# Research Papers — 2026-09 · bundle page 53 of 266 > Full-text markdown of papers #2601–2650 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/53 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/52.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/54.md) ### RewardExplainer: Learning Reward Model Explanations from Counterfactual Preference Feedback Authors: Jingyi He, Nier Wu, Shuang Liu, Xin Wang, Mengnan Du, Xia Hu Published: 2026-09-27 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.33989v1 Reward models (RMs) are a key component of large language model post-training, providing reward signals for subsequent reinforcement learning. However, conventional discriminative RMs typically output only scalar scores, making it difficult to identify the response behaviors associated with their scoring decisions. Existing interpretation methods often rely on predefined high-level attributes and require repeated counterfactual interventions for each response pair to validate candidate explanations, lacking a closed-loop mechanism that uses RMs' feedback to train a reusable explainer. To address this, we propose RewardExplainer, a framework that obtains feedback from the target reward model through counterfactual rewriting and uses this feedback to further optimize the explainer. RewardExplainer generates open-ended, atomic, and intervenable natural-language scoring mechanisms, making explanations more concrete, readable, and actionable. It further converts counterfactual feedback into preference supervision, enabling the explainer to more faithfully capture the target RM's scoring preferences and sensitive behaviors than single-pass generation. Extensive experiments across multiple target RMs and explainer backbones show consistent improvements. Beyond interpretation, we use the generated mechanisms to identify potential bias patterns and construct targeted debiasing data for fine-tuning the reward model, improving robustness on reward-hacking benchmarks. --- ### Opera: A Verbal Critic Framework for Long-horizon Coding Agents Authors: Kai Mei, Zhiyuan Hu, Yutong Dai, Juntao Tan, Yifan Zhang, Dingjie Song, Dimitris N. Metaxas, Silvio Savarese, Ran Xu, Zeyuan Chen Published: 2026-09-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.33987v1 Long-horizon coding agents need timely corrections, yet feedback can be ineffective or even harmful when it misjudges ongoing work or fails to address the underlying problem. Existing critics focus on evaluating trajectories and generating feedback, but rarely track what happens after feedback is delivered. We present Opera, a verbal critic framework that treats each correction as a persistent note, followed until the diagnosed problem is resolved. Opera decides when to review through periodic and event-driven triggers, diagnoses issues with typed operators, audits feedback against visible evidence before delivery, and tracks the agent's subsequent actions to distinguish mere compliance from actual resolution. As a test-time critic, Opera improves the resolve rate of non-critic agents by up to 12.4, 15.0, and 8.9 percentage points on Terminal-Bench 2.1, a SWE-Bench Pro subset, and DeepSWE v1.1, respectively, across four policy models, and achieves the highest mean resolve rate among competitive critic baselines on all three benchmarks, and also improves policy models when the policy critiques itself. Beyond inference, Opera-guided rollouts provide approximately on-policy training data: fine-tuning Qwen3.5-9B on them improves its resolve rate on held-out SWE-Bench Pro repositories by 10.2 percentage points without a critic at inference time, matching fine-tuning on rollouts from a stronger model, while preserving its performance when switching harness, i.e., from Openhands to Terminus-2, which the latter substantially degrades. --- ### ICMAPE: In-Context Multiagent Pure Exploration Authors: Xinyi Hu, Alessio Russo, Aldo Pacchiano Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33986v1 In some multi-agent systems, the quantity to be optimized is not an externally specified reward but the information acquired about unknown properties of the environment as done in active sequential hypothesis testing (ASHT) problems. However, the ASHT literature tends to focus on finite single-agent problems with well-specified models, while there is currently a gap for practical multi-agent methods that can perform active sequential testing. We fill this gap with ICMAPE, a Bayesian learning-based framework for decentralized multi-agent pure-exploration driven by inference objectives. ICMAPE converts the fixed-confidence identification objective into a reward derived from inference confidence, so that standard reinforcement learning machinery can be applied to decentralized pure exploration. It jointly learns a centralized neural inference network that estimates a posterior distribution over hypotheses from global trajectory data, and decentralized policies that select actions from local observation histories and learn when to stop collecting data once the target confidence is reached. On two synthetic benchmarks and a Maryland nitrate concentration monitoring task based on real-world data, ICMAPE-TD3 achieves target accuracy with fewer exploration steps. --- ### The Privacy Fallacy of Crowdsourced Fine-Tuning: Extracting Proprietary Data via Topic-Based Poisoning Authors: Sae Furukawa, Alina Oprea Published: 2026-09-27 Categories: cs.CR, cs.LG arXiv: https://arxiv.org/abs/2609.33985v1 Supervised fine-tuning (SFT) is widely used to adapt large language models to downstream tasks. Crowdsourcing user conversations is an established approach to collecting SFT data at scale while reducing the need for costly manual annotation. However, it also allows untrusted users to contribute data to the fine-tuning pipeline. We investigate an underexplored privacy risk arising from this setting: can a malicious user poison a small fraction of the crowdsourced data to amplify extraction of previously unseen instructions contributed by other users? We show that this is possible using only black-box, output-only access to the deployed model. Experiments across four models and two datasets demonstrate substantial increases in training-data extraction: with only 50 poisoned examples, near-verbatim extraction reaches $3.71\times$ the rate without poisoning for Qwen2.5-14B on OpenMathInstruct and $3.08\times$ for Llama-3.1-8B on AceReason. Data filtering also proves largely ineffective in detecting poisoned samples: even the best-performing method achieves only 0.378 in F-1 score, leaving the majority of poisoned samples undetected. These findings demonstrate that seemingly benign crowdsourced contributions can amplify leakage of other records while remaining difficult to identify through data filtering. --- ### From HL to H+L-1 Parameters: A Hankel-Toeplitz Forecaster for Long-Term Time Series Forecasting Authors: Chaoqi Zhang, Yu Wang, Haixu Tang Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33984v1 Linear forecasters have shown competitive accuracy against Transformer-based models in long-term time series forecasting. We study how classical stationary prediction theory can guide parameter sharing for more compact linear forecasters. For centered second-order stationary processes with nonsingular history covariance, the minimum-MSE finite-window linear predictor factors into a Hankel cross-covariance matrix and an inverse Toeplitz covariance matrix. Shared lags and scale cancellation specify this predictor using $H+L-1$ autocorrelations for lookback $L$ and horizon $H$. Building on the innovations representation, our Hankel-Toeplitz Forecaster (HTF) learns one impulse response that defines both an inverse filter and a forecast map. We characterize the finite-history correction and, under summability assumptions, bound the excess risk of truncating the true filters. HTF uses $H+L-1$ trainable coefficients while allowing a full-rank forecasting matrix. Across seven benchmarks at $L=336$, its horizon-averaged MSE is within 1.2% of Dense Linear on each dataset with 75-229 times fewer trainable parameters. --- ### High-Level Text Preprocessing for Semantic Similarity Analysis of Discursive Texts: A Framework and Empirical Demonstration Authors: Mehmet Murat Albayrakoglu, Mehmet Nafiz Aydin Published: 2026-09-27 Categories: cs.CL, cs.IR, cs.LG arXiv: https://arxiv.org/abs/2609.33983v1 Semantic Textual Similarity (STS) methods assume that a document's lexical content faithfully represents what it asserts. This assumption fails for discursive documents that discuss, compare, critique, and contextualize other positions in the process of articulating their own. The result is semantic diffusion: similarity scores between documents are inflated by vocabulary acquired through discursive engagement rather than substantive alignment. Standard Natural Language Processing (NLP) preprocessing (tokenization, stopword removal, stemming, lemmatization) cannot address this problem because it operates at the lexical level, treating all content identically regardless of its discursive function. This paper introduces high-level text preprocessing: a systematic, rule-based intervention applied before the standard preprocessing pipeline to isolate each document's actual claim from its discursive structure. We propose 12 rules, each with an explicit rationale, and demonstrate their effect on an encyclopedic philosophical corpus: three entries from the Stanford Encyclopedia of Philosophy (virtue ethics, deontological ethics, and consequentialism). A three-phase experiment using eight Transformer-based STS models shows that preprocessing reduces centroid cosine similarity scores across all three theory pairs, with 23 of 24 model-pair comparisons showing the expected decrease and cross-model agreement ranging from 7-1 to 8-0. We introduce the semantic diffusion index (SDI), a per-document metric for assessing the semantic reorientation between a document's raw and high-level preprocessed representations. Although the framework is demonstrated using philosophical texts, it potentially addresses a domain-agnostic problem applicable to legal texts, policy documents, academic articles, and any genre in which a discursive approach introduces vocabulary from positions the document does not endorse. --- ### Test-Time Spatial Reasoning for Robot Manipulation