# Research Papers — 2026-09 · bundle page 139 of 266 > Full-text markdown of papers #6901–6950 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/139 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/138.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/140.md) ### Refinement Is Inherently Editable: Training-Free Prompt-to-Prompt Image Editing with Generative Refinement Network Authors: Yulong Chen, Ziqian Zhang, Haoyu Zhang, Ao He, Senmao Li, Kai Wang Published: 2026-09-17 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.20633v1 Text-guided image editing must introduce the requested changes while preserving unrelated source content. Diffusion-based editors rely on spatial controls whose inaccuracies can leave edits incomplete or alter unrelated regions. Causal autoregressive editors face a further constraint: their fixed decoding order limits revision of earlier decisions. We introduce RefineEdit, a training-free prompt-to-prompt image editing framework built on a Generative Refinement Network. Our key idea is to couple edit localization with content generation through the global refinement of binary image codes, allowing editing evidence to be reassessed as the image evolves. RefineEdit initializes an editing branch from an intermediate source state, reusing the emerging layout. We compare the probabilities assigned by the two branches to the same source-sampled bits, using their signed differences to select editable positions and bits. Selected bits follow editing refinement, while the remaining bits copy the evolving source state. To stabilize editing across refinement steps, adaptive spatial freezing limits unnecessary mask expansion, while finite bit locking keeps recently selected bits editable. The framework requires no additional training, external masks, or attention control. Across nine editing categories of PIE-Bench, RefineEdit achieves the best background-preservation scores in PSNR, LPIPS, MSE and SSIM, together with the highest whole-image and edited-region CLIP scores among the evaluated methods. --- ### UniPolicy: Unified Objective-Specific Policies for Generative Search Advertising Authors: Kun Yao, Yuhang Zhou, Yichi Zhang, Zeliang Tong, Shengri Xue, Haitao Wang, Siyu Lu, Qianlong Xie, Xingxing Wang Published: 2026-09-17 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.20630v2 Search advertising connects user intent with commercial content and plays a critical role in platform monetization. Recent systems typically align pretrained generative models with a single business reward, such as eCPM, or use naive reward fusion for preliminary multi-objective alignment. However, an ideal search advertising system must jointly account for heterogeneous objectives, including relevance, click propensity, and commercial value, to balance user experience and business value while mitigating globally suboptimal performance caused by gradient competition. We propose UniPolicy, an objective-aware multi-policy alignment framework. UniPolicy combines objective-specific prefix tokens, sparse MoE-LoRA routing, and objective-specific residual FFNs to hierarchically decouple parameters within a shared backbone, providing differentiated parameter and policy-expression spaces for different business objectives. It further constructs pairwise preferences from multi-stage behavioral feedback, supplementing the relative preference information in exposed-but-unclicked samples and strengthening the relative advantage of clicked candidates in the generation distribution. At inference, UniPolicy supports parallel, business-customizable multi-policy beam search, flexibly allocating candidate quotas across objectives under a fixed retrieval budget. Large-scale offline experiments show that UniPolicy delivers balanced improvements across multiple metrics while preserving retrieval quality, outperforming single-objective reinforcement learning and naive reward-fusion baselines. In a 7-day online A/B test on a real search advertising system, UniPolicy improves CTR by 0.71%, RPS by 1.58%, and advertising revenue by 1.32%, while maintaining stable serving latency. --- ### UniPolicy: Unified Objective-Specific Policies for Generative Search Advertising Authors: Kun Yao, Yuhang Zhou, Yichi Zhang, Zeliang Tong, Shengri Xue, Haitao Wang, Siyu Lu, Qianlong Xie, Xingxing Wang Published: 2026-09-17 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.20630v1 Search advertising connects user intent with commercial content and plays a critical role in platform monetization. Recent systems typically align pretrained generative models with a single business reward, such as eCPM, or use naive reward fusion for preliminary multi-objective alignment. However, an ideal search advertising system must jointly account for heterogeneous objectives, including relevance, click propensity, and commercial value, to balance user experience and business value while mitigating globally suboptimal performance caused by gradient competition. We propose UniPolicy, an objective-aware multi-policy alignment framework. UniPolicy combines objective-specific prefix tokens, sparse MoE-LoRA routing, and objective-specific residual FFNs to hierarchically decouple parameters within a shared backbone, providing differentiated parameter and policy-expression spaces for different business objectives. It further constructs pairwise preferences from multi-stage behavioral feedback, supplementing the relative preference information in exposed-but-unclicked samples and strengthening the relative advantage of clicked candidates in the generation distribution. At inference, UniPolicy supports parallel, business-customizable multi-policy beam search, flexibly allocating candidate quotas across objectives under a fixed retrieval budget. Large-scale offline experiments show that UniPolicy delivers balanced improvements across multiple metrics while preserving retrieval quality, outperforming single-objective reinforcement learning and naive reward-fusion baselines. In a 7-day online A/B test on a real search advertising system, UniPolicy improves CTR by 0.71%, RPS by 1.58%, and advertising revenue by 1.32%, while maintaining stable serving latency. --- ### Chronicle: Cut-Point Replay for Regression Testing of LLM Agents Authors: Tisha Chawla, Susheem Koul Published: 2026-09-17 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.20625v1 Large language model responses are non-deterministic, so failures in LLM agents are hard to reproduce: a failure depends on inference that is not bitwise reproducible, on tools that read changing state, and on a multi-step trajectory that a re-run rarely repeats. Record-and-replay makes a run reproducible, but existing agent tooling records runs only to trace or score them, not to test a code change against them. We present Chronicle, which records an agent run at its non-deterministic boundaries as immutable envelopes and replays it from the record. Its central operation, cut-point replay, serves a chosen subset of boundaries from the record and executes the complementary subset live with new code, turning a recorded incident into a regression test that runs in continuous integration. On a benchmark of 6 recorded failures with simulated model boundaries, recording adds 23 μs per crossing (0.008% of an assumed 300 ms model call), full replay issues zero model calls and is bit-stable across 20 repetitions, and cut-point tests fail on faulty code and pass on guarded and benign changes for all 6 incidents. In a mutation study of the guarded tools, cut-point tests catch every mutant that lets the recorded unsafe action through, while a baseline that stubs every boundary, using the same assertion, catches none. Chronicle and the benchmark are publicly available at https://github.com/theagentplane/chronicle. --- ### PhGS: Post-Hoc Pruning and Refinement of Single-View Feed-Forward 3D Gaussian Reconstructions Authors: Rinto Yagawa, Han Cheng, Dieter Schmalstieg, Hideo Saito, Shohei Mori Published: 2026-09-17 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.20623v1 Recent single-view feed-forward 3D Gaussian Splatting (3DGS) generation predicts a fixed number of Gaussians per camera ray, introducing severe spatial redundancy. Most existing compaction strategies target multi-view setups to exploit cross-view consistency and are incompatible with single-image models. Instead of retraining the base feed-forward network to directly output compact representations, our insight is to keep the base models frozen and apply post-hoc pruning and recurrent refinement to the generated Gaussians. Consequently, we propose a backbone-agnostic compaction pipeline for single-view feed-forward 3DGS that couples an importance-score-based pruning mechanism with a trainable, lightweight recurrent refinement module, which iteratively updates the surviving primitives to restore image quality. Our results demonstrate seamless integration with existing baselines while preserving novel-view rendering fidelity and achieving high memory reduction. Furthermore, our method supports flexible inference-time keep ratios for application needs. --- ### A Simulation Platform for AUV Fault Recovery: Exploring LLM-Based Diagnostic Strategies Authors: Khalid Halba, Kylie Cooper, James G. Bellingham Published: 2026-09-17 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.20620v1 Autonomous underwater vehicles (AUVs) operating beyond reliable communications must recover from failures without human intervention. We investigate an architecture in which conventional deterministic layered control autonomy manages normal operations, while an invokable large language model (LLM) serves as a diagnostic and recovery planner when onboard anomaly detection identifies performance outside expected limits. Because language models are stochastic, rigorous evaluation requires ensemble testing rather than individual demonstrations. We present a closed-loop simulation architecture that couples real-time C vehicle software with a higher-level orchestration layer for physics-based fault injection, structured prompting, language-model interaction, mission file generation, validation, execution, and LLM-judge scoring. The framework, which we call SPAR (Simulation Platform for AUV Recovery), supports evaluation across fault realizations, prompt structures, reasoning models, and mission