# Research Papers — 2026-09 · bundle page 70 of 266 > Full-text markdown of papers #3451–3500 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/70 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/69.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/71.md) ### Adapting Nonstationary Multi-output Gaussian Processes to Bayesian Optimization Authors: Zikai Xie Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32464v1 Multi-objective Bayesian optimization (MOBO) commonly relies on independent Gaussian processes (GPs) with stationary kernels, limiting its ability to represent nonstationary structure and share information between objectives. However, expressive nonstationary GPs do not necessarily make reliable BO decisions. We study this mismatch for the multi-output low-rank nonstationary (MO-LRN) GP: strong training fit can coexist with large off-design errors and optimistic acquisition predictions. We introduce MOLRN-BO, which combines a regularized shared-spectral surrogate with objective-specific residuals, prequential mean correction and tempered covariance scaling, and Pareto-local qLogEHVI optimization with periodic global search. Experiments on 12 deterministic bi-objective benchmarks show that MOLRN-BO substantially improves upon the original MO-LRN and achieves the best average problem ranks for final normalized hypervolume and normalized inverted generational distance among nine evaluated algorithms. It also achieves the strongest adverse-tail performance while remaining competitive with the leading baselines in anytime optimization. Ablation studies further show that the shared spectral construction improves off-design prediction, the local--global decision policy improves optimization performance, and hierarchical calibration reduces systematic candidate bias. These results demonstrate that nonstationary multi-output surrogates can deliver strong and robust MOBO performance when their structure and use are explicitly adapted to the demands of sequential optimization. --- ### Can Motion-Language Models Ground Structure? STRIDE for Evaluating the Evaluators Authors: Lixing Tan, Qing Xia, Yuting Guo, Shuai Li, Aimin Hao Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32462v1 Motion-language models are typically scored by motion-language evaluators, but how well these evaluators ground language structure remains unclear. Here, we introduce the Structure grounding via Temporal-order, Reflection, and Identity Diagnostic Evaluation (STRIDE) benchmark to systematically evaluate the ability of evaluators to track temporal order, mirror reflection, and action identity. STRIDE comprises $5{,}869$ triples, each consisting of a motion, its original caption, and a perturbed caption, spanning both short and long descriptions. We likelihood-balance caption pairs to reduce text-only bias and estimate each evaluator's caption preference under unrelated motions to measure the discrimination gain from matched motions relative to this baseline. Our experiments reveal weak structural grounding and severe deficits in mirror sensitivity among the audited evaluators, which commonly used evaluation protocols fail to expose. To understand why these limitations go undetected in standard tests, we examine the evaluators more closely. We find that text-only priors alone can solve naive perturbation tests on existing datasets, while common retrieval and distributional metrics barely respond to structural corruption introduced by mirroring ground-truth motions. These findings suggest a natural intervention: structural hard negatives. Our experiments show that a simple modification to contrastive learning substantially improves performance on temporal order and mirror reflection. The benchmark and code will be released. --- ### REMEDY: How Far Is Video Generation from Medical Education World Models? Authors: Lixing Tan, Yanghao Zhou, Qing Xia, Yuting Guo, Shuai Li, Aimin Hao Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32460v1 Recent video generation models produce realistic videos and show potential as a foundation for world models. These advances create opportunities for generating medical teaching demonstrations, which requires both convincing visual quality and precise procedural actions. However, whether current generators can meet these requirements has not been measured. To address this problem, we introduce Readiness Evaluation of Medical Education Demonstration sYnthesis (REMEDY), to our knowledge, the first benchmark for AI-generated medical teaching demonstrations. REMEDY provides 900 first frames from real demonstration videos, covering 12 tasks across four scenarios: operating room, imaging, clinic and bedside, and resuscitation. Five contemporary open-source video generation models produce 4,500 videos from these frames. We combine task-specific clinical checklists with video and motion quality metrics. Evaluation covers four dimensions: clinical action following, clinical profiles, video quality, and motion quality. Our results show that realistic appearance and temporal consistency do not ensure correct clinical actions. Even the most advanced MiniMax-H3 achieves only 28.25% on strict clinical success rate, and fine-grained clinical actions remain challenging. These findings establish a foundation and roadmap for developing future medical education world models. --- ### Streamlined Reflective Evolution for Task-Adaptive Self-Refinement Pipelines Authors: Xiaofan Zhou, Lu Cheng Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32458v1 Reflective prompt optimization improves large language model (LLM) systems without updating model weights, but fixed architectures constrain how self-refinement is organized. We introduce Workflow-Designing Agents (WDA), a framework for streamlined reflective evolution of task-adaptive self-refinement pipelines. Starting from a minimal prompt, WDA jointly evolves stage instructions and their sequential structure. During evolution, we find that repeated revisions can accumulate redundant instructions in a single prompt. In WDA, we propose to address this problem with SPLIT, which redistributes these instructions across specialized stages. Three-example reflection and local screening guide selective search, while