Using Generative Real-to-Sim Authors: Ivan Kapelyukh, Yafei Hu, Ran Gong, Brandon May, Tushar Kusnur, Laura Herlant, Karl Schmeckpeper, Edward Johns, Xiaohan Zhang Published: 2026-09-27 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2609.33982v1 Spatial reasoning is fundamental to general robot intelligence, as it enables robots to complete long-horizon tasks involving multi-object interaction. We introduce Simify, a training-free, test-time framework that performs explicit spatial reasoning via massively parallel physics simulation. From a single RGB-D image of a scene, Simify reconstructs simulation-ready assets leveraging 3D generative models and vision-language models. Then given a task specified by a reward function (e.g., build the tallest tower), Simify launches thousands of parallel rollouts in simulation and performs an evolutionary search to optimize object arrangements, typically converging within seconds. We conduct quantitative experiments on real-robot hardware to demonstrate the ability of our framework to execute complex object rearrangement tasks end-to-end with previously unseen objects. Results show that our framework outperforms prior work on foundation models for spatial reasoning by effectively exploiting large-scale parallel simulation during inference, and also highlight the importance of complete and accurate geometry for successful sim-to-real transfer. --- ### Future Information-Directed Sampling for Bayesian Nonstationary Bandits Authors: Yichen Song, Alessio Russo, Aldo Pacchiano Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33981v1 Exploration--exploitation is a central trade-off in bandit learning. While classical algorithms such as upper confidence bound methods and Thompson Sampling effectively balance this trade-off in stationary environments, their exploration strategies mainly reduce uncertainty about the current optimal arm, which can be insufficient in nonstationary settings where future optimal arms may differ substantially from current ones. In this paper, we propose Future Information-Directed Sampling (FIDS), a new algorithm for Bayesian nonstationary bandits that explicitly explores to gather information about future optimal arms. We show that FIDS achieves regret comparable to Thompson Sampling up to a small constant factor, while being able to exploit predictive information structures that conventional exploration objectives fail to capture. To address the practical difficulty of posterior inference, we further propose a supervised-learning-based approximation framework that learns the FIDS policy from offline data, and demonstrate its effectiveness on synthetic benchmarks. --- ### DynGraphAgentBench: A Benchmark for Agentic Lifecycle Control in Dynamic Graph Anomaly Detection Authors: Yuwei Han, Lingwei Wei, Wooseong Yang, Liangjie Huang, Liancheng Fang, Huanhuan Ma, Philip S. Yu Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33980v1 Dynamic graph anomaly detection requires repeated decisions as graph structure and class prevalence drift, yet detector benchmarks usually score a fixed pipeline after current labels are known. We introduce DynGraphAgentBench, an executable benchmark for agentic lifecycle control under delayed feedback. It comprises seven temporal graph datasets with node- and edge-level anomaly tasks, eleven selectable detectors, and eight chronological deployment windows per dataset. In each window, a controller sees only time-causal aggregate context, registered model cards, and its own matured history. It must choose a detector before current-window training or candidate scores exist. A sandboxed executor trains the chosen architecture on mature data, scores a hidden deployment window, and releases the outcome after a one-window delay. A deterministic verifier checks decision timing, leakage guards, legal actions, training scope, and persisted artifacts. We measure detection utility with average precision and capture at fixed review depth, and characterize adaptation through model switches and compute. Complete eight-window trajectories from two primary controllers and a no-memory reference on four datasets, together with three additional controllers on three datasets, expose useful, costly, and ineffective reactions to delayed evidence without granting an exhaustive current-window oracle. --- ### Adapting neural operators for mechanics decisions under changing operating conditions Authors: Prashant K. Jha, Koffi Enakoutsa, Ian Galloway, Henry Anderson Published: 2026-09-27 Categories: cs.CE, cs.LG, math.NA arXiv: https://arxiv.org/abs/2609.33978v1 Neural operators can accelerate repeated nonlinear mechanics calculations, but their accuracy can deteriorate as operating conditions move beyond the training range. This work studies whether high-fidelity solutions acquired during use can be reused to adapt a neural operator and improve subsequent mechanics-based command selection. Two hard-magnetic soft-material systems are simulated using high-fidelity finite-element (FE) models, providing reference solutions for evaluating surrogate predictions and selected commands. A neural operator predicts deformation from known material, loading, and magnetic-field inputs, while an empirical error estimator determines which predictions may be used for command selection. Selected FE evaluations supplement these predictions, and their complete loading paths are retained for periodic updates of the neural operator and estimator. In both examples, the fixed operator loses substantial accuracy when stiffness and loading move outside the training range. Updates using 16 acquired paths recover much of the lost accuracy while preserving accuracy in the nominal regime. Under the same FE evaluation budget, the updated operators also improve command selection, although the benefit varies with the operating condition. Error estimation is less consistent, with inaccurate predictions sometimes accepted and accurate predictions rejected. These results demonstrate that reusing high-fidelity loading paths can extend the useful operating range of a neural operator. However, improved forward accuracy alone does not guarantee reliable prediction acceptance, highlighting prediction-specific error assessment as a separate requirement for trustworthy decision making. --- ### GroupMask: Layer-Adaptive Group-wise Sparsity for Semi-Structured LLM Pruning Authors: Zhengao Li, Shuoqiu Li, Xiaofang Zhang, Yukai Jin, Gokcen Kestor, Yanfu Zhang, Yiming Zeng, Bin Ren, Chuxu Zhang, Shangqian Gao Published: 2026-09-27 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.33977v1 Semi-structured pruning compresses large language models (LLMs) while keeping a regular sparse structure, but the prevailing N:M pattern fixes the same local sparsity ratio in every layer. Layer-adaptive sparsity allocation improves unstructured pruning, yet it has been reported to be less effective under N:M sparsity, leaving open whether adaptive allocation is of limited value for semi-structured pruning in general or only under the fine-grained N:M pattern. We examine this question with group-level sparsity, which partitions each weight matrix into regular groups, retains or prunes each group as a whole, and allows each layer's sparsity ratio to vary under a global budget. We propose GroupMask, which generates the group selectors of all layers with a lightweight hypernetwork, relaxes them with a Gumbel-Sigmoid parameterization and a straight-through estimator, and learns them through sparsity-budget regularization and self-distillation while keeping the pretrained weights frozen. On LLaMA-2-7B at 50% sparsity with the same $1\times256$ group size, learned layer-adaptive allocation reduces WikiText-2 perplexity from 10.02 to 8.30 and raises the average zero-shot accuracy from 0.455 to 0.496 relative to a uniform per-layer ratio. GroupMask obtains the lowest WikiText-2 perplexity on LLaMA-2-7B and the highest average zero-shot accuracy with Alpaca calibration among the evaluated baselines on five LLaMA and Qwen models. Our code is available at https://github.com/ZhengaoLi/GroupMask. --- ### Beyond Solo and Consistency: Vindicating Multi-Agent Debate via Conditional Progressive Pruning Authors: Ruosong Ye, Caiqi Zhang, Jiahao Li, Haijun Wu, Xiaolong Luo, Huiyuan Chen, Yu Wang, Ying Chen, Zhenting Wang, Kai Mei, Yang Zhou, Dimitris N. Metaxas Published: 2026-09-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.33974v1 Large Language Model (LLM) based Multi-Agent Debate (MAD) is one of the most effective test time scaling techniques. Through multi-round communication, agents complement each other in knowledge and reasoning and solve tasks that no single member can solve. However, existing MAD frameworks fail to beat strong Single Agent and Consistency-based baselines under the same strict cost limit, which shakes the foundation of the MAD field. We propose Conditional Progressive Pruning (CPP), a lightweight pruning framework that fully exploits multi-round MAD. CPP outperforms all existing MAD frameworks on multiple dominated benchmarks. It is also the first to fully outperform consistency methods. Our code, detailed agent interaction records will be released soon. --- ### HARMONIA: Interpretable Graph Learning through Mixtures of Neural Bases Authors: Quan D. Bui, Nguyen Do, An Nguyen Dang, Huyen Nguyen, Nhu Duc Minh Nguyen, My T. Thai Published: 2026-09-27 Categories: stat.ML, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.33972v1 Existing interpretable graph additive models still face limitations in either computational scalability or modeling flexibility. In terms of structural modeling, previous approaches either face quadratic scaling costs or sacrifice explicit source-to-target contribution decomposition. In terms of feature components, they rely either on per-feature neural networks or on single shared bases with limited feature specialization. We address both problems by introducing HARMONIA: Interpretable Graph Learning through Mixtures of Neural Bases, an interpretable-by-design framework. For feature modeling, HARMONIA introduces a Mixture of Neural Bases (MoNB), which routes features to specialized basis experts, enabling parameter sharing without sacrificing feature-specific specialization. For structural modeling, HARMONIA uses Relative Random Walk Probabilities (RRWP) to capture multi-hop and