conditions. We vary these for a mass-shift fault over 480 SPAR trials, evaluating a frontier model and three off-the-shelf locally deployable LLMs. Model choice dominates diagnosis: the frontier model places the CG-shift mechanism in its top three hypotheses in 85-90% of trials, versus 60-78% for the best local model. Reasoning analysis indicates that local-model success is associated with following the complete diagnostic procedure, whereas weaker models often commit prematurely to elevator failure even though the actuator tracks its command. Diagnosis and operational decision performance do not appear to be coupled in this dataset. The contributions are an architecture extending unanticipated-fault recovery from detection to mitigation and an ensemble methodology for evaluating LLM-assisted mission management on low-power AUVs. --- ### INSPECT: Learning Robot View Selection from Assistant Use Authors: Di Wen, Kailun Yang, Wenhao Guo, Yitian Shi, Junwei Zheng, Yufan Chen, Ruiping Liu, Jiale Wei, Rania Rayyes, Kunyu Peng Published: 2026-09-17 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2609.20615v1 Robots inspecting an assembly must determine which parts are present and whether they are correctly installed. During egocentric assembly assistance, head motion and workpiece handling reveal evidence for these checks, while spoken state confirmations link observations to procedural outcomes. We introduce INSPECT, which learns robot view preferences from records of a smart-glasses assistant that answers part queries and provides next-step guidance. Presence-Invariant TwinSwap (PI-TwinSwap) calibrates object evidence through paired identity interventions. Claim-indexed supervision separates evidence requirements from camera-reproducible observation changes. Object-centered calibration adapts relative view preferences to robot poses, while clause-level screening checks predicted evidence. The robot selects views using only its current observation and known poses, without candidate images. Evaluation uses annotated assistant-video replay to simulate state feedback, without target-domain view labels for policy training. On images of physical gearbox assemblies, INSPECT achieves the highest view utility among the compared non-oracle policies and raises human-rated full verifiability from 34.8% to 41.7% compared with keeping the current view. On commercial angle-grinder recordings in IMPACT, the transferred relative-view selector increases the correct decision rate from 50.6% to 54.3% with a frozen perception head. The source code is available at https://github.com/Kratos-Wen/INSPECT. --- ### Inference-Engine Fingerprinting Attacks are Practical: Exploring Model-Driven Environmental Discovery, Exploitation, and Escape Authors: Sarah Radway, Andrew Cheng, Vijay Janapa Reddi, James Mickens Published: 2026-09-17 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.20614v1 Frontier AI models are rapidly gaining the ability to exploit vulnerabilities in complex pieces of software. The risk is not theoretical, as evidenced by recent sandbox escapes performed by frontier models at OpenAI and Anthropic. Discussions of how to sandbox inference stack components often focus on components other than the inference engine itself (e.g., network proxies or code execution environments). However, the inference engine is an attractive target for a misaligned model. For example, if a model can trigger exploits in that engine merely by generating specially-crafted output tokens, the model can initiate a multi-step, to-the-bare-metal exploit chain in the engine, without relying on vulnerabilities in other components of the inference stack, and without assistance from externally-provided, maliciously-crafted input tokens. In this paper, we show that a misaligned model can perform inference engine fingerprinting to determine the specific engine (e.g., vLLM, SGLang) which executes the model. Once the engine has been fingerprinted, the model can leverage engine-specific exploits to take control of the engine using only carefully-selected output tokens. We provide concrete examples of model fingerprints in five popular engines, and demonstrate how realistic agentic harnesses allow a model to leverage those fingerprints to identify the local engine. We also describe a proof-of-concept, to-the-bare-metal exploit chain that originates from a fingerprinted (and subsequently compromised) inference engine. We conclude by discussing several ways that inference engines could be changed to make fingerprinting attacks more difficult. --- ### What Does Privileged Information Add to On-Policy Self-Distillation? Authors: XiuYu Zhang, Wei Chow, Junfeng Fang, Xingyu Zhu, Zhenkai Liang, Tat-Seng Chua Published: 2026-09-17 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.20612v2 On-policy self-distillation (OPSD) lets a language model learn from a frozen copy of itself that sees an answer or a worked solution. Giving the teacher this extra information seems to offer the student more to learn, but how much does it add beyond distillation itself? To isolate that contribution, we construct AMPLE-Math, a reusable suite of 5,319 mathematical problems with six reasoning views that share the same answer, and compare each view with matched reference-free distillation. With a thinking-enabled teacher supervising direct-response rollouts, reference-free distillation accounts for much of Qwen3-1.7B's improvement under thinking-enabled evaluation, both in domain and on external benchmarks. Evidence for an additional reference benefit is modest in Qwen, strongest for a polished solution, whereas complete traces add two percentage points in SmolLM3-3B at step 50. These benefits depend on the student being trained. At the same checkpoint, replacing short direct-response rollouts with long thinking-enabled rollouts turns gains into losses in both families while the problems, references, and evaluation stay fixed. Teacher profiles and matched loss interventions in Qwen further show that changing token-level supervision can leave student behavior largely unchanged. Together, these findings suggest that OPSD can improve access to existing reasoning capabilities through parameters shared by direct-response and thinking-enabled inference. The value of a privileged reference is what it adds to this cross-mode transfer, not how much of the solution it reveals. --- ### What Does Privileged Information Add to On-Policy Self-Distillation? Authors: XiuYu Zhang, Wei Chow, Junfeng Fang, Zhenkai Liang, Tat-Seng Chua Published: 2026-09-17 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.20612v1 On-policy self-distillation (OPSD) lets a language model learn from a frozen copy of itself that sees an answer or a worked solution. Giving the teacher this extra information seems to offer the student more to learn, but how much does it add beyond distillation itself? To isolate that contribution, we construct AMPLE-Math, a reusable suite of 5,319 mathematical problems with six reasoning views that share the same answer, and compare each view with matched reference-free distillation. With a thinking-enabled teacher supervising direct-response rollouts, reference-free distillation accounts for much of Qwen3-1.7B's improvement under thinking-enabled evaluation, both in domain and on external benchmarks. Evidence for an additional reference benefit is modest in Qwen, strongest for a polished solution, whereas complete traces add two percentage points in SmolLM3-3B at step 50. These benefits depend on the student being trained. At the same checkpoint, replacing short direct-response rollouts with long thinking-enabled rollouts turns gains into losses in both families while the problems, references, and evaluation stay fixed. Teacher profiles and matched loss interventions in Qwen further show that changing token-level supervision can leave student behavior largely unchanged. Together, these findings suggest that OPSD can improve access to existing reasoning capabilities through parameters shared by direct-response and thinking-enabled inference. The value of a privileged reference is what it adds to this cross-mode transfer, not how much of the solution it reveals. --- ### COIN-GP: Cooperative Online Learning in Networked Distributed Systems with Partial Measurements via Gaussian Process Regression Authors: Zewen Yang, Xiaobing Dai, Zhenxiao Yin, Hang Zhao, Zhijun Li, C. C. Chan Published: 2026-09-17 Categories: cs.LG, cs.MA, eess.SY arXiv: https://arxiv.org/abs/2609.20598v1 In this paper, we tackle the problem of jointly estimating the system states and partially unknown dynamics within distributed sensor-equipped networks, particularly in scenarios where only partial state observations are available. To address this issue, we propose an observer-based dynamic cooperative learning framework incorporating online distributed Gaussian Process (GP) regression, which enables accurate estimation despite incomplete in measurements and deficient GP models. In addition, a novel data collection strategy is introduced, with theoretical conditions ensuring feasible data acquisition. Moreover, we also derive an error upper bound encompassing state estimation and model estimation, leveraging the deterministic error bounds of GPs. Empirical simulations demonstrate the superiority of our approach compared to existing distributed GP-based methods. --- ### Recursive Quantum Long Short-Term Memory for Stable Short-Horizon Temperature Forecasting Authors: Mu-En Lee, Yen-Ku Liu, Samuel Yen-Chi Chen, Yun-Cheng Tsai Published: 2026-09-17 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.20594v1 Quantum long short-term memory (QLSTM) models extend recurrent sequence learning with variational quantum circuits, but their optimization behavior can vary substantially across random initializations and temporal contexts. This paper evaluates a recursive QLSTM architecture against a standard QLSTM for one-step-ahead prediction of daily minimum and maximum temperature. Using daily weather observations from Toronto and identical training settings, we compare convergence, predictive accuracy, and generalization across input windows of 8, 16, and 32 days over 20 random seeds. The recursive model consistently reaches a near-optimal test loss earlier, reduces mean absolute error and root mean squared error, and exhibits a smaller