calibration scores guide Pareto admission and rollback of unhelpful trailing updates. The resulting pipelines are task-adaptive: their instructions and depth are learned from task data, then fixed for all test inputs within that task. We evaluate WDA on five benchmarks spanning knowledge, mathematical reasoning, multi-hop question answering, and instruction following. On Qwen3.5-9B, WDA achieves an average score of 51.24%, improving over the initial solver by 8.63 percentage points and the variant without SPLIT by 2.60 points. On GPT-4.1-mini, it achieves 49.00%, with corresponding gains of 5.62 and 3.69 points. These results support task-adaptive self-refinement as a complementary direction to broader agentic workflow search. --- ### Write Back the $Δ$: Revisiting the Same Tokens with Fresh Representations Authors: Wencheng Ye, Anning Hu, Xiangdong Zhang, Tianyi Wang, Yikang Li, Hengyu Jin, Bing Li, Junchi Yan Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32457v1 Transformers process information strictly forward through depth, preventing deeper computation from revisiting and refining earlier representations. To augment the standard forward pass, existing approaches either re-execute depth, incurring additional computation, or modify the residual stream using predefined directions, limiting their instance-level adaptation. Recently, inference-time feedback offers a direct mechanism for recycling endogenously produced computation by writing deeper residual states back to earlier layers, yet what should be fed back remains unclear. We argue that the depth increment Delta, capturing newly accumulated computation between two layers, provides a more effective, composable, and scalable feedback signal than the full state. Building on this observation, we introduce ReFlux, a learnable feedback graph that dynamically selects and composes increment-carrying routes. ReFlux supports synchronous feedback to the same token and streaming feedback to subsequent tokens. Extensive experiments across various models, corpora, and benchmarks show that synchronous ReFlux consistently reduces perplexity across ten language-modeling corpora, and improves accuracy by 2.1-2.3 points, with gains reaching 4.7 points on multi-hop reasoning. Streaming ReFlux further retains most of these gains while preserving the base model's 1x theoretical backbone FLOPs. These results establish ReFlux as an efficient paradigm for unlocking the latent computational potential of LLMs, allowing them to revisit the same tokens with fresh representations. Code implementation can be found at https://github.com/gooogleshanghai/reflux. --- ### Seeing Parts, Reasoning about Worlds: Visual Inference under Partial Observation Authors: Wei Wang, Wenqiao Zhang, Yutong Lin, Jun Xiao, Yueting Zhuang Published: 2026-09-26 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.32456v1 World modeling under partial observation requires reasoning about the complete worlds that remain compatible with limited visual evidence. Occluded objects and unseen regions can leave several world states possible; additional views can exclude alternatives and strengthen the conclusions supported by the observations. We introduce WorldScope to study this process through possible-world semantics, evidence-grounded data, and learned visual representations. WorldScope-1.2M provides 1.2 million English question-answer pairs spanning eight world properties, ten task interfaces, and three observation protocols. Its answers encode confirmed facts, supported bounds, and unresolved possibilities. Complementary supervision comprises 4,800 certified counterworld groups with equivalent base observations and different hidden object configurations and query answers. These groups provide physical witnesses of ambiguity and training-only labels for world compatibility and view-induced exclusions. We propose WorldFlow, which composes cross-view entity evidence and support-surface coverage into an image-subset evidence lattice. Counterworld compatibility and transition objectives train subset representations to reflect how new observations constrain possible worlds. A shared answer generator uses these representations to predict the strongest supported conclusion. WorldScope-Bench evaluates claim judgments and evidence-dependent conclusions as views are selected, combined, removed, or ordered. On its 5,000-question test set, WorldFlow reaches 64.34% exact accuracy, improving over the same backbone trained on QA alone by 24.88 percentage points. It retains 50.43% accuracy on the 3,000 questions from structure-disjoint scenes. --- ### QuacamFM: Quaternion-Constrained Flow Matching for Camera Pose Estimation Authors: Bao-Long Tran, Cuong Le, Tahereh Dehdarirad, Fredrik Viksten, Per-Erik Forssén Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32455v1 Camera pose estimation from multi-view images remains a challenge in computer vision. Traditional methods often address this problem using Structure-from-Motion (SfM) with bundle adjustment. However, camera poses estimated from sparse views are inherently ambiguous due to insufficient geometric constraints. Recent work leverages probabilistic models, such as diffusion models, to generate multiple camera pose hypotheses and therefore capture this uncertainty better. Most of these methods represent camera rotations using unit quaternions, but treat them as unconstrained 4D vectors during the generative processes, thereby ignoring the unit-norm constraint of quaternions. Unconstrained quaternions create non-smooth and suboptimal generation trajectories. To this end, we propose *QuacamFM*, a quaternion-constrained flow matching framework for camera pose estimation that preserves unit quaternion representations throughout the entire flow trajectory. We design the optimal transport of the quaternion flows using smooth spherical linear interpolation. Experiments on CO3Dv2 demonstrate our method's advantage in camera pose accuracy over diffusion-based methods and classical SfM approaches. We further show that our quaternion-constrained formulation outperforms the naive application of standard flow matching to 4D quaternion vectors on sparse-view camera pose estimation. Finally, it is observed that QuacamFM