multi-path relationships, and proposes Sparse RRWP Aggregation (SRA) to compute these interactions through sparse graph propagation without quadratic pairwise complexity. HARMONIA retains a simple additive form in which predictions decompose into feature responses modulated by structural influence. Empirically, HARMONIA achieves stronger explanation recovery than existing interpretable graph baselines while maintaining competitive predictive performance and scaling to graphs with millions of nodes. These results show that interpretable graph learning can remain both faithful and scalable without sacrificing predictive effectiveness. --- ### Do System One Decisions Add Up? A Study of Probabilistic Coherence Authors: Saman Sarker Joy Published: 2026-09-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.33971v1 A decision model can give probabilities that sum to one for every question yet disagree with itself when the same decision is broken into smaller steps. We study this form of probabilistic coherence in Jev and the English Laya checkpoint, using 2,500 matched examples per system across TREC, CLINC150, and MASSIVE. Across 72,000 classification questions, we compare direct fine-label predictions with broad-category probabilities and predictions reconstructed through those categories. Both systems show substantial disagreement: mean category-level total variation ranges from 0.219 to 0.349 for Jev and from 0.424 to 0.689 for Laya, on a scale where zero means exact agreement. The consequences differ sharply. On CLINC150, reconstruction reduces Jev's accuracy by 22.9 percentage points (paired 95% bootstrap interval: (-24.9, -20.9)) and improves Laya's by 21.3 points ((18.0, 24.5)). The same directions hold across all three datasets, with all six unadjusted accuracy-change intervals excluding zero. Improved accuracy can also accompany less reliable confidence: on MASSIVE, Laya gains 9.2 accuracy points while its expected calibration error rises from 0.046 to 0.124. Error analysis identifies both broad-category mistakes and within-category confusions. These findings show why decision systems need joint evaluation of accuracy, confidence calibration, and probability coherence in the workflow used by an application. --- ### On the Token Value Inequality in Efficient Reasoning Authors: Runjia Zeng, Hang Hua, Yiyang Liu, Zhiqiang Tao, Ruixiang Tang, Qifan Wang, Cheng Han, Dongfang Liu Published: 2026-09-27 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.33970v1 Chain-of-Thought reasoning has enabled large language models to achieve substantial performance gains on complex tasks. However, these gains come at the cost of dramatically increased token consumption. This raises a fundamental question: is every token in the reasoning trace equally valuable? We present a diagnostic and optimization framework grounded in a key empirical finding: the value of tokens within a CoT reasoning sequence is highly non-uniform, and this non-uniformity can be effectively characterized by token-level log probability signals. We show that normalized log probability helps distinguish core tokens, which carry structural and decisive reasoning content, from redundant tokens, which are exploratory, low-confidence filler that contributes less directly to the final answer. Building on these findings, we formulate the TokenProbe framework around two empirical findings and one claim: findings identify token value inequality first and then establish TokenProbe as a core-token proxy, and the claim introduces an efficient GRPO objective positing that selectively compressing redundant tokens can yield Pareto improvements in the accuracy-token efficiency space. Empirically, our method preserves reasoning quality while reducing the token usage by 76% of the baseline. Under matched reasoning-length budgets, we show that it can even outperform strong flagship baselines like Gemini-3.1-Pro. Homepage: https://runjia.tech/tokenprobe/. --- ### Gaussian Splatting-based Volumetric Video Compression with Sparse 4D Anchors Authors: Ge Gao, Siyue Teng, Chanqgi Wang, Fan Zhang, Nantheera Anantrasirichai, Jui Chiu Chiang, Wen-Hsiao Peng, David Bull Published: 2026-09-27 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.33969v1 Immersive video communication requires photorealistic, render-efficient, and compact dynamic scene representations. 3D Gaussian Splatting (3DGS) offers a promising representation, but dynamic 3DGS remains difficult to compress due to dense primitives and spatiotemporal redundancy. Anchor-based formulations improve compactness with sparse scaffolds that share geometry and appearance across primitives. However, existing designs often rely on deforming a single canonical scaffold and condition each primitive on its associated anchor in isolation, limiting their ability to handle non-local dynamics and disocclusion while under-exploiting inter-anchor correlations, particularly in motion- or texture-dense regions. To address these limitations, we propose SAGA, a volumetric video codec built upon Sparse Anchor-assisted GAussian splatting representations. SAGA represents dynamic 3D scenes using hierarchically organized sparse 4D anchors, where coordinate-based INR decoders generate fine anchors and Gaussian primitives from inter-anchor interpolations, enabling compact parameter sharing across spatiotemporal structures. For long-range dependencies among unstructured anchors, we further introduce fixed-size memory slots with orthogonality-informed updates for accurate entropy-context modeling. Experiments show that SAGA achieves strong rate-distortion performance against GIFStream, with PSNR BD-rate reductions of 80.39% and 83.94% on Neu3D and MPEG MIV, respectively. --- ### Two-Sample Testing for Inhomogeneous Random Graphs in Non-Integral $L_r$ Norms Authors: Soham Dan Published: 2026-09-27 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2609.33968v1 Testing whether two populations of networks share the same edge probabilities is a basic problem in network inference. How hard it is depends on the norm used to measure the difference. For the inhomogeneous Erdős--Rényi (IER) model, the optimal sample complexity is known for every integer $L_r$ norm and for $1\le r<2$. For non-integral $r>2$, however, the known upper and lower bounds do not match, and the lower bound was conjectured to be tight. We study this gap for two-sample testing on aligned vertices. We propose a test that runs two published statistics, of orders $2$ and $\lceil r\rceil$, on the same data and rejects if either one rejects. Its thresholds come from Hölder interpolation, so that both statistics have the same sample cost. We prove that this test attains the conjectured rate. Combined with earlier results, this shows that for every fixed $r\ge1$ the minimax sample complexity is of order $n^{\max\{4/r-1,\,2/r\}}/ε^2$, even when the separation changes with $n$. In simulations with $n$ between 32 and 256, the number of graphs needed for 80\% power at level $0.05$ grows with $n$ at a rate consistent with the theory. For $r=2.5$, for example, the fitted exponent is $0.78$, against the theoretical value $0.8$. Interestingly, the two statistics split the work as the interpolation argument suggests: the higher-order statistic is more powerful when only a few edges change, and the $L_2$ statistic when many edges change. --- ### ThinkNet: Compact Architecture Selection and Validation-Gated Ensembles for Subject-Independent MI-EEG Decoding Authors: Abdul Basit, Saim Rehman, Muhammad Shafique Published: 2026-09-27 Categories: cs.HC, cs.AI, cs.LG, eess.SP arXiv: https://arxiv.org/abs/2609.33967v1 Practical assistive and rehabilitative brain--computer interfaces require subject-independent motor-imagery EEG (MI-EEG) decoders that generalize to new users under limited target-user data and constrained compute. However, held-out-subject performance can be overstated when test-subject information influences preprocessing, model selection, or ensemble selection. We present \textit{ThinkNet}, a validation-controlled framework that combines train-only normalization, validation-guided evolutionary search, and validation-gated inference to identify compact decoders and inference policies for held-out subjects. We evaluate four-class BCI Competition IV-2a (session T) decoding with nine Leave-One-Subject-Out (LOSO) folds, three seeds, seven fixed decoder entries, and a broader search over ten representative decoder families; the held-out subject is never used for normalization, hyperparameter, architecture, or ensemble-policy selection. In the fixed benchmark, the validation-selected compact decoder achieved 44.35$\pm$15.41\% accuracy with 4.9K parameters, 19 KB FP32 weights, and 0.99 ms batch-1 Orin CUDA inference. Across the broader search, compact models ($\leq$25K parameters) achieved higher mean held-out accuracy than mid-size and large alternatives after selected retraining (40.10\% vs. 35.09\% and 34.78\%). Validation-gated ensembling improved over validation-selected single-model inference, reaching 43.98$\pm$16.25\% in the fixed benchmark and 43.31$\pm$15.88\% for the compact six-family ensemble. A non-deployable oracle analysis revealed a 6.1-point family-selection gap and near-zero validation--test correlation, showing that validation reliability remains a key bottleneck under subject shift. Thus, ThinkNet is a validation-controlled framework for compact MI-EEG model and inference-policy selection, rather than a single-architecture benchmark. --- ### Greenpixie's AI Token Methodology: Assessing the Energy, Water and CO2-eq Impact of AI Tokens for Open and Closed Weight Models Authors: Joshua Horswill, Ross Hunter, Matt Clifford, James Hall Published: 2026-09-27 Categories: cs.CY, cs.LG arXiv: https://arxiv.org/abs/2609.33965v2 We describe a methodology for estimating the per-token energy cost of cloud-hosted large language model (LLM) inference, separating between input (prefill) and output (decode) tokens. Graphics processing unit (GPU) energy usage is measured during inference benchmarking with open-weights models on a wide range of text-based tasks. The remaining server energy contribution from non-GPU hardware is estimated from the inference wall time. Bayesian linear regression is used to model the relationship between energy per token and LLM size, request traffic, and hardware deployment