generalization gap. These results indicate that recursive quantum feature transformations can improve stability and out-of-sample performance for compact hybrid quantum--classical temporal models. --- ### WiC is Not WSD: A Study on LLMs and Lexical Ambiguity Resolution Authors: Yi Zhou, Kiamehr Rezaee, Danushka Bollegala, Mohammad Taher Pilehvar, Jose Camacho-Collados Published: 2026-09-17 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.20593v1 Word-in-Context (WiC) remains challenging for language models, despite recent progress on lexical-semantic tasks. We hypothesise that this difficulty arises not only from comparing two contextual uses of a word, but also from the absence of an explicit sense inventory that specifies the relevant level of semantic granularity. We evaluate open LLMs on WiC and traditional Word Sense Disambiguation (WSD) under similar settings. We find that providing candidate senses, similar to what is done in traditional WSD, improves WiC performance in all settings. In general, explicit sense information helps models make more consistent and targeted judgements. Human evaluation further shows that many apparent WiC errors reflect label ambiguity or mismatches between model and annotator sense boundaries rather than simple failures of lexical understanding. In particular, results show that LLMs overthink the sense distinction often leading to errors based on overly fine-grained distinctions. --- ### CrystalMO-TuRBO: Multi-Objective Trust-Region Bayesian Optimization for High-precision Joint Crystal Structure Refinement Authors: Joseph Agada, Yishu Wang, Arpan Biswas Published: 2026-09-17 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.20592v1 Crystal structure refinement is a fundamental inverse problem in materials characterization, where structural parameters are optimized to reproduce experimental diffraction data. Conventional approaches, such as least-squares and likelihood-based optimization, rely on local search and often struggle with non-convex, noisy, and highly correlated parameter landscapes, particularly when integrating multiple diffraction modalities. Joint refinement of X-ray and neutron data is especially challenging due to their complementary but competing sensitivities, which are typically combined through scalarized objectives requiring manual weighting and leading to suboptimal solutions. We propose CrystalMO-TuRBO, a multi-objective trust region Bayesian optimization architecture for joint crystal structure refinement. The method models X-ray and neutron discrepancies as separate objectives and transforms the problem into a normalized maximization setting. A two-phase optimization strategy is introduced: Phase 1 performs global exploration using parallel trust-region Bayesian optimization across multiple scalarizations to identify promising regions of the parameter space, while Phase 2 conducts localized refinement within a shrinking region to achieve high-precision solutions. This design explicitly separates global search from fine-grained optimization, addressing the unique accuracy requirements of refinement tasks. We evaluate the proposed method on experimentally collected X-ray and neutron diffraction data from single-crystal Ho2Ti2O7. Results demonstrate improved convergence, robustness, and parameter precision compared to classical refinement methods and Bayesian optimization baselines on refinement of a single-crystal pyrochlore material system. --- ### RawSLAM: Online HDR Gaussian SLAM from Linear Radiance Authors: Marina Orozco González, Luis Merino Published: 2026-09-17 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.20589v1 Current dense visual SLAM systems rely almost exclusively on 8-bit tonemapped Low Dynamic Range (LDR) inputs, limiting their robustness in extreme lighting where shadows and highlights trigger tracking drift and mapping collapse. Conversely, existing raw and High Dynamic Range (HDR) reconstruction pipelines operate strictly offline. They depend on Structure-from-Motion preprocessing and are not suited for large inter-frame motion. We present, to the best of our knowledge, the first online Gaussian SLAM framework that tracks and maps directly on single-exposure 16-bit linear HDR imagery. Our method rests on three core components: an architecture-agnostic HDR Gaussian Splatting module featuring an MLP-free logarithmic parameterization of Gaussian color features; a Reinhard range-compressed photometric objective; and structure-guided spatial gradient weighting. Combined, these components allow our approach to outperform a direct HDR adaptation of MonoGS in both trajectory and reconstruction accuracy, while rendering natively in linear scene radiance for post-rendering processing. The same formulation runs unchanged on standard 8-bit inputs, roughly halving the MonoGS baseline error. Furthermore, our HDR Gaussian module transfers seamlessly to SplaTAM, Gaussian SLAM, and DROID-W, eliminating all tracking failures these systems suffer on challenging illumination sequences. To enable this research, we introduce RawSLAM: a dataset of 10 real-world indoor sequences featuring 16-bit RAW imagery, aligned depth, IMU measurements, and external OptiTrack poses. Code and dataset will be made publicly available soon. --- ### CoRef-GS: Cooperative Referring Gaussian Splatting for Multi-Agent Scene Understanding Authors: Zhikun Zhou, Kunyu Peng, Runyi Yang, Junhao Cai, Di Wen, Ruiping Liu, Danda Pani Paudel, Yi Zhou, Luc Van Gool, Kailun Yang Published: 2026-09-17 Categories: cs.RO, cs.CV, eess.IV arXiv: https://arxiv.org/abs/2609.20586v1 Referring scene understanding for embodied robots requires grounding object- and relation-centric language queries from a designated viewpoint. While a local semantic Gaussian map can support such grounding within one agent's observations, cooperative settings require this ability to remain effective after independently reconstructed maps are aligned and fused. In this setting, the referred target or its contextual landmark may come from another agent's observations, while spatial relations must still be interpreted from the querying robot's viewpoint. We formulate this problem as cooperative referring Gaussian grounding over fused maps, which requires geometric alignability, instance-level semantic comparability, and view-conditioned relation reasoning. Existing language-aware Gaussian methods mainly focus on single-map querying, whereas Gaussian registration methods optimize geometric or photometric alignment without preserving language-grounding-oriented semantic compatibility. We propose CoRef-GS, a cooperative referring Gaussian splatting framework. CoRef-GS constructs local open-vocabulary instance-aware Gaussian maps, then aligns partially overlapping maps with a cross-agent alignment module by geometric and semantic consistency, and grounds queries using a view-conditioned mask relation graph. We further introduce CoQuad-Ref, a dual-quadruped benchmark spanning both real-world and simulated indoor scenes. Experiments show that, on simulated scenes, CoRef-GS reduces the rotation error from 2.58° after coarse initialization to 0.15° after refinement, and improves real-world referring mIoU over ReferSplat from 52.6% to 68.8%. The established benchmark and source code will be publicly released at https://github.com/ruojiruoli17/CoRef-GS.git. --- ### SAFARI: An Industrial Benchmark for LLM-Assisted Hazard Analysis and Risk Assessment Authors: Chenxi Wu, Zimu Wang, Haiyang Zhang, Wei Wang, Zhijie Xu Published: 2026-09-17 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.20584v1 Large language models (LLMs) are increasingly considered for safety-critical engineering, yet their reliability in regulated functional-safety workflows remains underexplored. We introduce SAFARI (Safety-Aware Functional Automotive Risk Inference), the first industrial benchmark for LLM-assisted automotive Hazard Analysis and Risk Assessment (HARA) under ISO 26262. It contains 3,000 de-identified industrial HARA cases and evaluates two coupled tasks: open-ended hazard analysis and standards-grounded risk assessment. To evaluate open-ended HARA artifacts, we propose the first reference-anchored LLM-as-a-judge protocol with high expert correlation. Experiments with nine frontier LLMs show that models often produce plausible hazard narratives but remain weak at ISO 26262 risk classification, with the best ASIL macro-F1 reaching only 0.261. Chain-of-Thought prompting provides limited benefit and often degrades categorical risk assessment. Error analysis further localizes major failures to scenario-critical context omissions during hazard generation and to controllability misjudgments during risk assessment, indicating where expert oversight should be concentrated. The dataset can be obtained from https://github.com/xixi47520-hash/HARA. --- ### Limits of Confidence in Diffusion Authors: Russ Webb, Amitis Shidani, Alice Bizeul, Dan Busbridge Published: 2026-09-17 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.20581v2 Discrete diffusion, including remasking and uniform-state samplers, generate a sequence by writing multiple token positions per step, drawing each from a per-position distribution and choosing which positions to write from those same distributions. For domains of general interest (pixels, phonemes, or words) there are inherent dependencies between tokens. We show that a step matches the training distribution only when the positions it writes are conditionally independent given the tokens already fixed, that no product of per-position distributions can match a dependent group, and that per-position distributions do not determine whether a group is dependent: two joint distributions can have identical per-position marginals while differing in which combinations of values occur. On ScanAndAdd, a synthetic task whose joint distribution is available in closed form, we verify that every group of two or more undetermined positions a confidence ranking writes is dependent, and measure the generated distribution to be $29\times$ the sampling-noise floor total variation while per-sample metrics are $1.0$. --- ### Limits of Confidence in Diffusion Authors: Russ Webb, Amitis Shidani, Alice Bizeul, Dan Busbridge Published: 2026-09-17 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.20581v1 Discrete diffusion, including