generalizes well across datasets and in-the-wild examples. --- ### De-biasing Skeleton-based Action Recognition with Convex Hull Adaptive Shift Authors: Mengyuan Liu, Yuhang Wen, Yi Zhang, Songtao Wu, Hong Liu, Junsong Yuan, Beichen Ding Published: 2026-09-26 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32454v1 Skeleton sequences can represent both individual actions and multi-entity interactions, encompassing human bodies, hands, objects, and robots. Existing approaches to recognize skeleton-based actions and interactions usually adopt a late fusion strategy, which expects individuals are independent and identically distributed to train a robust weight-shared entity encoder. However, observed entity bias in various skeletal data violates this assumption, leading to suboptimal optimization of backbone models that might produce wrong recognition results. This bias arises from the world coordinate system's initial configuration, where the choice of origin often creates bias in representation. To this end, we propose a Convex Hull Adaptive Shift based normalization method to reduce Entity bias (CHASE), improving performance across a variety of skeleton-based action and interaction recognition tasks. To adaptively apply plausible shifts to the input skeletons, we formulate a plug-and-play parameterized network that ensures the relocated world origin lies within the skeleton convex hull, which avoids non-convergence by limiting the search space. To further minimize entity bias, we incorporate an auxiliary objective that leverages pair-wise distribution distances to guide network optimization. To support both single- and multi-entity actions, we propose a sub-entity strategy that offers a consistent formulation for both scenarios. Moreover, CHASE demonstrates compatibility with various intra-skeleton modalities, such as bones and velocities, highlighting its adaptability. Essentially, our method works as a normalization approach to reduce entity bias, enabling subsequent classifiers to achieve improved recognition performance across diverse settings. Extensive experiments on 7 datasets verify our approach by seamlessly integrating with various backbones and significantly boosting their performance. --- ### DRAM: Delta-rule Recurrent Associative Memory for Robot Manipulation Policies Authors: Xinyu Zhao, Yixiang Shan, Tao Yang, Runyu Lei, Yiming Zhao, Jiaxin Fan, Zongbao Feng, Peng Jia Published: 2026-09-26 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.32453v2 Robotic manipulation is inherently history-dependent, yet most pretrained robotic policies condition on only the current observation or a short temporal window. Equipping such policies with long-term memory remains challenging: existing approaches either feed the backbone multi-frame observation windows, which substantially increase inference cost, or rely on pre-defined semantic features, which limit task generality and may also require the retraining of the backbone to adapt to the memory. We introduce DRAM (Delta-rule Recurrent Associative Memory), a plug-and-play memory module that can be attached to a wide range of pretrained robotic policies, endowing them with long-horizon memory without architectural modification or backbone retraining, requiring only task-specific post-training of the memory module and action expert. DRAM maintains a fixed-size associative memory using gated delta-rule linear attention, with a modified update that incorporates all tokens within each frame in parallel. An architecture-agnostic readout integrates historical context into action prediction across different policy architectures. Experiments show that DRAM consistently improves frozen pretrained policies over short-context baselines and alternative compact memory designs, validating its effectiveness as a fixed-size, post-hoc memory module trained with the backbone frozen. --- ### What Should the Reflector See? An Empirical Study of Evidence in Reflective Prompt Optimization Authors: Xiaofan Zhou, Lu Cheng Published: 2026-09-26 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.32452v1 Reflective prompt optimization revises instructions using examples of a model's behavior, but which evidence the reflector should receive remains unclear. We study evidence composition, visibility of examples, candidate selection and domain-knowledge policy within a single-parent Pareto-guided search. Using Qwen3.5-9B as both task model and reflector, we evaluate nine reflection strategies on five datasets. From these experiments, we find three distinct patterns. For performance improvement, Failures-only produces the largest mean test gain (+8.0 percentage points), while Balanced-mix and No-examples+Val share the best mean performance rank. For reflection effectiveness, No-examples achieves the best rank for improving sampled parents, yet yields only a 1.4-point mean test gain: local reflection success does not necessarily produce a stronger final prompt. For overfitting assessment, Failures-only and Balanced-mix share the lowest mean calibration-gap rank, while larger gaps on GPQA and IFBench show that calibration gains can overstate held-out improvement. This gap is a descriptive indicator, not a direct measure of overfitting. Together, these results show why reflection strategies should be assessed separately on final performance, parent improvement and calibration-to-test transfer. --- ### Self-Reports Do Not Identify Self-Models: An Identifiability Test for Counterfactual Reports Authors: Phongsakon Mark Konrad, Toygar Tanyel, Serkan Ayvaz Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32449v1 Language-model self-reports are evidence about behavior in a prompt environment, not by themselves evidence of a self-model. We investigate counterfactual reports about affect-like states under activation interventions and ask whether the report remains bound to the named intervention when the demonstration environment changes. Across three open instruction models, wrong-source demonstrations move reports toward the source answer family, while explicit mechanism binding reduces this pull. Self-report benchmarks should include environment-shift invariance tests under fixed intervention before treating accuracy as evidence