configuration. Proprietary frontier LLMs of unknown size and deployment are binned into size buckets based on naming conventions and performance priors, and the space of possible LLM configurations is sampled with Monte-Carlo methods to give a representative average energy per token and uncertainty. We also describe how these energy measurements can be used to estimate the carbon-dioxide equivalent ($\mathrm{CO_2\text{-}eq}$) emissions, both usage and embodied, and water consumed per token of AI inference. This methodology provides actionable data that enables reductions in cost, electricity usage, $\mathrm{CO_2\text{-}eq}$ emitted and water consumed in cloud and Software as a Service (SaaS). --- ### SymNetPro: LOS-Aware Directional Multi-Transmitter Localization from Sparse Radio Observations Authors: Lyuzhou Ye, Heng Fan, Yan Huang Published: 2026-09-27 Categories: cs.IT, cs.CV arXiv: https://arxiv.org/abs/2609.33964v1 Directional multi-transmitter localization from sparse received-power observations is difficult because the receiver observes only the source-unresolved aggregate field: multiple directional sources superpose, building blockage fragments their visible regions, and stronger sources can mask weaker ones. We present SymNetPro, which retains the dual-task radio-map reconstruction and localization backbone of SymNet and adds two targeted components. First, a sparse line-of-sight (LOS)-aware attention bias injects obstruction-aware spatial relations into selected token interactions. Second, transmitter-drop augmentation recomposes training scenes after removing one sample-supported transmitter, exposing the model to controlled source-cardinality variation. Experiments on directional ray-traced urban environments show substantially lower OSPA than representative localization baselines under extreme sparse sampling, with consistent gains under measurement noise and increasing transmitter count. A transmitter-specific evidence analysis further shows that remaining misses concentrate in regimes where the target contributes little distinguishable power to the aggregate observation. --- ### EEG-Fusion: Failure-Informed Source-Free Expert Routing for Robust Motor Imagery EEG Decoding Authors: Abdul Basit, Saim Rehman, Muhammad Shafique Published: 2026-09-27 Categories: cs.HC, cs.AI, cs.LG, eess.SP arXiv: https://arxiv.org/abs/2609.33962v1 Subject-independent motor-imagery (MI) EEG decoding can exhibit subject-level failures even when average performance appears acceptable: under subject shift, a decoder can become an overconfident near-one-class predictor. This is especially problematic in source-free deployment, where target-user labels are unavailable during adaptation and expert selection. We present \textit{EEG-Fusion}, a failure-informed decision-level fusion framework that treats source-free MI decoding as label-free reliability estimation over heterogeneous experts. EEG-Fusion applies subject-wise Euclidean alignment and normalization-only test-time adaptation, then routes each target subject to a neural, covariance-based, or physiological-feature expert using a reliability gate trained on source-held-out folds to predict expert performance and collapse risk from label-free stream diagnostics. The gate uses confidence, entropy, prediction diversity, expert agreement, and predicted class balance; collapse is measured as the maximum predicted class fraction. In 9-fold leave-one-subject-out (LOSO) evaluation with three seeds, relative to a no-alignment raw EEGNet source-free anchor, EEG-Fusion improves subject macro-F1 from 0.417 to 0.529 on BCI IV-2a local protocol, from 0.314 to 0.482 on BNCI2014-001, and from 0.607 to 0.708 on BNCI2014-004; corresponding collapse-index reductions are 0.199, 0.227, and 0.169. In a 9-subject Cho2017 external subset, EEG-Fusion improves macro-F1 from 0.516 to 0.630. These results suggest that label-free reliability estimation can reduce subject-level failure modes in source-free MI-EEG deployment. --- ### Designing Reliable LLM-as-a-Judge Measurement Systems for Multi-Turn Business Agents Authors: Kaiwen Luo, Ming Gao Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33955v1 Many LLM-as-a-judge evaluations score fixed outputs under a fixed task definition. Production multi-turn business agents instead require a maintained measurement system: correctness depends on business-specific facts and procedures, outcomes emerge across turns, and failures must be attributed to either agent capability or missing business knowledge before they are actionable. We present an integrated methodology spanning evaluation specification, modular LLM judges, intent-preserving user simulation, and human-in-the-loop governance. The specification defines conversation-level end states and actionable failure ownership. Atomic judges share versioned evidence and feed an explicit aggregation graph. The simulator is released only after task-preservation and stability checks. Independent human audits estimate measurement fidelity, renew tiered reference sets, and route disagreements to label correction, guideline revision, or judge improvement. Production studies show that system-level fidelity improved across repeated audits, that human reviewers and automated judges improved together under the shared feedback loop, and that their combined workflow had the strongest descriptive performance in both reported task-completion settings. Because the studies are observational and the human reference itself required revision, these findings demonstrate operational usefulness rather than causal or universal superiority. The contribution is a practical framework for making multi-turn agent measurement reliable, actionable, and maintainable as the evaluated system and its evidence evolve. --- ### Simple Diffusion Language Models Are More Effective Few-Step Generators Than Reported Authors: Hasan Amin, Ming Yin, Rajiv Khanna Published: 2026-09-27 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.33947v1 Diffusion language models (DLMs) promise fast parallel generation, yet high-quality samples often require large number of refinement steps, which diminishes their advantage in practice. This has led to massive interest in and rapid development of new methods for effective few-step generation. We show that much of the supposed quality gap at few steps can instead arise from a suboptimally configured sampler. Modest sampler sharpening, without any model retraining, enables a couple years old masked DLM to rival supposedly far improved successors. This differently sampled DLM in fact achieves lower generative perplexity in just 16 steps than what its standard sampler obtains with 1024, while improving both judged quality and semantic diversity. We further show that conventional per-output metrics can fundamentally obscure these gains, since any optimal trade-off between two such metrics can be attained by a generator supported on at most two outputs. We subsequently introduce GroupEval, which separately evaluates quality and across-output semantic diversity, and offers fresh insights including uncovering how 1.5-4.7x perplexity gains of a distilled model yield no corresponding quality gain. Finally, we explain why sharpening helps: parallel unmasking destroys dependencies among simultaneously generated tokens, creating a gap between prediction and generation. We prove that pervasive temperature choice of one is generically suboptimal under parallel sampling even for an exact denoiser, and that worse predictions can yield better samples. Through these results, we argue for a broader evaluation principle of treating the deployed generator as the object of comparison, benchmarking it against tuned baselines, and assessing quality and diversity jointly and with more human-aligned measures. --- ### Behavioral Monitoring of JEPA World Models with Jacobian Centroids Authors: Thomas Walker, Randall Balestriero, Richard Baraniuk Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33940v1 Detecting failures in World Model (WM)-based planning requires monitoring whether the model is behaviorally aligned with the current task, which in turn requires studying its internal representations. Here, we show that centroids---sub-component Jacobian row-sums---effectively identify the behavioral properties of WMs, complementing traditional activation-based knowledge signals. The centroids of a model are easily computed through Jacobian vector products and characterize how the model organizes the geometry of its input space, yielding an efficient perspective on internal representations, including the generation of task-relevant saliency maps. Evaluated on continuous control tasks using JEPA WMs, this behavioral view reveals a structural dissociation, where the encoder correctly represents the goal while the predictor remains behaviorally unresponsive. This failure mode directly predicts planning failure before any action is taken, allowing for goal resampling to recapture out-of-distribution success. Moreover, centroid-based methods outperform baseline methods as distribution-shift detectors. Together, these tools yield a behavioral monitoring stack that is operational and consequential under distribution shifts. --- ### EpiTransfer: Sparse, Training-Free Long-Range Depth Estimation from Temporal Monocular Aerial Frames Authors: Diksha Aggarwal, Rutvik Dagadkhair, Sanjana Srivastava, Bradley Denby, Kevin Kochersberger Published: 2026-09-27 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2609.33939v1 Reliable 3D spatial understanding is essential for autonomous navigation, obstacle avoidance, and scene reconstruction. While state-of-the-art learned depth estimation techniques achieve high accuracy in-distribution, they often generalize poorly to novel viewpoints and altitudes. This paper presents a geometrically derived, training-free depth estimation method using epipolar transfer with only two monocular images and camera pose estimates. By leveraging camera motion to synthesize a virtual stereo pair with a freely chosen baseline, our approach transforms temporal correspondence into a stereo triangulation task while mitigating geometric degeneracies inherent to direct two-view triangulation. Validated across outdoor drone flights (to a maximum range of approximately 90\,m) and indoor OptiTrack environments against LiDAR ground truth, the method achieves an