remasking and uniform-state samplers, generate a sequence by writing multiple token positions per step, drawing each from a per-position distribution and choosing which positions to write from those same distributions. For domains of general interest (pixels, phonemes, or words) there are inherent dependencies between tokens. We show that a step matches the training distribution only when the positions it writes are conditionally independent given the tokens already fixed, that no product of per-position distributions can match a dependent group, and that per-position distributions do not determine whether a group is dependent: two joint distributions can have identical per-position marginals while differing in which combinations of values occur. On ScanAndAdd, a synthetic task whose joint distribution is available in closed form, we verify that every group of two or more undetermined positions a confidence ranking writes is dependent, and measure the generated distribution to be $29\times$ the sampling-noise floor total variation while per-sample metrics are $1.0$. --- ### NS3Learn: Transferring 5G NR Mode-2 Reception Realism from ns-3 to the Veins/SUMO Stack for Connected-Vehicle Safety Assessment Authors: Rasheed Bello, Arthur Mukwaya, Gurcan Comert, Varghese Vaidyan, Vijay Bendigeri, Anthony Dontoh, Jagruti Sahoo, Judith Mwakalonge Published: 2026-09-17 Categories: cs.NI, cs.LG, cs.MA, eess.SY arXiv: https://arxiv.org/abs/2609.20578v1 Connected-vehicle safety evaluations rely on coupled traffic and network simulations, but standard channel models ignore radio resource competition in 5G NR sidelink Mode-2, reporting unrealistically high message delivery in dense traffic. This study introduces resource-competition losses without requiring full protocol reimplementation. We labeled 10.5 million reception outcomes from ns-3 5G-LENA traces (calibrated on 3GPP scenarios and driven by SUMO trajectories) to fit NS3Learn - a closed-form model capturing half-duplex loss, scheduling collisions, receiver capture, and decoding. Evaluation spanned two signalized urban networks, six penetration levels (1-100%), and five random seeds per condition. NS3Learn achieved a mean absolute deviation of 0.06 in per-instant delivery compared to ns-3 5G-LENA, outperforming alternative models (0.44 and 0.55 deviation). Fitted parameters transferred to a distinct intersection with only 20% additional error. Crucially, using realistic communication models reversed simulated traffic speed trends and more than doubled predicted hard-braking events. The framework transfers reception realism between simulators via model distillation instead of full reimplementation. Every stage maps directly to an explicit physical mechanism. Researchers and transportation agencies can maintain existing simulation pipelines while accurately accounting for dense-traffic packet loss and denial-of-service impacts. Adapting to new radio configurations requires only offline refitting rather than code modification. --- ### TAP Accuracy Below the Fluctuation Scale and Universal Posterior Geometry in Spherical Linear Models Authors: Jingbo Liu, Zhiyuan Yu Published: 2026-09-17 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2609.20577v1 We study the Bayes-optimal spherical linear model as the ambient dimension and sample size grow proportionally, under a quantitative Marchenko--Pastur spectral-regularity condition on the design. This condition is satisfied by normalized i.i.d. designs with standardized entries of finite fourth moment, but does not require entrywise independence or impose conditions on the singular vectors. Under this condition, we prove a quantitative all-temperature TAP approximation and characterize the posterior geometry. For the natural finite-aspect-ratio TAP functional, the normalized spherical free energy and the TAP optimum differ by $O_P(p^{-1})$. Each is within $O_P(p^{-1/2})$ of its explicit deterministic equivalent, and this fluctuation scale is sharp. Uniformly over all global TAP maximizers, the normalized squared Euclidean distance to the spherical posterior mean is $O_P(p^{-1})$. We also prove that the posterior mass outside a data-dependent band determined by the ridge estimator has sharp exponential order. More precisely, uniformly over sufficiently small band widths $\varepsilon$, the logarithm of this mass is at most $-cp\varepsilon^2+O_P(1)$. For every fixed geometrically admissible width, a spherical-cap construction gives a matching exponential-order lower bound on this mass. For every deterministic sequence of widths $\varepsilon_p\gg p^{-1/2}$, the corresponding bands capture asymptotically all posterior mass. --- ### Accelerating Visual Policy Learning with Sampling-Based Model Predictive Control Authors: Yilang Liu, Haoxiang You, Qian Wang, Daniel Rakita, Ian Abraham Published: 2026-09-17 Categories: cs.RO, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.20575v1 Learning visual policies for locomotion and manipulation requires coordinating contact with the environment and can incur substantial computation and GPU memory costs. First-order policy gradients (FoPG) reduce training cost through differentiable simulation, but local optimization can converge to unintended contact patterns. To address this shortfall, we propose Sampling-Guided Policy Search (SGPS), which couples recurring action-target refinement by sampling-based model-predictive control with first-order policy optimization. Behavior cloning initializes the policy from sampled actions; training then alternates sampling-based refinement with short-horizon FoPG updates under perturbed initial states and randomized dynamics. For visual policy training, we use a decoupled FoPG formulation that excludes rendering from the computation graph, enabling direct learning from depth observations without a state-policy teacher. On a single GPU, SGPS learns policies for locomotion, obstacle traversal, crate pushing, and bimanual carrying on simulated Unitree Go2 and G1 robots. Our experiments further show that refinement improves policy learning beyond initialization and tracking alone. For hardware deployment, the distilled policy transfers zero-shot to a real Go2 and uses onboard depth to autonomously trot, crawl, clear hurdles, and switch between these behaviors. --- ### DocAttriBench: Benchmarking Answer Grounding in Document Visual Question Answering Authors: Luca De Grandis, Silvia Cappelletti, William Raccagni, Marcella Cornia, Lorenzo Baraldi, Rita Cucchiara Published: 2026-09-17 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.20574v2 Answer grounding in document visual question answering remains an open challenge: most benchmarks lack grounding annotations or provide limited-quality labels, while constructing grounded datasets still requires costly manual effort. We introduce DocAttriBench (DAB), a large-scale benchmark for fine-grained, element-level source attribution in Document VQA, grounding answers to specific layout elements such as text blocks, tables, and images. To build DAB, we propose a Mask-based Perplexity-Derived Attribution method (MAPPET) that combines document layout and language modeling to identify the most informative element for each answer. MAPPET measures the increase in perplexity after masking candidate elements and attributes the answer to the element contributing most to model confidence. Applying MAPPET to multiple existing Document VQA datasets yields DAB, with 237k documents and 296k question-answer pairs with element-level grounding. We benchmark grounding-capable multimodal LLMs on DAB, evaluating answer accuracy, attribution accuracy, and overall answer quality. Results show that while larger models generally achieve higher answer accuracy, even the strongest models often fail to localize the supporting elements. DAB provides a scalable benchmark for developing grounded, verifiable, and trustworthy Document VQA models. Dataset and code are available at https://aimagelab.github.io/DocAttriBench/. --- ### DocAttriBench: Benchmarking Answer Grounding in Document Visual Question Answering Authors: Luca De Grandis, Silvia Cappelletti, William Raccagni, Marcella Cornia, Lorenzo Baraldi, Rita Cucchiara Published: 2026-09-17 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.20574v1 Answer grounding in document visual question answering remains an open challenge: most benchmarks lack grounding annotations or provide limited-quality labels, while constructing grounded datasets still requires costly manual effort. We introduce DocAttriBench (DAB), a large-scale benchmark for fine-grained, element-level source attribution in Document VQA, grounding answers to specific layout elements such as text blocks, tables, and images. To build DAB, we propose a Mask-based Perplexity-Derived Attribution method (MAPPET) that combines document layout and language modeling to identify the most informative element for each answer. MAPPET measures the increase in perplexity after masking candidate elements and attributes the answer to the element contributing most to model confidence. Applying MAPPET to multiple existing Document VQA datasets yields DAB, with 237k documents and 296k question-answer pairs with element-level grounding. We benchmark grounding-capable multimodal LLMs on DAB, evaluating answer accuracy, attribution accuracy, and overall answer quality. Results show that while larger models generally achieve higher answer accuracy, even the strongest models often fail to localize the supporting elements. DAB provides a scalable benchmark for developing grounded, verifiable, and trustworthy Document VQA models. Dataset and code are available at https://aimagelab.github.io/DocAttriBench/. --- ### OmniMimic: Dynamics-completed Motion Augmentation for Multi-style Omnidirectional Quadruped Locomotion Authors: Sheng Wu, Guoqiang Zhao, Zhe Yang, Fei Teng, Zhikun Zhou, Yanlin Yang, Zheng Fang, Hong Zheng, Yaonan Wang, Kailun Yang Published: 2026-09-17 Categories: cs.RO, cs.CV, eess.IV arXiv: https://arxiv.org/abs/2609.20566v1 Animal demonstrations provide quadruped robots with natural and distinctive gait styles that are difficult to specify through hand-crafted rewards. However, their narrow directional coverage leaves little style-consistent supervision for backward, lateral, and turning commands. We present OmniMimic, a training framework