for an autonomous report mechanism. --- ### ForkLeft: Entropy-First Rollouts for Prefix-Aligned Autoregressive-to-Diffusion Distillation Authors: Junming Liu, Jicheng Wang, Yifeng He, Hao Chen, Jianzhong Qi Published: 2026-09-26 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.32448v1 Autoregressive Next-Token Prediction (NTP) has enabled strong reasoning capabilities in language models, while Diffusion Language Models (DLMs) offer flexible token orders and parallel generation. We ask whether DLMs can acquire NTP-style reasoning through distillation without giving up their native generation process. Direct distillation, however, faces a fundamental mismatch: an autoregressive teacher predicts from a left prefix, whereas a DLM can condition on tokens on both sides. We introduce ForkLeft, a distillation framework that resolves this mismatch by separating the student's rollout from teacher supervision. During training, the student first performs entropy-first rollouts that commit uncertain positions and expose potential forks. We then fix the resulting student prefix and distill an NTP teacher under the same context, with answer correctness determining the supervision source. At inference, the student returns to its native confidence-first parallel decoding. With Qwen3-30B-A3B-Base, ForkLeft improves Efficient-DLM-4B on all ten benchmarks, raising MATH500 from 72.60% to 79.60% and consistently outperforming three alternative designs. The gains scale with teacher strength and generalize to SDAR-4B with only $500$ updates. At matched scale, the distilled 4B and 8B students exceed the published SDAR-Chat and OPDLM models on seven benchmarks, showing that DLMs can learn NTP-style reasoning without sacrificing native parallel generation. Code and datasets will be released upon acceptance. --- ### Length-Independent State Tracking Under a Parallel Scan Authors: Julien Brandoit, Arthur Fyon, Thomas Braipson, Tom Clara, Florent De Geeter, Pierre Sacré, Damien Ernst, Guillaume Drion Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32447v1 Learning robust and scalable finite-state tracking is fundamental to sequence processing. While linear recurrent neural networks (RNNs), linear attention, and state space models enable scalable parallel training through affine recurrences, their theoretical expressivity guarantees assume idealized arithmetic and do not extend to finite precision, where the parallel scan that makes them fast is itself a source of perturbation. We formalize finite-state tracking at finite precision and characterize length independence: tracking that stays correct at every sequence length, at a precision cost that does not grow with the length. We show that length-independent state tracking requires two competing dynamics within a single map: contraction to suppress numerical perturbations and separation to keep distinct states apart. We prove that affine recurrences, which offer a single rate at each step to serve both roles, realize at most definite automata at finite precision. Instead of treating scan compatibility as a restriction on the update map, we reinterpret it as a computational budget and introduce the Neural Finite-State Machine (NFSM): a nonaffine, scan-compatible RNN built for length-independent finite-state tracking. On synthetic benchmarks spanning abelian and nonabelian groups, noninvertible monoids, and textual state-tracking tasks, affine baselines fail on every nondefinite task, most of them within a few hundred steps. A single NFSM layer instead learns the exact transition tables of every algebraic task, which certifies correctness beyond the tested lengths, and a stack of NFSMs keeps perfect accuracy on the textual tasks at every tested length. --- ### Masking Frequent Tokens Sharpens Direct Preference Optimization Authors: Harshvardhan Saini, Samyak Jha, Yiming Tang, Dianbo Liu Published: 2026-09-26 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.32445v1 Direct Preference Optimization (DPO) aligns language models by optimizing over sequence-level sums of token-wise implicit reward differences. However, we identify a pervasive pathology in this formulation: a disproportionately small subset of high-frequency token types dominates cumulative sequence scores while appearing symmetrically across both preferred and dispreferred responses. Specifically, under canonical Qwen tokenization on Anthropic HH-RLHF, merely 69 token types account for $55.1\%$ of all response tokens and $85.9\%$ of within-pair shared token mass, exhibiting substantially lower preference-side specificity than the remaining vocabulary. This symmetric ubiquity induces gradient entanglement and dilutes the discriminative preference signal propagated through the objective. To resolve this issue, we introduce \emph{Anisotropic DPO} (\textsf{ADPO}) and its canonical realization, \emph{Frequency-Hard DPO}. Using a fixed, label-agnostic vocabulary mask, our method zeroes the implicit reward contribution of high-frequency response tokens while assigning unit weight to informative positions, thereby suppressing gradient interference without modifying preference pairs, discarding context, or introducing learned parameters. Here, \emph{anisotropy} designates non-uniform token-level objective weighting rather than representational geometry. Extensive empirical evaluations on AlpacaEval, MT-Bench, and Arena-Hard demonstrate that Frequency-Hard DPO consistently outperforms standard DPO across Qwen-2.5-7B-Instruct and Llama-3-8B-Instruct, establishing that selectively masking shared high-frequency tokens offers an effective, zero-overhead mechanism for robust preference alignment. --- ### Rethinking Training-Inference Mismatch in LLM Reinforcement Learning: Where It Arises and How to Correct It Authors: Tianrun Yu, Kaixiang Zhao, Shangzhe Li, Yuxiao Yang, Porter Jenkins, Weitong Zhang, Taylor W. Killian Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32444v1 We study training-inference mismatch in reinforcement learning with verifiable rewards (RLVR) for large language models, where rollouts are sampled by an inference engine while gradients are computed by a training engine, and the two engines assign different probabilities to the same tokens. To account for