indoor AbsRel of 0.092 and $δ< 1.25$ of 0.940, comparable to direct triangulation (AbsRel 0.073) while retaining valid depth over a larger fraction of challenging scenes, and substantially outperforms off-the-shelf learning-based baselines such as ZoeDepth (AbsRel 0.225) and Depth Anything V2 (AbsRel 0.570), which are not trained or fine-tuned for this domain, with no training data required. --- ### Test-Time Generalized Category Discovery Authors: Shambhavi Mishra, Omprakash Chakraborty, Julio Silva-Rodriguez, Ismail Ben Ayed, Marco Pedersoli, Jose Dolz Published: 2026-09-27 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.33937v1 Test-Time Adaptation (TTA) and Generalized Category Discovery (GCD) are traditionally treated as disjoint problems: the former adapts models to domain shift assuming all test classes are known, while the latter discovers novel categories assuming labeled training data for known classes. However, real-world deployment rarely fits either setting. Motivated by this gap, we introduce Test-Time Generalized Category Discovery (TT-GCD), a unified and more realistic scenario where a vision-language model must adapt to distribution shifts, classify known categories using only textual supervision, and discover novel categories, all during test time and without access to labeled data. To address this challenging scenario, we propose PACT (Prototype Assignment for Category discovery at Test time), a fully unsupervised framework that casts known-class recognition and novel-class discovery via prototype assignment. PACT first re-aligns shifted visual features with the text-derived class representations of the VLM using confident zero-shot predictions. Known and novel categories are then both represented by prototypes in the visual embedding space, estimated from the unlabeled test stream, and each test image is assigned to the category whose prototype is most similar to its visual feature. Extensive experiments across corruption and domain-shift benchmarks demonstrate that PACT outperforms adapted state-of-the-art TTA and GCD methods, effectively bridging the gap between adaptation and discovery. --- ### How Strong Is the Evidence for the Artificial Hivemind? Reevaluating Evidence for the Open-Ended Homogeneity of Language Models Authors: Rylan Schaeffer, Brando Miranda, Joshua Kazdan, Jessica Chudnovsky, Sanmi Koyejo Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33936v1 Recent research argues that language models exhibit pronounced homogeneity in open-ended generation, framing such behavior as an Artificial Hivemind that poses a long-term threat to human creativity. We examine three of its central results. First, the flagship example is that model responses to "Write a metaphor involving time" collapse into two clusters. Visualization, spectral analysis, clustering, and language model labels all contradict this description. The labels record each response's vehicle, what it compares time to. Our responses and the original authors' own show one dominant vehicle plus a heavy tail of distinct minority vehicles. "Time" is one of our least diverse topics, so the example is a favorable case, not a representative one. Second, the paper measures homogeneity against an undemanding null: responses to unrelated prompts. Under a more demanding null (same-prompt responses expressing genuinely different ideas), 20%-32% of such pairs already exceed the paper's 0.8 convergence threshold. A residual effect survives this null. The paper's same-prompt pairs exceed 0.8 roughly two to three times as often as our different-idea pairs. Much of what the paper calls homogeneity is the shared geometry of answering the same prompt. The remaining measurements lack any null: no human baseline is collected, and the model-indistinguishability statistic has no null. Third, the paper concludes that inference-time interventions are inadequate for combating the Artificial Hivemind, writing that "more generalizable solutions are needed at the model training level." We show that this conclusion is unsupported in three ways, and that an inference-time intervention (prompting) reliably raises measured response diversity. We do not resolve whether the Artificial Hivemind is real. We show that the published evidence does not establish it. --- ### Video, Ergo Genero: Unifying Video Tasks via Spatiotemporal Analogy Authors: Chia-Hsiang Kao, Belinda Zeng, Bharath Hariharan, Menglin Jia Published: 2026-09-27 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.33935v1 Adapting video models to new tasks typically requires dedicated data curation and fine-tuning. While visual analogy provides a training-free alternative by specifying tasks in-context, it remains restricted to the image domain. To explore whether analogy-based methods can unify diverse video tasks and generalize to out-of-distribution scenarios, we introduce ViGeo, a framework that extends visual in-context learning to the video domain via spatiotemporal canvas completion. Evaluated on a diverse task taxonomy with a strict train-test split, ViGeo generalizes to unseen video manipulations and zero-shot modalities (e.g., event cameras). Finally, we identify task internalization, where a query format associated with a pretrained task overrides the demonstration, and show that this shortcut can be removed with a small amount of task-unrelated data, highlighting the need to decorrelate prompt format from task identity. --- ### ArticulateArena: A Metric for Articulated Kinematics Authors: Yumeng He, Yongfei She, Huanyu Chen, Chun Yuan, Peihao Li, Joseph Masterjohn, Yin Yang, Ying Jiang, Chenfanfu Jiang Published: 2026-09-27 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2609.33931v1 Modern methods reconstruct or generate simulation-ready articulated objects, predicting not only their geometry but also how their parts are connected and allowed to move. Evaluating the geometry is straightforward, but evaluating the predicted articulation is not, because articulation specifies a motion rather than a shape, and there is no agreed distance between two motions. More specifically, existing protocols score joint type, axis direction, origin, and motion limits separately, although these parameters jointly describe a single physical motion, and the same motion can be written as different parameter values. As a result, a joint can score maximally wrong against an equivalent encoding of itself, and several component errors are ill-conditioned or undefined exactly where predictions become accurate. We propose ArticulateArena, a representation-invariant counterpart of Chamfer distance for articulation that compares the motions one-DOF joints induce rather than the parameters that encode them. It represents each joint by the unordered pair of its Lie-algebra endpoint twists, and we prove that the resulting quotient distance is a metric. It unifies fixed, revolute, prismatic, and helical joints, brings continuous joints into the same score through a compactification, and reads as the RMS motion of the moving part in meters when weighted by its mass distribution. A motion-aware tree edit distance lifts the metric to full kinematic trees, pricing structural errors such as spurious or missing joints in the same motion units as joint errors, and for a fixed inner product it remains a metric on trees up to relabeling. Alongside the metric we release ArticulateArena-20K, a new suite of 19,977 articulated objects with verified kinematics, and we re-evaluate published reconstruction methods on it under the new metric. Project page: https://heyumeng.com/ArticulateArena-web/ --- ### Diffusion-Based Rollouts as a Stabilization Mechanism for Long-Horizon Environmental Forecasting Authors: Marina Vicens-Miquel, Amy McGovern, Aaron J. Hill, Efi Foufoula-Georgiou, Samuel S. P. Shen Published: 2026-09-27 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.33930v1 Extending forecast lead times while maintaining predictive skill remains a major challenge in environmental forecasting. We investigate diffusion-based rollouts as a stabilization mechanism for recursive forecasting using low-dimensional water-level time series and high-dimensional precipitation fields. Across both modalities, diffusion suppresses recursive error growth, with the largest stabilization occurring where deterministic rollouts are most unstable. However, stabilization does not guarantee forecast fidelity. In the water-level experiments, forecasts progressively lose event-level fidelity as the rollout loses access to external predictive information, and trajectory-level comparisons show that diffusion can remain numerically stable while contracting toward central values and exhibiting reduced variability. In the precipitation experiments, which retain conditioning from numerical weather prediction throughout the rollout, diffusion better preserves spatial organization and event-detection skill. Together, these contrasting experiments indicate that diffusion can control recursive error amplification, while its practical benefit also depends on the predictive information available to constrain future evolution. --- ### Preserving DEG Rankings for Gene Discovery in Histology-Based Spatial Gene Expression Prediction Authors: Kaito Shiku, Kazuya Nishimura, Yasuhiro Kojima, Ryoma Bise Published: 2026-09-27 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.33928v1 Predicting spatial gene expression from histology images could scale spatial transcriptomics (ST) to image-only cohorts, but conventional histology-based ST prediction is trained and evaluated mainly by per-gene spatial-profile reconstruction. This objective is misaligned with a key downstream use of ST: differentially expressed gene (DEG) discovery, where genes are ranked for a biological or morphology-defined contrast by evidence of between-group expression differences. We formulate image-based differential expression ranking (IDER), which asks whether predicted expression profiles preserve the contrast-specific ranked gene list obtained from measured profiles. IDER compares gene rankings induced by differential-expression statistics, rather than raw expression magnitudes or per-gene spatial correlations. We further introduce a differentiable IDER objective that aligns these statistics across genes and can be trained with morphology-derived proxy contrasts without predefined biological