that turns directionally limited animal demonstrations into a single multi-gait policy over target per-axis velocity ranges. OmniMimic first combines temporal reversal, constrained dynamics completion, and sagittal reflection to construct robot-specific kinematic and physical supervision beyond the observed directions. It then expands commands progressively from the demonstrated velocity distribution toward the target per-axis bounds, and uses a shared actor with soft-gated, gait-specialized residual experts to balance reusable locomotion skills with gait-specific corrections. Across four gaits in simulation, OmniMimic reduces mean foot-position RMSE at forward and backward reference velocities by 12.9% and velocity-tracking RMSE on a uniform Cartesian command grid by 63.1%, compared with the matched APEX baseline. The project page is at https://OmniMimic.github.io. --- ### Steering the Compass: Aligning Dynamic Psychological Counseling Conversations with Cognitive Behavioral Therapy Strategies Authors: Zimu Wang, Yiwen Jiang, Xiangyu Zhao, Yaling Shen, Jiahe Liu, Stephanie Fong, Maxmartwell H Cheng, Guilherme C Oliveira, Anh Nguyen, Robert Desimone, Barnaby Nelson, Dominic Dwyer et al. Published: 2026-09-17 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.20565v1 Recent advancements in large language models have revolutionized the field of psychological counseling, especially in the context of Cognitive Behavioral Therapy (CBT). While the success of CBT relies heavily on dynamic decision-making informed by the client's real-time mental state, this aspect has often been overlooked in current research, limiting both flexibility and therapeutic outcomes. In this paper, we introduce StratCBT, a dataset specifically designed for psychological counseling conversations with CBT Strategies, consisting of 9,688 sessions and around 256K utterances, with each counselor's response aligned with one of eight distinct strategies. The creation of StratCBT involves modeling clients based on their negative thoughts and generating high-quality counseling conversations through self-chat, incorporating realistic sessions as guidance, thereby significantly surpassing existing datasets in both general counseling and CBT-specific skills. We conduct extensive experiments to demonstrate the effectiveness of strategy-aligned generation and evaluate its efficacy in delivering professional and effective counseling with LLM-simulated clients to reflect real-world scenarios. The dataset can be obtained from https://github.com/zimuwangnlp/StratCBT. --- ### A Dual-Stream Regulated Reconstruction and Segmentation Network with Hierarchical Artifact-Prior Modeling for Ultra-Low-Field Pediatric Neuroimaging Authors: Bahram Jafrasteh, Leo Milecki, Qingyu Zhao Published: 2026-09-17 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.20562v1 Automated quality assessment, enhancement, and segmentation of multiple structures in $0.064\,\mathrm{T}$ ultra-low-field pediatric MRI are limited by a low signal-to-noise ratio, weak anatomical boundaries, and frequent artifacts. We present a unified framework for the LISA 2026 Challenge that performs all three tasks together within one inference pipeline. A network with two coupled streams, built on a 3D U-Net, first reconstructs an enhanced uLF volume and then combines the original and enhanced images for subcortical segmentation. To improve boundary stability, we add an auxiliary class covering brain tissue outside the target structures, derived from whole brain masks. A head conditioned on an artifact graph predicts the seven artifact ratings from reconstruction residuals and frozen segmentation features. We address the scarcity of dense annotations using diffeomorphic registration from atlas to target for label propagation and to regularize anatomical reconstruction. We report validation results across all three tasks. --- ### Mitigating Retaliatory Algorithmic Collusion in Repeated Games Authors: Karthik Sivachandran, Rohan Paleja Published: 2026-09-17 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.20548v1 Reinforcement learning agents trained to maximize their own reward in repeated interactions can converge to supra-competitive outcomes resembling explicit collusion, without communication or shared design. Existing mitigation approaches are largely tied to specific economic settings, like two-sided platforms and auctions, leaving open how to design interventions for general repeated games. We address this gap by formalizing the connection between empirical observations from prior work on Q-learning collusion and classical theory of Simple Penal Codes (SPCs). We show any non-trivial SPC induces a quantifiable conditional dependence in agents' policies, detectable via the total variation distance between an agent's action distributions across cooperation and defection histories. Building on this connection, we propose CURB (Collusion Unwinding via Reward shaping and Belief injection), a reward-shaping framework that penalizes this Total Variation (TV) distance signal during Q-learning and is guaranteed to convert any SPC fixed point of the dynamics into a trivial one, thus precluding collusive equilibria sustained by punishment threats. Empirically, CURB substantially reduces collusion by Q-learning agents in both Bertrand and Cournot Competition Repeated Games. We further demonstrate that CURB extends to deep Q-network agents in Bertrand competition, suggesting the mechanism generalizes beyond tabular Q-learning. --- ### Language-model groups overstate consensus when replaying human deliberation on a reasoning task Authors: Tengfei Shao Published: 2026-09-17 Categories: cs.AI, cs.CL, cs.CY, cs.MA arXiv: https://arxiv.org/abs/2609.20543v1 Full-consensus rates are often treated as indicators of collective cognition, yet depend on how participation and final states are operationalized. We replayed 100 held-out human Wason groups with matched large language model (LLM) agent groups, seeding one belief-anchored agent per participant's pre-discussion answer and scoring agents and people with the same code. Across human scoring definitions, estimates ranged from 24.0% to 57.0%; about one fifth of participants never posted, whereas agents almost always did. Agent groups remained more consensual in two post-unblinding sensitivity analyses: the submit-based comparison (n = 98) yielded gaps of 34.0 and 43.9 percentage points for chat and reasoning modes, and the participation-matched comparison (n = 45) yielded gaps of 34.1 and 44.4 points. These complementary routes reduced different measurement asymmetries yet converged within 0.5 percentage points. The gap persisted without early stopping and under a reparameterization removing the memorizable answer; reasoning-mode groups then agreed nearly unanimously, mostly on incorrect answers. Simulated consensus did not track collective accuracy, and belief-anchored agent groups were biased estimators of the human group-outcome distribution in this setting. These analyses provide a scoring-explicit basis for assessing simulated-group estimates of human deliberative outcomes. --- ### An Analysis of Training-Free Self-Reported Confidence in Language Models Authors: Lukas Meyer, Sofia Rossi, Wei Chen, Thomas Laurent, Yiming Li Published: 2026-09-17 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.20541v1 Large language models can report a numerical confidence together with generated content, but it is unclear whether this report is more than calibrated rhetoric. We analyze three training-free signals: confidence verbalized with the answer, post-hoc $P(\mathrm{True})$, and agreement with three additional generations on the same 100 TriviaQA questions for two model families. Direct verbalization is a surprisingly strong baseline: after auditing benchmark errors, it reaches AUROC 0.956 and 0.937 for correctness prediction. Three-sample agreement is substantially weaker (0.765 and 0.790), and a fixed interpolation with verbalized confidence has no statistically reliable benefit. Four of nine errors from one model and two of eight from the other receive unanimous sample support, showing that self-consistency can amplify shared misconceptions. Re-eliciting confidence for the same fixed answers with equivalent prompts changes scores by 0.043 to 0.084 on average and flips 4\% to 9\% of decisions at a 0.8 threshold. An exploratory audit of 100 confidence-tagged biography claims further finds only a modest confidence gap between supported and contradicted claims. These results argue that useful self-reports remain sensitive to elicitation, correlated errors, and benchmark noise. --- ### Parallelism, critical windows, and separations among diffusion language models Authors: Sitan Chen, Liye Wang Published: 2026-09-17 Categories: cs.LG, cs.DS, math.ST, stat.ML arXiv: https://arxiv.org/abs/2609.20539v2 A popular selling point of diffusion large language models (dLLMs) is their capacity for parallelism: the ability to generate sequences of text far more efficiently than autoregressive models, which require one forward pass per token. Yet among the many competing paradigms for dLLMs, from masked to uniform to Gaussian diffusion, principled understanding of how these different proposals compare in parallelism remains limited. In this work, we initiate a fine-grained comparison of the capacity for parallelism among these three leading approaches and prove the following: - Uniform and Gaussian diffusion can sample in a number of forward passes which scales with the dual total correlation of the underlying distribution, a measure of intrinsic complexity which can be much smaller than the context length. Previously, it was only known how to achieve this using masked diffusion. - For a certain family of random empirical measures, we show that $\widetildeΘ(\sqrt{d})$ forward passes are necessary and sufficient to sample using uniform or Gaussian diffusion, yet there exist approximate score oracles for which $\widetildeΩ(d)$ forward passes are needed for masked diffusion. This establishes the first provable separation in parallelism between the three prevailing dLLM paradigms. Contrary to popular intuition that masked diffusions are harder to parallelize because they must commit to token values, the latter separation instead comes from the fact that the critical windows in