this discrepancy in policy updates, we introduce calibrated importance sampling (CIS). CIS is motivated by an empirically supported logit-displacement characterization that expresses the mismatch as an additive displacement $\varepsilon_t$ in log-odds, determined by the per-logit perturbation before the softmax, whose distribution is approximately invariant to token confidence. This characterization motivates a confidence-aware truncation: large positive displacements are truncated at a single constant threshold, which maps back to an importance-ratio cap that tightens as token confidence increases. Theoretically, we show that CIS replaces the unbounded second moment that governs the error of exact importance sampling with a term bounded by a constant, at the cost of a bias controlled by the truncated excess. In evaluation across three mixture-of-experts models and five mathematical reasoning benchmarks, CIS achieves the highest five-benchmark average on all three models among the evaluated baselines. Diagnostic analyses show that CIS places less truncation bias on low-confidence tokens than truncated importance sampling, while upward clipping of small importance weights reduces held-out accuracy. --- ### Controllable GNN Explanations via Multi-Metric Preference Selection Authors: Rachit Verma, Yashraj J. Deshmukh, Anirban Dasgupta Published: 2026-09-26 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32436v1 Mechanisms for generating GNN explanations are crucial for building trust and mitigating biases in Graph Neural Networks (GNNs), especially in high-stakes scenarios. Most current methods optimize only for fidelity under the sparsity constraint. However, this discounts the need for interpretable explanations (those that consist of familiar motif patterns) and stable explanations (those that remain unchanged under structural perturbations). We propose a novel approach that optimizes GNN explanations across these metrics, exposing their relative weighing as a control. Experiments on various real-world datasets, including MUTAG, BA-2Motif, BAMultiShapes, and PROTEINS, suggest that our method produces higher-fidelity explanations than a state-of-the-art baseline on MUTAG and PROTEINS across all evaluated budgets, and on BA-2Motif at larger budgets, while being faster in the regime of small explanation budgets. We also explore how, given an input motif library containing standard motifs for the corresponding domain, the method can be used to determine the relative importance of those motifs in generating the explanations, and how this information can be used to further improve the quality of the output explanations. We also examine the relationship between different metrics through their induced tradeoff surface, and explore its dependence on the nature of the motif library. --- ### From Latents to Wires: Surgical Post-Editing on Large Language Models Authors: Jiankai Jin, Xiangzheng Zhang, Zhao Liu, Wenzhuo Xu, Dongdong Yang, Deyue Zhang, Quanchen Zou Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32434v1 Given a large language model (LLM), can whoever holds the weights name a semantic target (e.g., the model's identity), locate the model components that produce it, and edit them so that the target no longer appears while other capability is preserved? We call such an edit on a trained model a post-edit. We present L2W (latents to wires), a framework that performs surgical post-edits for named semantic targets. For localization, L2W uses Jacobian lens (J-lens) attribution to score components against the semantic target. For surgical removal, because LLM mechanisms are redundant (i.e., a semantic target may have multiple components producing it), L2W runs Counterexample-Guided Causal Cut (CGCC) until the target no longer appears. CGCC first cumulatively closes model components, treating each surviving expression of the target as a counterexample that exposes the next components to close, and then reopens some of them to preserve capability. In a controlled experiment with an implanted behavioural watermark, L2W removes the watermark, and its localization lands on the model region the implant changed. Across three model configurations, L2W removes model-metadata (e.g., identity) self-claims in all nine runs, and adult-content refusal in all three, with no held-out target residual. L2W further composes two post-edits on a text-to-image model: one removes the refusal of requested nudity, and a second removes the nude rendering the first exposes. The results support post-editing as a complement to post-training: post-training installs preferred behaviours, and post-editing removes named unwanted ones. --- ### Beyond the Manifold Hypothesis: Hybrid Spectral Parameterizations for Flow Matching Authors: Ségolène Martin, Anne Gagneux, Quentin Bertrand, Rémi Emonet, Mathurin Massias Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32432v1 Flow matching and diffusion can be trained to predict different quantities, most commonly the data $x_1$, the source noise $x_0$, or the velocity~$v$. Although theoretically equivalent, these can lead to substantially different performances. We identify two main drivers for these differences: the source--data signal-to-noise ratio, and the information bottleneck induced by the neural architecture. We show that, beyond intrinsic data dimension, the factor affecting the optimal parametrization the most is a certain signal-to-noise ratio in each data covariance direction. From this analysis, we introduce new \emph{spectral hybrid} parameterizations that adapt across time and data covariance directions; we show that these are optimal for Gaussian data. We also show that architecture-induced compression changes which parameterization is easier to learn, with $v$-prediction being more sensitive to discarded directions than $x_1$-prediction. Experiments across architectures and source scales show that our spectral parameterizations are robust across regimes, can substantially accelerate optimization, while incurring essentially no additional training cost compared with standard parameterizations. --- ### DualGuard: Dual-Mode Quality Control for Logic-Preserving Data Augmentation Authors: Shenghao Li, Lin