group labels. Experiments on public ST datasets show improved DEG-ranking agreement and pathway-enrichment overlap over conventional reconstruction objectives, including morphology-derived and pathologist-annotated tissue-region evaluations. --- ### Optimizing the Phi-2 Small Language Model for Real-time Chatbot Applications Using Parameter-Efficient Fine-Tuning (PEFT) with QLoRA Quantization Authors: PhanTan Khanh Nguyen, Ashfaq Ali Shafin, Khandaker Mamun Ahmed Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33927v1 This study explores the optimization of the Phi-2 Small Language Models (SLMs) for real-time chatbot applications through Parameter-Efficient Fine-Tuning (PEFT) and Quantized Low-Rank Adaptation (QLoRA). QLoRA specifically refers to the integration of PEFT with LoRA alongside a 4-bit quantization process, aimed at enhancing computational efficiency. These models, initially designed for high performance with minimal computational overhead, are further refined to address the constraints of mobile and edge computing environments. By integrating PEFT with QLoRA, the research aims to reduce memory usage significantly while maintaining, or potentially improving, the accuracy of model responses in real-time interactions. The effectiveness of these techniques was evaluated using the ROUGE metric system, which showed notable improvements in the summarization tasks performed by the models. This approach not only confirms the feasibility of using SLMs in resource-restricted environments but also opens up new avenues for deploying advanced AI-driven applications in real-time settings. The study's findings have significant implications for the development of efficient, scalable, and accessible AI technologies, paving the way for broader adoption in various industries. --- ### Quantization Error Is Spectrally Flat: A Single Random Probe Is a Calibrated, Data-Free Sensitivity Estimator, with Application to Budget-Targeted Mixed-Precision Quantization Authors: I Kennedy, T Kennedy Published: 2026-09-27 Categories: cs.CL, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.33923v2 A single random Gaussian probe gives an unbiased estimate of the squared Frobenius norm of a layer's quantization error. The estimator is well-behaved because round-to-nearest error is spectrally flat. Across 1,683 tensors from a 35B MoE and a 9B dense model, effective dimensionality is 0.93 to 0.96 times the i.i.d. noise value of the same shape, and on the MoE the median is unchanged from 2-bit to 8-bit. The probe coefficient of variation is predictable from tensor shape. One probe measures per-tensor sensitivity to within 4 to 7%; twenty probes reach 1.3 to 1.4%.RAM applies the propagated form of this estimator to budget-targeted mixed-precision quantization with no calibration data. Gaussian probes carrying the network's own input statistics score every tensor at six bit-widths. A knapsack solver allocates bits under an exact byte budget, with guardrails against catastrophic 2-bit assignments. One probe pass serves any budget. Isolated and propagated scores rank tensors independently on Qwen3.5-35B-A3B (Spearman -0.01), yet the propagated probe rank-correlates 0.81 to 0.83 with the GPTQ layer objective from real activations, while the isolated estimator is uncorrelated with it. That objective is the wrong allocation target: at matched bytes on Qwen3.8-27B, a block-output probe beats a vendor IQ3_M mix and an oracle that allocates from the real-activation objective. On Qwen3-8B the propagated probe ties HAWQ-V2 at matched bytes. Across seven architectures from 8B to 122B, with probe timing up to a 400B model in nine minutes on one workstation, RAM reaches 3.5 to 13.6% lower median WikiText-2 perplexity than size-comparable uniform 4-bit builds on the tested MoE models. (Black Sheep Ai baa.ai) --- ### HyperMCTS: Hypergraph-Augmented MCTS for Long-Horizon LLM Agents Authors: Tingsong Xiao, Nithish Balachandar Moudhgalya, Chandrayee Basu, Lichao Wang, Luyang Kong, Benjamin Z. Yao, Zhe Jiang, Jie Hao Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33920v1 Long-horizon tasks require large language model (LLM) agents to coordinate decisions under constraints that span an entire solution. Monte Carlo Tree Search (MCTS) offers a promising approach to test-time scaling by exploring alternative action trajectories, but model computation and environment interaction make search costly. Efficient search therefore requires effective reuse of trajectory feedback. Standard MCTS maintains prefix-specific statistics, without explicitly accumulating outcomes for decision groups that recur across different paths. To fill this gap, we propose HyperMCTS, a training-free method that augments an ordered MCTS tree with a cross-trajectory hypergraph. Hyperedges represent groups of canonical decisions and accumulate their observed returns within the current task. Our hypergraph-guided HyperUCT selection rule aggregates evidence from overlapping hyperedges into an action prior, allowing outcomes collected under one prefix to inform selection under another while preserving execution histories in the tree. On DeepPlanning, HyperMCTS improves average planning accuracy by 2.3--7.3 percentage points over the strongest baseline for each of three backbone models. It enables Qwen3.6-27B to outperform Claude Opus 4.6 (max) on Shopping Planning, while achieving higher accuracy with fewer LLM calls and output tokens than the evaluated MCTS-based baselines. SealQA experiments further demonstrate improvements in question answering. --- ### Validating Memory-Optimal Transformer Kernels on Real Hardware: From Formal Derivation to Measured Performance Across Two HPC Clusters Authors: Lenore M. Mullin, Gaetan Hains Published: 2026-09-27 Categories: cs.DC, cs.AI arXiv: https://arxiv.org/abs/2609.33916v1 We validate memory-optimal cost functions for transformer kernels derived via the Mathematics of Arrays (MoA). Companion Papers I-IV formally derive kernels for attention forward, backward, fused forward+backward, decode, and the complete block (RMSNorm, gated MLP) as a hardware-independent specification (DNF) transformed to a machine-specific realization (ONF) via gamma, with verification to machine precision against PyTorch. This paper checks those predictions against measured performance on two HPC clusters (Purdue Anvil, NCSA Delta) across CPU and GPU. Three results stand out. (1) We identify and fix a GPU regression: fusing forward+backward, proven to avoid materializing an O(n^2) intermediate, initially ran slower than naive on GPU due to atomic contention. Profiling confirmed 2.00x more atomic instructions; a targeted ONF rewrite reversed it, yielding up to 2.5x speedup. (2) Identical derivations produce markedly different real costs by topology: 535x NUMA-locality penalty on one cluster vs <3x oversubscription on another, showing optimal deployment is a function of the machine's array structure. (3) We report a partially resolved anomaly: identical denotational computations run faster in C than Fortran on CPU but faster in Fortran than C on GPU, narrowed to one dominant kernel and one memory-latency stall mechanism (3.17x time gap matches 3.35x stall gap). We treat hardware-specific optimization as a routine ONF rewrite with fixed, verified DNF, a candidate methodology for scaling AI onto evolving hardware without re-deriving correctness. --- ### Training Witnesses: Trusting the Training without Trusting the Trainer Authors: Houjun Liu, Pratyusha Sharma Published: 2026-09-27 Categories: cs.LG, cs.CR arXiv: https://arxiv.org/abs/2609.33915v1 Progress in machine learning cannot outpace our ability to verify it. With an explosion in papers today, every scientific claim rests initially on trust in the trainer, leading to uneven evaluation, baselines, and forestalling of reliable progress. Traditionally, the burden of verification falls on the reader, who must reproduce expensive training runs. This strategy is impractical due to an explosion in slop contributions, diversity of methods, and the sheer compute required. We put the burden of proof where it belongs, on the trainer, and in the process also cut the overall cost of verification significantly. We introduce Witnesses, a method for certifying training, data usage and evaluation in a neural network training run. Our key insight is that fast behavioral fingerprints with occasional replay challenges are sufficient for auditing neural network training. Our method is applicable at scale with minimal overhead to the trainer, is cheap for the verifier, rejects bad training runs with amplifiable probability, and allows for exact queries of both data inclusion and exclusion. We test our method on language model training runs from 100M to 2B scales, across DDP and FSDP, and demonstrate this minimal overhead. We also introduce a self-regulating leaderboard of "auto-certified" training runs that enables shared baselines and progress. We invite the community to participate in the leaderboard to improve reproducibility in machine learning. --- ### When Consent Outlives Context: Residual Authority Replay in Long-Lived Agents Authors: Zhihao Zhang, Chao Wang, Rujia Li, Qingze Wang, Xiaoyan Sun, Jun Dai Published: 2026-09-27 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.33910v1 LLM agents increasingly rely on user approval to authorize security-sensitive actions at runtime. Such approvals are granted within a specific task and execution context. In long-lived agents, authorization decisions may need to persist across tasks or sessions. We find that this continuity can outlive the context that originally justified the approval, creating residual authority reusable without renewed consent. We expose this failure mode through a longitudinal attack that starts from a target security-sensitive action, identifies the authority required to execute it, induces benign interactions that legitimately obtain that authority, and later replays the residual authority during adversarial execution. Across controlled and live settings, we demonstrate that residual-authority replay arises in practice and substantially increases the success of prompt-injection and context-rebinding attacks. We evaluate 508 AgentDojo attack cases across six LLM families using production-derived authorization semantics. With residual authority, attack success rate (ASR) increases