masked diffusion sampling are asymptotically narrower than those in uniform and Gaussian diffusion sampling. --- ### Parallelism, critical windows, and separations among diffusion language models Authors: Sitan Chen, Liye Wang Published: 2026-09-17 Categories: cs.LG, cs.DS, math.ST, stat.ML arXiv: https://arxiv.org/abs/2609.20539v1 A popular selling point of diffusion large language models (dLLMs) is their capacity for parallelism: the ability to generate sequences of text far more efficiently than autoregressive models, which require one forward pass per token. Yet among the many competing paradigms for dLLMs, from masked to uniform to Gaussian diffusion, principled understanding of how these different proposals compare in parallelism remains limited. In this work, we initiate a fine-grained comparison of the capacity for parallelism among these three leading approaches and prove the following: - Uniform and Gaussian diffusion can sample in a number of forward passes which scales with the dual total correlation of the underlying distribution, a measure of intrinsic complexity which can be much smaller than the context length. Previously, it was only known how to achieve this using masked diffusion. - For a certain family of random empirical measures, we show that $\widetildeΘ(\sqrt{d})$ forward passes are necessary and sufficient to sample using uniform or Gaussian diffusion, yet there exist approximate score oracles for which $\widetildeΩ(d)$ forward passes are needed for masked diffusion. This establishes the first provable separation in parallelism between the three prevailing dLLM paradigms. Contrary to popular intuition that masked diffusions are harder to parallelize because they must commit to token values, the latter separation instead comes from the fact that the critical windows in masked diffusion sampling are asymptotically narrower than those in uniform and Gaussian diffusion sampling. --- ### Refuse, Decompose, Refresh: A Claim-Safe Protocol for Closed-Loop AI Evaluation Authors: Peiying Zhu, Sidi Chang Published: 2026-09-17 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.20538v1 An AI evaluation can be perfectly reproducible and still support the wrong claim. This risk is acute in closed-loop systems: policy determines visited states, observable components, and which failures leave a measurable trace. We propose a claim-safe protocol with three actions. Refuse: abstain when a clean reference stream or matched runtime comparison lacks support. Decompose: report protocol execution, operational false admission, and structural hypotheses separately rather than as one PASS/FAIL label. Refresh: treat distribution-shift alarms as requests to invalidate and recompute a reference map, not as fault evidence. We instantiate the protocol in an aggregate-only simulator with 24 policy components, three demand regimes, two fault-mask families, and independent development and heldout seeds. The preregistered heldout contains 1,440 cases and 21,600 partition rows. Only 55/72 regime-component units were reference-admitted and 54/55 remained runtime-admitted, making abstention part of the result. Stable false admission was 0/20 represented components, with a one-sided exact 95% upper bound of 0.1391 under a frozen 0.20 rule. Within admitted units, affected clean traffic outpredicted nominal fault-cell fraction: across 540 unit-arm rows nested in 20 component clusters, the cell-minus-traffic negative-log-likelihood difference was 0.1264 nats per row, with a 95% component-cluster interval of (0.0593, 0.1918). A drift log shows why "null" must be reference-relative: clean fault-null streams triggered 15/15, 0/15, and 14/15 alarms across three regimes, while only the middle regime matched the frozen detector reference. Rather than a universal threshold, we contribute an executable contract linking observable support, statistical calibration, and justified claims. --- ### FreqCondNorm: Towards Cross-domain Predictive Maintenance through a Frequency-Conditioned Transformer Foundation Model Authors: Zaynab Raounak, Camille LHermine, Zhiguo Zeng Published: 2026-09-17 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.20535v1 Deep learning predictive maintenance models suffer from poor transferability across machines and operating conditions, especially when labelled data are scarce and signals span five orders of magnitude in sampling frequency (1 Hz to ~100 kHz). We propose FreqCondNorm, a Transformer-based architecture that introduces a FiLM-style frequency-conditioned normalization layer to unify heterogeneous time-series within a single model. The architecture is pretrained on five public predictive maintenance datasets (CWRU, MFPT, UOC18, PRONOSTIA, CMAPSS) using masked auto-encoding and contrastive learning with balanced domain sampling. On fault diagnosis, the model achieves 99.2% accuracy on CWRU (+6.4 pp over CNN) and 82.1% zero-shot accuracy on MFPT, demonstrating strong transfer across sampling frequencies. However, the approach does not improve remaining useful life prediction, suggesting a mismatch between pretraining and RUL objectives that warrants future investigation. --- ### Relational Attention for Data-Efficient Language Modeling Authors: Adrian Brasoveanu, Ece Takmaz, Jakub Dotlačil Published: 2026-09-17 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.20530v1 We present Relational BabyLM, a system submission to the BabyLM 2026 challenge that combines two cognitively motivated inductive biases in a single decoder-only Transformer. Architecturally, we replace standard self-attention with a Dual Attention Transformer (DAT), which separates the routing of object-level ("sensory") lexical features from structural/relational information (Altabaa and Lafferty, 2025; Altabaa et al., 2024; Webb et al., 2024; Kerg et al., 2022; Webb et al., 2021). Relational attention (RA) disentangled from self-attention greatly increases data efficiency and out-of-training-sample generalization on purely relational tasks, but language modeling requires object-level and relational information to be integrated as well as disentangled, and RA-based LMs have remained largely unexplored. BabyLM's data-constrained training and comprehensive evaluation is an ideal testing ground for whether that data efficiency transfers. As a training intervention, we add a Next-Latent Prediction (NextLat; Teoh et al. 2026) objective that encourages hidden states to compress history incrementally into a dense belief state. Architecture is the dominant factor for structural linguistic generalization; the objective is secondary but still significant. DAT's three relational attention types (full RA vs. the simpler RCA and DisRCA variants) are largely interchangeable at 10M words; full RA pulls ahead at 100M. We also introduce a novel symbol-retrieval mechanism (RoPE-based, as opposed to learned, relative symbols) that matches learned symbol libraries while adding no parameters. On the strict (100M-word) track, our best model ranks 6th of 55 overall and 3rd of 55 on the leaderboard's NLP-task subset at the time of writing; our two strongest models outperform the GPT-2 baseline on most benchmarks, with one attaining the highest EWoK score among strict-track entries. --- ### Noise-Robust Quantum State Characterization for Remote State Preparation with Deep Learning Authors: Bo Tang, Zixuan Liao, Hao Li, Yilin Yang, Jiani Lei, Zengya Li, Jing Qiu, Zhaohui Dong, Zhengyang Mao, Yuanhua Li, Yuanlin Zheng, Xianfeng Chen Published: 2026-09-17 Categories: quant-ph, cs.LG, physics.optics arXiv: https://arxiv.org/abs/2609.20523v2 Quantum communication underpins secure information processing and scalable quantum networks. In particular, remote state preparation (RSP) enables efficient quantum state transfer, but accurately estimating target states under complex noise remains challenging. Here, we propose a Transformer-based Quantum State Characterizer (TQSC) model for noisy RSP experiments. Our model reconstructs experimentally prepared pure and mixed photonic polarization states from noisy measurements in complex scattering environments, while its attention patterns provide physically grounded insights into correlations among the measured observables. The method achieves a mean estimator-target fidelity exceeding 99.999% under complex scattering and dynamic Gaussian noise, while its robustness and generalization are further examined using Qiskit-simulated Bloch-ball states. Furthermore, in a practical MNIST image transmission task with held-out states, the decoded bit error rate is reduced from 50.34% to zero after TQSC post-processing. The TQSC model enables accurate tomographic characterization under dynamic noise and provides physically grounded post-hoc insights, holding promise for intelligent quantum information processing applications. --- ### Noise-Robust Quantum State Characterization for Remote State Preparation with Deep Learning Authors: Bo Tang, Zixuan Liao, Hao Li, Yilin Yang, Jiani Lei, Zengya Li, Jing Qiu, Zhaohui Dong, Zhengyang Mao, Yuanhua Li, Yuanlin Zheng, Xianfeng Chen Published: 2026-09-17 Categories: quant-ph, cs.LG, physics.optics arXiv: https://arxiv.org/abs/2609.20523v1 Quantum communication underpins secure information processing and scalable quantum networks. In particular, remote state preparation (RSP) enables efficient quantum state transfer, but accurately estimating target states under complex noise remains challenging. Here, we propose a Transformer-based Quantum State Characterizer (TQSC) model for noisy RSP experiments. Our model reconstructs experimentally prepared pure and mixed photonic polarization states from noisy measurements in complex scattering environments, while its attention patterns provide physically grounded insights into correlations among the measured observables. The method achieves a mean estimator-target fidelity exceeding 99.999% under complex scattering and dynamic Gaussian noise, while its robustness and generalization are further examined using Qiskit-simulated Bloch-ball states.Furthermore, in a practical MNIST image transmission task with held-out states, the decoded bit error rate is reduced from 50.34% to zero after TQSC post-processing. The TQSC model enables accurate tomographic characterization under dynamic noise and provides physically grounded post-hoc insights, holding promise for