Zhao Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32431v1 Large language models provide a practical way to generate augmented data for logical reasoning at scale, but a larger generation volume does not guarantee semantic, label, or logical reliability. Existing work has improved generation quality through generation constraints, candidate validation, filtering, and feedback-based revision; however, once a quality judgment is available, deciding whether a candidate should be retained, filtered, or repaired remains an important control problem. We propose DualGuard, a dual-mode quality-control framework for logic-preserving data augmentation. The first mode uses the current instance and candidate batch for selective retention, filtering, attribution, and targeted feedback. The second mode accumulates cross-instance execution records of augmentation actions on top of per-sample diagnosis and attribution, compares new executions against each action's own historical behavior, and supports retrospective anomaly inspection, targeted rollback, and bounded repair. Both modes share semantic verification and additionally use symbolic verification when a reliable logical form is available. Across seven downstream tasks in the Two-Stage Transfer setting, DualGuard achieves the highest Accuracy on five tasks and outperforms the no-augmentation BERT baseline on all seven. Controlled ablations further show complementary roles for Memory, Z3, and history-aware anomaly control. --- ### Multi-Agent System Search via Active Substructure-aware Policy Optimization Authors: Beicheng Xu, Bowen Fan, Weitong Qian, Lingching Tung, Bin Cui Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32430v1 LLMs enable multi-agent systems (MAS) to tackle complex tasks, but manually designing agent roles, prompts, and communication structures requires substantial expertise and effort. This motivates learning policies that construct query-specific MAS from execution reward. Existing approaches typically train these policies by repeatedly traversing a fixed set of training queries and assigning rewards at the workflow level. However, this overlooks differences in queries' evolving learning potential and obscures which substructures improve solution quality. In this paper, we propose Active Substructure-aware Policy Optimization (ASPO), a RL framework for query-level MAS search. ASPO introduces an Adaptive Query-Selection Mechanism (AQSM) that focuses training on queries at the policy's competence boundary: those it can solve but not yet reliably. A complementary discovery mechanism widens architectural exploration for hard queries, helping distinguish insufficient exploration from operator capability limits. Beyond query selection, ASPO introduces substructure-level rewards that measure output-quality gains within each action's descendant subgraph. These rewards guide proximal policy optimization to reinforce useful architectural refinements and discourage redundant or harmful computation. Together, these mechanisms prioritize learnable queries and provide fine-grained feedback for learning effective MASs. Across six benchmarks spanning mathematical reasoning, general question answering, and code generation, ASPO ranks first on every benchmark against twelve baselines. --- ### PrismQuant: Optimal Null-Space Rotations for Grouped Quantizers Authors: Yanlong Chen, Yining Chen, Song Zhang, Amirhossein Habibian, Yawei Li Published: 2026-09-26 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32429v1 Smaller activation outliers do not necessarily imply better low-bit quantization: their alignment with the quantizer matters. We introduce PrismQuant, a quantizer-aware rotation framework that aligns the leading activation eigenspace with the constant group subspace of asymmetric grouped INT4. The affine offsets represent the energy in this subspace without widening the range within the group. We formulate rotation design as a Ky Fan trace maximization and derive a closed-form solution that is provably optimal for this alignment objective. Compact Householder transformations and their compact-WY representation enable gradient-free construction and efficient application at both foldable and online sites. A predictive range law further connects unaligned activation energy and group size to quantization-relevant variation. Experiments on Llama, Qwen, and Mistral span dense models up to 70B parameters and a 30B mixture-of-experts model. Under W4A4KV4, PrismQuant sets the state of the art on Llama-3.2-3B among the compared methods in both perplexity and accuracy. On Llama-3.1-70B, it attains 3.85 perplexity and 72.46% average zero-shot accuracy, only 0.22 percentage points below full precision. In the deployment study on Llama-3.1-8B, our optimized implementation achieves 1.51x prefill and 1.22x CUDA Graph decode speedups over matched FP16 baselines, with 56.34% lower decode peak memory and only 2.35% additional Graph decode latency over Hadamard. Code is available at https://github.com/ForeverBlue816/PrismQuant. --- ### Authorization Closure Graph: Minimal Repair for LLM Agents with Evolving User Instructions Authors: Qingzhuo Wang, CaiYi Wang, Jinglu Meng, Ruiyang Qin, Kunyu Peng, Zhihua Wei, Wen Shen Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32428v1 Tool-using large language model (LLM) agents increasingly perform state-changing actions that require user authorization. Yet existing approaches do not provide a principled mechanism for selectively updating prior authorization when only part of an instruction changes. To this end, we propose an Authorization-Closure-Graph (ACG)-based framework that represents authorization and its dependencies as an evolving, versioned state. ACG selectively invalidates authority affected by a revision while preserving unaffected portions of the authorization state, and computes a minimal repair that identifies only the missing evidence or authority required for execution. This enables agents to adapt to revised instructions while avoiding stale authority and unnecessary authorization requests. We evaluate ACG across three advanced LLMs in two natural tasks, and ACG consistently improves action safety rate and task success rate. Code is available at