averaged over all models. Furthermore, stating a small set of basic properties without proving them can improve model judges' capability to detect definitions with compatibility issues, at only about 11% of the definition formalization cost. These results are verified with ReuseBench, a benchmark we curated for formalizing new mathematical fields and theories.

---

### REPORT-OPS: Hybrid Evidence-Guided Reporting with Human-Supervised Refinement for HPC

Authors: Mah Kadidia Konate, Mayur Akewar, Janki Bhimani, Steven Farrell
Published: 2026-10-13
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.19222v1

Operational reporting for high-performance computing (HPC) facilities requires combining information from monitoring services, schedulers, incident and maintenance records, storage measurements, periodic reports, and operator investigations. Producing a reliable report requires more than retrieving relevant records: the evidence must refer to the correct system, metric definition, time period, and source of authority. We present REPORT-OPS, a hybrid reporting agent in which an LLM interprets requests, plans evidence collection, evaluates possible explanations, and drafts reports, while author-seeded operational policies and automated checks determine whether the available evidence supports the requested conclusion. Depending on the evidence state, the system answers the request, reports a qualified conclusion, or withholds an unsupported claim. During offline refinement, an LLM critic proposes bounded policy changes that a human approves or rejects without editing; accepted candidates are promoted only after automated target, held-out, and regression testing. We evaluate REPORT-OPS on 68 sanitized reporting queries constructed from 50 base scenarios in a frozen HPC operational corpus, using three agent models and two LLM judges. REPORT-OPS achieves 68.63% Pass@80, compared with 7.02% for the strongest retrieval baseline. Evidence gating increases claim-support precision from 40.58% to 90.06%. On the 13-query explanatory subset, explanation accuracy reaches 55.54%. Across five rounds of human-supervised refinement, localized policy updates correct 91.17% of targeted failures and improve performance on related held-out cases by 9.33 percentage points, with no observed regression breaks under release gating. These results demonstrate improved evidence-grounded reporting and within-environment policy reuse; they do not establish unrestricted causal discovery or cross-facility generalization.

---

### Riding the Fixed-Point Curve: Image Classification in the Continuous Thought Machine

Authors: Thomas Hannagan
Published: 2026-10-13
Categories: cs.LG, cs.CV, cs.NE
arXiv: https://arxiv.org/abs/2610.19220v1

The Continuous Thought Machine (CTM) is a recurrent network that attends to inputs and makes predictions based on the synchronization between its neurons. On tasks spanning many domains, it displays emergent thought processes that have not been explained. We report a number of findings about how the CTM classifies images, focusing on ImageNet. Each thinking step is an attracting but strongly non-normal map exhibiting balanced amplification (Murphy and Miller, 2009); while attraction is already present at CTM initialization, training highly promotes non-normality. By Banach's theorem, the neural state has a locally unique attracting fixed point, which moves at every step, and the thought trajectory is the curve these fixed points trace, followed one tick behind. Most of that curve's motion is shared across images, and visible as a previously reported traveling wave of activity across neurons. The attention schedule along the thought is largely generic: shuffling queries across images at every tick costs only 7 percent of classification accuracy, while image-independent increments predict the next query with R^2 = 0.98 (held out). Our results suggest that the CTM has found a single strategy to process all ImageNet images.

---