intelligent quantum information processing applications. --- ### Sharp spectral norm concentration of sparse random tensors Authors: Zhixin Zhou, Yizhe Zhu Published: 2026-09-17 Categories: math.PR, math.CO, math.ST, stat.ML arXiv: https://arxiv.org/abs/2609.20520v1 We prove a sharp concentration inequality for the spectral norm of sparse random tensors with independent Bernoulli entries. Let $T$ be an order-$k$ tensor of dimension $n\times\cdots\times n$ with independent Bernoulli$(p)$ entries, where $k$ is fixed. For any $c,r>0$, we show that $\|T-\mathbb E T\|\le C_{k,r,c}\sqrt{np}$ with probability at least $1-n^{-r}$ whenever $np\ge c\log n$. We extend this bound to inhomogeneous Bernoulli sampling with deterministic entrywise weights. This removes the logarithmic factor in the work of Zhou and Zhu (2021). The proof follows the Kahn--Szemerédi light--heavy decomposition with a refined estimate on the heavy tuple part. We also obtain a log-free second eigenvalue bound for the random hypergraph model of Friedman and Wigderson (1995). --- ### SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness Authors: Haozhe Liu, Tian Ye, Sensen Gao, Qihang Cao, Yitong Li, Mingchen Zhuge, Duomin Wang, Ruihua Zhang, Ping Luo, Jiawang Bian, Lei Zhu, Ligeng Zhu et al. Published: 2026-09-17 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.20519v1 As coding agents move from supervised code completion to unattended, around-the-clock exploration, their work expands from isolated predictions into long trajectories of reasoning, tool use, and feedback. Token efficiency therefore becomes important for scaling recursive self-improvement. We take an RSI-inspired approach at the harness layer, scaling auto-research loops across increasingly numerous and diverse environments for harness rollouts. At this scale, the process yields reusable improvements that transfer beyond their development setting, moving automated harness discovery toward production-level outcomes. Four mechanisms survive selection and form SoL-Pi, spanning action execution, context compaction, observation handling, and delegated reading. On the 51-task EdgeBench evaluation, SoL-Pi achieves performance comparable to Pi across GPT-5.6 Sol and Opus 5 while reducing recorded token traffic by 44.7-49.0% and API cost by about one third. In other words, estimated hourly savings are \$8.75-\$13.50 relative to native Codex and Claude Code harnesses, and \$4.36-\$5.71 relative to Pi. --- ### When EOS Tokens Disagree: Understanding Length Inflation in On-Policy Distillation Authors: Yuxiao Yang, Tianrun Yu, Shangzhe Li, Kaixiang Zhao, Xuchao Zhang, Chetan Bansal, Huaxiu Yao, Taylor W. Killian, Weitong Zhang Published: 2026-09-17 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.20511v1 We study length inflation in on-policy distillation (OPD), where student responses can become excessively long and even exhaust the generation budget. We identify \emph{termination-token mismatch} between base students and post-trained teachers as an important source of this behavior. Across Qwen3, Llama, and Gemma, the two models can place their stopping probability on different EOS tokens, even when their declared stopping sets are identical. This mismatch can suppress the student's preferred termination action without reliably transferring the teacher-preferred alternative. We show that aligning the decoding stopping set alone is insufficient, while treating functionally equivalent EOS tokens as a shared semantic stopping action substantially mitigates mismatch-induced length inflation across all three model families. To further understand how termination behavior evolves over training, we study OPD across different K2-Horizon training stages. This stage-wise analysis shows that termination preferences can shift substantially during training, while also revealing a distinct length inflation late in the OPD run that persists beyond termination alignment. Together, these results identify termination mismatch as an important, but not exhaustive, source of OPD length dynamics. We release an implementation incorporating the proposed termination-handling corrections. --- ### Truncated automatic sparse differentiation for machine learning interatomic potentials Authors: Marcel F. Langer, Adrian Hill, Michele Ceriotti Published: 2026-09-17 Categories: physics.chem-ph, cond-mat.mtrl-sci, cs.LG arXiv: https://arxiv.org/abs/2609.20510v1 Machine learning interatomic potentials (MLIPs) learn the mapping from atomic positions to potential energy. The forces, the negative gradient of this energy, drive molecular dynamics and are readily obtained using automatic differentiation. Higher-order derivatives, most notably the Hessian, describe collective motion and allow the direct prediction of experimental observables, but are considered computationally inaccessible for large systems. We suggest a solution: in physical systems, interactions decay with distance, and most MLIPs build on this locality through message passing up to a finite receptive field. This implies both sparsity of higher-order derivatives and their decay with distance. This structure can be exploited using automatic sparse differentiation (ASD). We explain how to compute the sparsity pattern for MLIP derivatives and demonstrate that, for multiple foundation MLIPs, ASD computes full Hessians of large porous materials exactly, but with modest speedups at best. The larger gains come from truncated ASD: discarding small, but nonzero, Hessian entries between distant atoms yields order-of-magnitude speedups with negligible impact on predicted observables. --- ### Automated Goldsmith's Mark Retrieval in Silverware Authors: Atmik Tiwari, Vincent Christlein, Mark Fichtner, Freya Gohlke, Birgit Schübel, Theresa Witting, Heike Zech, Mathias Zinnen Published: 2026-09-17 Categories: cs.CV, cs.DL arXiv: https://arxiv.org/abs/2609.20509v1 For art historians, goldsmith marks play a critical role in the identification and dating of artifacts. In practice, experts must manually compare a query mark against hundreds of documented examples, a process that is both tedious and highly dependent on specialist knowledge. To address this, we present an AI-assisted retrieval pipeline that combines mark localization with metric-learning fine-tuning across three backbone architectures: an ImageNet-pretrained ResNet-50, a supervised ViT-S/16, and a self-supervised DINOv2 ViT-S/14. We conduct a systematic evaluation of cropping strategies, where we measure the impact of no cropping, manual ground-truth cropping, and learned detection-based cropping, and assess their interaction with each backbone. Our strongest configuration, DINOv2 ViT-S/14 with manual crop and metric-learning fine-tuning, achieves an mAP of 62.63% and a Top-1 accuracy of 73.74%. Our experiments show that self-supervised pretraining and mark localization are the two most impactful factors, with learned cropping recovering the majority of the gain from manual cropping without requiring ground-truth annotations at inference time. To enable reproducibility and adoption in the digital humanities, we release our manually annotated dataset and codebase, and deploy the system via a public web interface. --- ### Grounded Product Understanding in Livestream Videos Authors: Xinyu Zhang, Junjie Chen, Jiawei Ge, Qianlong Li, Libin Ma, Baokun Pan, Yahui Luo Published: 2026-09-17 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.20508v2 E-commerce livestreams have emerged as an important channel for presenting products to online consumers, often featuring multiple products with relevant information distributed across different moments. This poses significant challenges for downstream product understanding applications, such as product-centric livestream clipping, where models need to identify the product and its relevant segments for information gathering. However, existing benchmarks for general product understanding typically evaluate product retrieval and temporal localization in isolation, leaving the critical correspondence between product identity and temporal evidence largely unassessed. To address this limitation, we introduce GPUB, a large-scale benchmark comprising 3,000 real-world e-commerce livestream instances with quality-controlled multi-moment temporal annotations and a catalog of over 31K fashion products. GPUB supports three evaluation tasks: given a livestream video and a candidate product set, the main task Grounded Product Understanding (GPrU) requires jointly identifying the product being presented and localizing its supporting moments; Product Retrieval and Product Moment Localization serve as two complementary subtasks. Evaluation of existing multimodal models shows that GPrU remains highly challenging, with the best-performing off-the-shelf baseline achieving only 10.13% Pair mAP@.3. To narrow the performance gap, we further develop UniPro, a unified product understanding model that derives product-aligned and temporally structured representations from shared multimodal encoding, improving Pair mAP@.3 to 24.58% while achieving 38.81% Joint R@1@.3 on GPrU. --- ### Grounded Product Understanding in Livestream Videos Authors: Xinyu Zhang, Junjie Chen, Jiawei Ge, Qianlong Li, Libin Ma, Baokun Pan, Yahui Luo Published: 2026-09-17 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.20508v1 E-commerce livestreams have emerged as an important channel for presenting products to online consumers, containing multiple products whose information is scattered in different moments. This poses significant challenges for downstream product understanding applications, such as product-centric livestream clipping, where models need to identify the product and its relevant segments for information gathering. However, existing benchmarks for general product understanding typically evaluate product retrieval and temporal localization in isolation, leaving the critical correspondence between product identity and temporal evidence largely unassessed. To address this limitation, we introduce GPUB, a large-scale benchmark comprising 3,000 livestream instances with quality-controlled multi-moment temporal annotations and a catalog of over 31K fashion products. GPUB supports three evaluation tasks: the main task Grounded Product Understanding (GPrU) requires jointly identifying the target product and localizing