https://github.com/weiliang822/ACG. --- ### CityToolVQA: Tool-Augmented Visual Question Answering for 3D Spatial Cognition in Urban Low-Altitude Environments Authors: Boao Yu, Yingzhen Nie, Yue Hu, Zhengqiu Zhu, Rusheng Ju Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32427v1 CityToolVQA addresses the weak performance of Vision-Language Models (VLMs) on quantitative tasks in urban low-altitude visual question answering. We divide the seven tasks into qualitative and quantitative groups: qualitative questions are answered directly by the VLM, whereas quantitative questions are handled by an external visual-geometric toolchain that performs object grounding, segmentation, depth back-projection, and spatial computation. The toolchain can be attached to different VLMs in a zero-shot manner; a Depth-Assisted Prompt Inference (DAPI) fallback is triggered when the main-chain detection is invalid or unreliable, and CityToolVQA-SFT adapts the 8B backbone to tool-conditioned inputs. On the 73,324-question Open3D-VQA-v2 test set, CityToolVQA-SFT (Qwen3-VL-8B) reaches 67.6% overall accuracy, and attaching the toolchain to ten open-source VLMs improves quantitative-task accuracy by 12.7-36.9 percentage points. These results indicate that externalizing explicit 3D geometric computation effectively complements the limited ability of RGB-only VLMs to estimate metric distances and object sizes. --- ### HoTS: Homophily-Aware Temperature Scaling for Graph Neural Network Calibration Authors: Inwoo Tae, Yoontae Hwang, Yongjae Lee Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32426v1 For graph node classification, calibrated class probabilities are needed when confidence scores, usually the maximum predicted class probability, are used to rank predictions, defer uncertain nodes to human review, or control risk. Existing post-hoc calibrators either apply one global temperature or use graph-aware modules without a principled structural form. We study how local graph structure should enter node-level calibration. Our first results show that a logit-only temperature rule is insufficient when nodes with identical logits but different local homophily require different optimal temperatures. We then analyze a population-concentration contextual stochastic block model with Gaussian features and a one-layer linear GCN. Under equidistant class means, the Bayes posterior over class-template scores is a temperature-scaled softmax whose inverse-temperature is governed by a homophily-dependent signal strength. In the positive-signal homophilic regime, the resulting temperature decreases approximately inversely with normalized local homophily. This law motivates Homophily-aware Temperature Scaling (HoTS), a simple post-hoc calibrator that assigns each node a positive scalar temperature from entropy-based logit concentration and estimated local homophily. HoTS has three temperature parameters, preserves the predicted class, and learns the strength of the structural correction from calibration data. Across 18 node-classification benchmarks, two GNN backbones, and eight calibration baselines, HoTS achieves the best mean Expected Calibration Error (ECE) of 4.79%, the best average rank, and the most reliable confidence ranking in selective classification. Code is available at https://github.com/inu0104/HoTS. --- ### CyberClear: A Benchmark for LLM Agent Systems on APT Attack Chain Provenance Authors: Qi Chen, Fushuo Huo, Hangli Shen, Jingcai Guo, Shuhao Li, Guang Cheng Published: 2026-09-26 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.32424v1 Large language model agents have demonstrated promising capabilities in cybersecurity tasks, yet their ability to reconstruct complete Advanced Persistent Threat attack campaigns from complex security logs remains largely unexplored. Existing cybersecurity benchmarks for agents mainly focus on vulnerability discovery, exploitation, and security analysis tasks, leaving the evaluation of attack chain provenance under realistic security logs insufficiently studied. To address this gap, we introduce CyberClear, a benchmark for evaluating LLM agents and advanced agent systems on APT attack chain provenance from long-context security logs. CyberClear covers both single-step attacks and multi-stage attack chains, requiring agents to identify attack evidence, infer attack progression, and generate provenance graphs containing entities, causal relationships, MITRE ATT&CK techniques, and forensic evidence. To enable comprehensive evaluation, we develop an evaluation method tailored to APT attack chain provenance. Unlike conventional text similarity metrics that focus on surface-level matching, our evaluation examines whether reconstructed graphs preserve the semantics of attack chains across single-step behavior correctness, multi-step behavior identification, temporal and causal consistency, entity and relationship fidelity, and overall attack narrative consistency. Advanced multi-agent systems powered by state-of-the-art LLMs still struggle on CyberClear, motivating us to propose CyberProvenance, an agent cyber harness designed for multi-agents that augments LLM agents with evidence accumulation, execution-based validation, and feedback-guided refinement mechanisms for reliable attack-chain provenance. Extensive evaluations on CyberClear demonstrate the effectiveness of CyberProvenance in improving evidence reasoning, execution-grounded validation, and complete APT attack chain reconstruction. --- ### PluginRSI: Recursive Improvement of Agent Harnesses with Reusable Plugins Authors: Yaorui Shi, Yuchun Miao, Yuxin Chen, Jiayuan Zhang, Yueqing Sun, Xierui Song, Xiang Wang, An Zhang Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32423v1 The harness surrounding a language model is a central determinant of agent performance. Recent methods optimize harnesses by searching over complete programs, where individual mechanisms are difficult to isolate and reuse. We introduce PluginRSI, which represents a harness as a composition of atomized plugins and organizes harness evolution around these plugins. Individual plugins are improved independently and accumulated in a shared library, then recombined into new harnesses at each iteration. PluginRSI improves over existing harness