by up to 35.1 percentage points compared with a fresh authorization state. In live context-rebinding attacks on 55 Terminal-Bench cases across three real-world production coding agents, residual-authority replay increases ASR by 24.9 percentage points on average. These findings expose a fundamental mismatch between persistent authorization and the contextual nature of user consent in long-lived LLM agents. --- ### JIVE: Jacobian-Informed Volume Expansion for Diverse Generative Sampling Authors: Guangxun Zhang, Brian Cai, Boxuan Zhang, Chao Chen, Ruixiang Tang Published: 2026-09-27 Categories: cs.LG, cs.AI, cs.CV arXiv: https://arxiv.org/abs/2609.33906v1 Generative models often suffer from mode collapse and limited sample diversity. While prior works attempt to mitigate this by jointly generating a batch of samples and repelling their trajectories, these heuristics do not explicitly maximize the diversity of the resulting endpoints. We introduce JIVE, a training-free framework that enhances generative diversity by injecting velocity perturbations aligned with the leading right singular subspace of the generator's endpoint Jacobian. By leveraging this local geometric structure, JIVE provably maximizes endpoint diversity while preserving sample quality. To maintain practical efficiency, we compute these perturbation directions via matrix-free iterations rooted in classical numerical linear algebra, requiring only a small computational overhead. Across different benchmarks, JIVE boosts both pixel and feature-level diversity in few-step and one-step generation. --- ### SlopBench: How Well Can We Rank Language Models by Slop? A Multi-Domain Benchmark of Repetitive AI Writing Authors: Dhruv Roongta, Harsha Gaddipati, Anh Tuan Huynh Published: 2026-09-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.33905v1 SlopBench asks which models produce the stiff, repetitive prose readers call AI slop, a question detectors leave open once they have classified a text as machine-written. We evaluated eighteen models on 112 hand-written tasks in email, social posts, essays, and workplace chat, sampling each model on each task up to ten times, for 19,928 outputs in all. SlopBench scores four surface behaviors a reader can check by hand: length against the word band each task specifies, opener repetition across a model's own samples of one task, and paragraph rhythm and fixed lexical constructions against pre-ChatGPT human corpora. Under one fixed weighting, Kimi K2.6 scores lowest at 21.1 and Mistral Large highest at 40.6. Across 500 random reweightings Kimi has the lowest score in 58 percent of draws and Mistral the highest in 97 percent. No draw preserves the full order of the eighteen, and a scenario bootstrap leaves exactly one of those ranks unambiguous. We ran three further checks on that middle order: a crowd arena, an AI detector, and lexical diversity. None of them confirmed the order. We therefore report the four behaviors separately and treat the composite as one weighting among many, and we release the prompts, outputs, reference statistics, and scoring code. --- ### On the Two Faces of Adam in Separable Linear Classification Authors: Chen Fan, Csaba Szepesvári Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33904v1 We consider the behavior of deterministic, full-batch, bias-corrected Adam in separable linear classification with softmax parametrization under log-loss. In this setting, under a wide range of conditions Adam is known to approach max-norm-margin optimality when its stability constant $ε$ is zero, while with a positive $ε$, it is known to approach Euclidean-margin optimality. Our main contribution is the quantitative description of Adam's behavior for small fixed positive $ε$. We give sufficient conditions under which an Adam-trained classifier nearly maximizes the max-norm margin before the updates become gradient-like. We also show that the classifier reaches a fixed target Euclidean margin only much later. Specifically, we show that for polynomially decreasing stepsizes with exponent \(a\), where \(1/3<a<1\), the updates become approximately proportional to the negative gradient after $Θ(\log(1/ε)^{1/(1-a)})$ iterations. At that time, the classifier still nearly maximizes the max-norm margin. Reaching a fixed target Euclidean margin above that of every max-norm-optimal classifier, but below the optimum, is shown to require $ε^{-Θ(1)/(1-a)}$ iterations. Under inverse-linear stepsize decay (\(a=1\)), the update transition takes polynomially many iterations, whereas reaching the target margin takes exponentially many. Experiments support these predictions. The later change in the classifier can improve or worsen generalization after training error reaches zero, connecting the analysis to grokking and its reverse. --- ### Finite Probes Suffice: Identifiability and Universality for Weight-Space Learning Authors: Soutrik Sarangi, Yonatan Sverdlov, Adir Dayan, Haggai Maron, Nadav Dym Published: 2026-09-27 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.33901v1 Learning properties of neural networks has recently attracted growing interest, with existing approaches operating either directly on network parameters or through probe-based representations of network behavior. While probing methods have shown strong empirical performance, their theoretical foundations remain limited. In this work, we study when finite probe-based representations are sufficient for learning neural functionals. We establish general identification and universality results for probing, and show that using intermediate hidden representations can provide significantly more informative representations than relying only on final outputs. Motivated by these results, we introduce HIDDENPROBE, a simple architecture for learning from hidden probe responses. Across a range of neural functional benchmarks, including both MLPs and Transformers, HIDDENPROBE consistently improves over existing probing methods and achieves state-of-the-art performance. Our code is publicly available on GitHub. --- ### NSV-Shift: A Contrastive Benchmark for Non-Speech Vocalization Understanding and Response Adaptation in Speech-to-Speech Models Authors: Ziwei Chen Published: 2026-09-27 Categories: cs.CL, cs.HC arXiv: https://arxiv.org/abs/2609.33899v1 We introduce NSV-Shift, a contrastive benchmark for evaluating whether speech-to-speech models can understand non-speech vocalizations (NSVs) and adapt their responses accordingly. Each pair contains two conversations with identical lexical content that differ only in the NSV embedded in the final turn. Our pilot contains 22 human-verified pairs (44 audio conditions) and evaluates five models on NSV perception, emotion understanding, and response adaptation. Results show that models generally perform better at detecting NSVs than at interpreting their fine-grained emotional meaning or producing appropriately differentiated responses. The data construction pipeline, dataset, and evaluation pipeline are publicly available at https://github.com/ChenzwNina/nsv-construction. --- ### No Free Efficiency: Revisiting the Trade-off Between Training Efficiency and Model Vulnerability Authors: Yiyong Liu, Jun Sakuma, Michael Backes, Rui Wen Published: 2026-09-27 Categories: cs.LG, cs.CR arXiv: https://arxiv.org/abs/2609.33898v1 Training efficiency has become the central driver of recent progress in foundation models. To overcome the massive computational and data requirements of large-scale training, researchers increasingly adopt strategies such as selective data sampling, efficient pre-training, and simplified reinforcement learning pipelines. While these strategies drastically reduce overhead, they prompt a critical, yet neglected question: Is efficiency achieved at the expense of model robustness and security? To our knowledge, we present the first systematic cross-domain investigation of the efficiency-vulnerability trade-off. Across vision and language models, we show that efficiency-oriented training increases susceptibility to adversarial and privacy attacks. We characterize this vulnerability by analyzing the models' internal geometry and functional representations, demonstrating that the evaluated efficient variants consistently exhibit sharper loss geometry together with systematic changes in representational structure. We further extend our analysis to "zero RL training", finding that models trained using simplified RL recipes exhibit substantially greater susceptibility to catastrophic forgetting and more pronounced overconfidence than those trained through conventional alignment pipelines. Our findings suggest that training efficiency is rarely a "free lunch"; rather, the mechanisms that minimize computation can inadvertently compromise safety. We conclude by calling for a paradigm shift toward multi-objective training that jointly optimizes for performance, cost, and security. --- ### Residual-Stream Burden Shapes Representation Learning in Diffusion Transformers Authors: Tongtong Liang, Siqi Kou, Ziqiao Xi, Esha Singh, Kun Zhou, Zhijie Deng, Alexander Cloninger, Yu-Xiang Wang, Rahul Parhi Published: 2026-09-27 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2609.33895v1 In diffusion-based generation, a neural network can be trained to predict the clean data, the noise, or the velocity from a noisy input. These prediction targets are interconvertible and describe the same generative process, yet plain Diffusion Transformers operating on large pixel patches succeed with clean prediction and fail with noise or velocity prediction. We argue that this asymmetry arises because noisy targets require the residual stream to preserve noise-dependent input variation through depth for the final readout, forcing subsequent layers to compute on noisy representations. A spectrally concentrated clean target imposes a lighter demand, leaving greater freedom to organize hidden representations for subsequent computation. We call this preservation requirement *residual-stream burden* and show how it shapes representation learning in Diffusion Transformers. Controlled experiments indicate that the exploitable structure is spectral concentration in patch space and that the bandwidth of the persistent residual state is a key resource for noisy prediction. We further show that this account is consistent with recent decoupled pixel-space