its supporting moments from a livestream video and a candidate product set; Product Retrieval and Product Moment Localization serve as two complementary subtasks. Evaluation of existing multimodal models shows that GPrU remains highly challenging, with the best-performing baseline achieving only 10.13% Pair mAP@.3. To narrow the performance gap, we further develop UniPro, a unified product understanding model that derives product-aligned and temporally structured representations from shared multimodal encoding, improving Pair mAP@.3 to 21.53% while achieving 37.23% Joint R@1@.3 on GPrU. --- ### Radio Frequency Detection and Classification of Microplastics in Water Authors: Jaden Tolbert, Md Saiful Islam, Pingshan Wang Published: 2026-09-17 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.20507v1 Micro- and nano-plastic particles (MPs/NPs) are ubiquitous environmental contaminants whose increasing abundance and potential health impacts have created an urgent need for rapid, label-free detection methods. As particle size decreases to the low-micrometer range, conventional optical and spectroscopic techniques become increasingly challenging because of limited throughput and/or complex sample preparation. In this work, we present a machine learning (ML)-assisted radio-frequency (RF) dielectric spectroscopic cytometry (DiSC) platform for the label-free detection and classification of MPs. Eight types of $ 10 $ μm nominal-diameter MP particles suspended in deionized (DI) water were characterized at four frequencies spanning $ 0.2\text{-}9\text{ GHz} $. The measured alterations in RF scattering parameters (S-parameters), referenced to the carrier medium, were used to train supervised ML models for material classification, including the identification of MPs in mixed samples and saline-water environments. For eight MP classes suspended in DI water, the proposed method achieved macro-average F1-score, precision, and recall values exceeding $ 0.71 $. Furthermore, PET classification performance was largely maintained in saline carrier media containing $3.3\% $ and $ 6.6\% $ sea salt. These results demonstrate the feasibility of ML-assisted RF DiSC for rapid, single-particle MP classification in aqueous environments. Future work will focus on improving classification performance through enhanced RF calibration, increased spectral coverage, larger training datasets, and validation using environmentally aged and biologically contaminated microplastics. --- ### Model-Agnostic and Language-Agnostic Voice Pipeline Improvement for the Agriculture Domain Authors: Aakash Singh, Lakshmi Pedapudi, Chandrashekar M S, Sanyam Singh, Naga Ganesh, Vineet Singh Published: 2026-09-17 Categories: eess.AS, cs.AI, cs.CL, cs.SD arXiv: https://arxiv.org/abs/2609.20504v1 FarmerChat is Digital Green's AI-powered agricultural advisory assistant for smallholder farmers, who access it in their own language through text, voice, or photographs. Voice is a critical channel for this population, yet field-recorded speech is challenging for general-purpose automatic speech recognition (ASR) because recordings frequently contain machinery noise, background media, competing speakers, and domain-specific agricultural vocabulary. These conditions disproportionately affect crop, pest, chemical, and quantity terms that carry the meaning of a farmer's query. We present a modular, model-agnostic pipeline for improving ASR quality in FarmerChat without fine-tuning or replacing the underlying ASR model. The pipeline combines gated audio enhancement, speaker diarization and target-speaker selection, ASR, domain-aware correction using a weighted agricultural lexicon, and a quality gate for detecting unreliable transcripts. Only the diarization stage is fine-tuned; all other stages use off-the-shelf models behind common interfaces. We evaluate the pipeline on human-annotated FarmerChat recordings in Hindi, Telugu, and Odia using word error rate (WER) and a domain-weighted error rate that gives greater importance to agricultural terminology. The largest improvements occur on multi-speaker recordings, where target-speaker selection prevents competing speech from entering the transcript. Across the full corpus, the pipeline reduces WER by 16-23% relative on three cloud ASR models and by 5% on an on-device model. On multi-speaker recordings, the reductions are 32-42% for the cloud models and 16% for the on-device model. All reported reductions are statistically significant. These results show that targeted preprocessing, speaker selection, and domain-aware post-processing can substantially improve agricultural speech transcription while preserving the underlying ASR model. --- ### Distributionally Robust Federated Learning with Multi-Source Data Authors: Yingzhu Liu, Zhongkui Li, Pengcheng You, Ashish Cherukuri Published: 2026-09-17 Categories: cs.LG, math.OC arXiv: https://arxiv.org/abs/2609.20501v1 Federated learning trains a shared model from private client data. In practice, data-generating distributions may differ, and the true mixture across clients is often unknown, making the underlying group distribution difficult to specify. Existing approaches address cross-client mixture uncertainty by optimizing against the worst-case mixture, yet assume accurate client-wise distribution estimates. However, these estimates can be unreliable when based on finite samples. To handle both cross-client mixture uncertainty and within-client distributional ambiguity, we construct a global ambiguity set as the union of admissible mixtures of local ambiguity sets. The construction allows client-specific ambiguity radii and admits a client-wise separable reformulation. Leveraging this structure, we establish a high-probability out-of-sample performance guarantee. We further develop a federated algorithm for a penalty-based reformulation and prove its convergence under milder regularity conditions. Simulations validate the algorithm's effectiveness. --- ### Resolution limits for process comparison from event data Authors: Antony R. Lee, Peter Tiňo, Iain B. Styles Published: 2026-09-17 Categories: cs.DB, cs.LG, stat.ML arXiv: https://arxiv.org/abs/2609.20489v1 One hospital runs bloods and imaging at the same time. Another runs them one after the other, in either order, equally often. Knowing which actually happened, and how it is recorded in data, is critical for all operational managers. In process mining, the standard approach is to construct an event log, and attempt to discover concurrent and sequential processes in a data-driven way. We show this standard approach, built on the stochastic language of an event log, reports only the assumptions of its discovery algorithm, because every such log is explained equally well by a model with no concurrency at all. Further, before any data is acquired, we characterise when data can and cannot distinguish concurrent behaviour. Where it cannot, the distinction is recoverable from evidence the stochastic language discards, such as the times at which activities start and end, or object-centric records that fix an order within an execution. The remedy is therefore a choice of what is recorded, rather than a larger sample. This impacts decision making, as planning resource for truly concurrent services is very different from sequential services. --- ### Edustories: A Collection of Real-world Case Studies from Classroom Practices Authors: Michal Štefánik, Jan Nehyba, Jirina Karasova, Martin Fico, Lucie Škarková, Markéta Košatková, David Kosatka Published: 2026-09-17 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.20484v1 Despite the widely recognized potential of AI in education, most prior work has focused on individualized student assistance. In contrast, the majority of educational practice worldwide still takes place in collective classroom settings. To enable researchers to study AI assistance in collective teaching, we introduce Edustories, a dataset of 1,492 teacher-written case studies describing real elementary and high-school classroom situations involving challenging student behavior, pedagogical interventions, and their outcomes. Among many other applications, Edustories enables evaluating LLMs' ability to predict the success of teacher interventions, crucial for providing practicing teachers with useful feedback. Comparing the latest models from four language-model families against expert assessments, we find that current models fall short of human expertise in predicting classroom outcomes; the strongest models reach 58% accuracy compared to 64% of human experts. This gap highlights both the limitations and the emerging potential of AI as assistants for practicing teachers. --- ### greCAPTCHA: Assessing Understanding as Evidence of Research Authorship Under Generative AI Authors: Justin Payan, Bálint Gyevnár, Atoosa Kasirzadeh, Nihar B. Shah Published: 2026-09-17 Categories: cs.DL, cs.AI, cs.HC arXiv: https://arxiv.org/abs/2609.20481v1 Conferences, journals, funders, schools, and universities are struggling with a surge of potentially AI-generated submissions from ostensibly human authors, who may not have exercised sufficient human oversight for their manuscripts. In turn, institutions evaluating submissions can no longer reliably credit expertise based solely on authors' names on submitted work. To address this problem, we propose greCAPTCHA, a proctored assessment approach that measures authors' understanding of research manuscripts via the construct of capacity to verify, which we define as the knowledge and reasoning required to critically assess the contents underlying one's contributions to a manuscript. greCAPTCHA generates questions assessing multiple levels of understanding and provides an evaluative report based on authors' responses. Using a prototype implementation, we conduct a user study and semi-structured interviews with $31$ researchers to evaluate greCAPTCHA. Its automated scores predict which papers were or were not authored by study participants with an AUC of $0.90$. Participants reported positive overall experiences with the system and remarked on the appropriate construct validity for author understanding, while also suggesting important changes to be made before deployment. Our results provide initial evidence that greCAPTCHA can assess manuscript-specific understanding under proctored conditions. ---