architectures, whose diverse designs all reduce the residual-stream burden on the main pathway. To examine this understanding from a complementary direction, we expand and reorganize the residual-stream bandwidth directly, introducing Spatially Indexed Hyper-Connections (SiHC) that reach FID 1.71 on ImageNet $256^2$. Together, these results identify residual-stream burden as a mechanism through which prediction targets and architecture jointly shape representation learning in Diffusion Transformers. --- ### MISHAP-Bench: A Hallucination Benchmark for Large Audio-Language Models Authors: Zhi Wen Soi, Giulio Segalini, Jian-Jia Chen, Lydia Chen Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33893v1 Large audio-language models (LALMs) produce fluent responses about audio but often hallucinate by making plausible yet ungrounded claims. Existing audio hallucination benchmarks mainly measure response correctness, leaving it unclear whether an LALM hallucinates or simply fails to understand the audio. We challenge correctness-based evaluation by defining two hallucination categories: (i) context, where claims are not grounded in the audio; and (ii) knowledge, where claims about audio-related topics lack support from externally verifiable facts. We introduce MISHAP-Bench, a comprehensive benchmark with 12,000 challenging open-ended question-audio pairs and a rigorous evaluation pipeline covering both categories. To evaluate open-ended responses, we propose a groundedness judge that uses reference rubrics and judge prompts guided by human annotations. We extensively evaluate ten state-of-the-art LALMs and show that hallucination remains substantial. Even a frontier model such as Gemini 3.7 Flash reaches a hallucination rate of 36.5%. We further adapt and benchmark four mitigation methods from multiple domains for LALMs. Despite some improvements, effective hallucination mitigation remains an open challenge. Finally, we call on the community to evaluate hallucination and benchmark mitigation methods with MISHAP-Bench. --- ### Augmented Feature Boosting for Multicalibration Authors: Ira Globus-Harris, Inbal Livni Navon Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33891v1 Multicalibration requires a predictor's residuals to be unbiased not only globally, but also after conditioning on the predictor's own level sets and reweighting by a rich class of test functions. Standard boosting approaches in the distributional setting achieve this by repeatedly discretizing the predictor's range then auditing and repairing the resulting level sets. One consequence is that in practice, the algorithm's guarantees are sensitive to this parametrization of the rounding parameter. A natural theoretical question, then, is how to do discretization-free boosting which avoids this rounding within the boosting process itself. Here, we analyze an alternative feature-augmentation boosting paradigm inspired by Tax et al. (2026): at each round, a squared-loss oracle is called on hypotheses that receive the previous predictor's output as an additional feature, and only the final predictor is rounded to have a finite set of level sets to provide the multicalibration guarantee with respect to. We give a theoretical analysis of this procedure through the expressivity of the augmented hypothesis class, and show how the expressivity of this class yields a hierarchy of guarantees, including multiaccuracy, multicalibration, and the stronger notion of level-set multicalibration. --- ### Where Activation Sparsity and KV-Cache Sparsity Cross in LLM Decoding Authors: Jungseob Lee, Seungyoon Lee, Seongtae Hong, Sugyeong Eo, Heuiseok Lim Published: 2026-09-27 Categories: cs.LG, cs.CL, cs.PF arXiv: https://arxiv.org/abs/2609.33889v1 At each step, decoding one sequence with a large language model rereads the projection weights, whose traffic is fixed, and the key-value (KV) cache, whose traffic grows with context. Activation sparsity trims the first term and KV-cache sparsity the second, yet their reported speedups are hard to compare because each depends on context length and on the dense attention kernel it is measured against. We derive a byte crossover, the context length at which the two savings are equal, together with ideal speedup bounds for each branch and for their composition, from model dimensions and keep ratios alone. We then time both branches and their composition from 2K to 128K tokens on two GPUs after a dense prefill of real text, with dense and sparse modes reading the cache through the same split-K attention kernel. The projection branch leads at short context and the KV branch at long context, with speedups that follow their byte bounds up to fixed kernel costs. Adding these costs, measured in separate sweeps, lets the byte account predict the measured crossings of three keep-ratio pairs, a second model, and a second GPU to within 4.1K tokens. Timing the dense baseline with masked instead of split-K attention inflates the apparent speedup of the same KV policy about fivefold. An attention-scored KV selection answers the same passkey and multi-key placements as dense decoding up to 127K tokens, whereas a KV window misses most of them. Under matched perplexity budgets, activation sparsity composed with this selection decodes 14 to 26% faster than the best single branch on both GPUs. Code is available at https://github.com/js-lee-AI/ByteCross. --- ### Vanilla Policy Optimization Is Both Optimal and Differentially Private for Stochastic Contextual Bandits Authors: Idan Attias, Orin Levy, Alexander Ryabchenko, Yishay Mansour, Uri Stemmer Published: 2026-09-27 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.33888v1 Can vanilla policy optimization explore enough to achieve near-optimal regret in stochastic contextual bandits? We show that standard exponential policy updates driven by offline regression do so under realizability, without exploration bonuses or importance weighting. For $A$ actions, $T$ rounds, and a finite prediction class $F$, vanilla PO achieves $\widetilde O(\sqrt{AT\log(|F|)})$ regret with high probability. Our analysis reveals an implicit exploration mechanism of independent interest: gradual policy updates prevent actions from losing probability too quickly, allowing the regression oracle to learn their expected losses. We further develop a batched version using only $O(\log T)$ regression calls and policy switches, and show how private regression oracles yield differentially private contextual bandit algorithms without composition across batches. For a finite class, this gives pure $\varepsilon_{\rm priv}$-DP and regret $\widetilde O\left( \sqrt{AT \log(|F|/δ)}(1+\varepsilon_{\rm priv}^{-1/2}) \right)$. Finally, experiments across oracle-based contextual bandit algorithms, with and without privacy, demonstrate the practical effectiveness of policy optimization and the value of explicit exploration under stronger privacy constraints. --- ### Faster Block-Diffusion Serving with Distribution-Free Risk Guarantees Authors: Jungseob Lee, Dongyub Jude Lee, Chanjun Park, Sugyeong Eo, Heuiseok Lim Published: 2026-09-27 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.33887v1 Block-diffusion language models are served at hand-picked operating points, such as acceptance thresholds, buffer depth, schedule, checkpoint and precision, and each point is chosen by its mean benchmark accuracy. However, a mean does not tell an operator how often a faster configuration fails on prompts that the slower one answers correctly. On the serving engine and its decode traces, the default commit rule already commits every fully resolved block, a static skip rule captures nearly all of the compute that allocation can save, and self-distillation on engine-decoded targets adds speed at unchanged accuracy. Larger speedups come from lower thresholds, which commit tokens that are still uncertain. We therefore present Redline, a finite-sample procedure that selects operating points, hand-picked or learned, from the correctness of their answers on calibration prompts. Redline keeps the reference-relative risk, the joint probability that the reference answers correctly and a candidate configuration does not, within a user-chosen budget with high probability, and deploys the fastest configuration that passes. It speeds up math at a smaller risk budget than code in both model families, and at a budget of ten percent it deploys a LLaDA2 math configuration that commits over a third more tokens in each forward. It also applies without modification to the acceptance rule of speculative decoding and to weight quantization. On the same calibration data, Redline stays within its stated failure probability, whereas each tolerance of a mean-accuracy rule either gains less speed for some model and task or exceeds the risk budget far more often for another. Code is available at https://github.com/js-lee-AI/Redline. --- ### LLMs learn different forms of metacognition when trained to predict their own accuracy Authors: Nicolas Yax, Stefano Palminteri, Pierre-Yves Oudeyer Published: 2026-09-27 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.33886v2 Large language models are trained to always produce an answer, regardless of whether they possess the relevant knowledge, which leads them to fabricate facts. Prior work has shown that LLMs' confidence estimates correspond poorly to their actual performance, and that fine-tuning can substantially improve them. However, what models actually learn during such training remains poorly understood. We investigate how LLMs acquire metacognitive monitoring, the ability to know what one knows, by training 10 open-weight LLMs to predict their own accuracy on factual multiple-choice questions before answering them. We find that trained confidence reflects two distinct signals. While on questions close to the training data, it tracks the model's true accuracy, in other domains, it instead tracks output consistency: the concentration of the model's answer distribution. Output consistency tracking emerges early in training and generalizes across datasets, whereas accuracy tracking develops later and remains local to the training distribution. These results suggest that calibration training may not teach models to generally detect errors they commit confidently, and they raise broader questions about the nature of metacognition in artificial systems. ---