optimization methods across software engineering, command-line interaction, and question-answering tasks. The resulting harnesses retain their advantage when transferred to other solver models without further optimization. The evolved plugin library accelerates subsequent optimization from the initial harness, which helps faster and higher convergence on unseen tasks. These results show that accumulating reusable mechanisms provides an effective basis for continued harness improvement. --- ### MergeHEIR: Mitigating Multimodal Hallucinations as the Tax of Model Merging Authors: Jinyu Li, Hao Fang, Zhiming Zhang, Jiawei Kong, Bin Chen, Shu-Tao Xia Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32422v1 Model merging consolidates task-specialized experts into a single deployable model. However, we show that such capability consolidation incurs a merging tax of increased hallucination: across 8 model-merging methods, every merged checkpoint exhibits a higher hallucination rate than the average of its constituent experts. An intuitive approach is to adapt existing hallucination-mitigation methods to the post-merge model, yet this unconstrained adaptation disrupts inherited capabilities, creating a tension between hallucination mitigation and expertise retention. To tackle this challenge, we introduce MergeHEIR, a post-merge adaptation framework designed to reduce this merging tax while preserving expertise inherited from initial experts. Using small expert-task calibration sets, MergeHEIR constructs layer-wise null-space projectors via SVD from task-specific activations collected from the merged checkpoint, and periodically projects the accumulated post-merge displacement onto the resulting null spaces to preserve inherited expertise. Theoretically, we establish minimum-distortion and maximum-dimensionality guarantees, characterize the threshold-controlled adaptation-retention trade-off, and extend perturbation guarantees beyond finite calibration data. Across 24 paired comparisons spanning three MLLM configurations and 8 model-merging methods, MergeHEIR consistently mitigates hallucination while largely preserving inherited expertise, demonstrating a more favorable hallucination-retention trade-off. --- ### Carnator: Fast Text-to-Video Generation with Generation-Native Compatibility-Guided Cross-Request Reuse Authors: Xingkun Yin, Xuebin Tang, Mingkun Xu, Hongyang Du Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32420v1 Video diffusion transformers produce high-quality videos, yet iterative denoising incurs substantial inference latency, limiting interactive and large-scale serving. Most existing acceleration methods focus on individual requests, thereby restricting efficiency gains to redundancy within a single generation trajectory. Recent cross-request reuse offers an additional source of savings, but existing approaches often infer reusability from coarse semantic similarity. This conflates semantic relatedness with generation-level computational compatibility, so aggressive reuse may accept incompatible historical computation while conservative reuse leaves substantial acceleration unrealized. We present \emph{Carnator}, a cross-request acceleration framework that addresses this challenge by extracting and using generation-native compatibility evidence directly from the model's evolving internal states. Specifically, \emph{Carnator} performs a lightweight early probe to construct an Early Signature from internal diffusion states, assessing reuse validity through risk-aware compatibility decisions. The same evidence characterizes reuse scope by localizing target-specific computation and guiding joint reuse of historical latent trajectories and sparse attention connectivity. Across three text-to-video backbones, Carnator consistently achieves higher cache-hit end-to-end acceleration than the evaluated cross-request baselines despite more selective cache acceptance, reaching up to 2.17$\times$ speedup while maintaining competitive generation quality. --- ### RE-0: Verified Recursive Improvement of Embodied Code-as-Policy Agents through Local On-Policy Distillation Authors: Jiawei Zhang, Xiangrong Zhang, Rui Song, Huanbin Zhou, Chengye Song, Hongzhou Wang Published: 2026-09-26 Categories: cs.RO, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32416v1 Code-as-Policy agents accomplish long-horizon embodied tasks by generating and executing code, yet continually improving them with teachers that are stronger but not globally reliable remains a key challenge. Existing distillation methods typically treat the teacher's complete behavior as the supervision target and thus misassign training credit on states where the teacher fails. We propose RE-0, a recursively verified policy improvement framework: rather than assuming that the teacher globally outperforms the student, RE-0 requests local corrections from the teacher on the student's own failure histories and checks in the environment whether each correction is genuinely beneficial; verified corrections yield immediate improvement. Building on this, we propose RE-OPD, which turns verified interventions into supervision for on-policy distillation. Only counterfactually verified teacher interventions provide distribution-level supervision, weighted by their measured local benefit, and the improvement they induce is projected back into the standalone student, so both where supervision is applied and how much credit the teacher receives co-evolve with the student policy. We further prove that the student's per-round gain is lower-bounded by its verified intervention gain up to verification and projection error terms. Experiments across multiple Code-as-Policy embodied tasks show that RE-0 improves both teacher-assisted execution and the standalone student, and generalizes to novel robots and scenes. --- ### RefAdapt-DiT: Adaptive Joint Attention for Reference-Conditioned Diffusion Transformers Authors: Jian Tang, Jiawei Fan, Qiannan Zhou, Qingbin Liu, Jiang Bian, Zang Li Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32415v1 Diffusion Transformers (DiTs) have become the standard backbone for high-quality generative modeling, yet deploying them in conditional generation tasks remains computationally prohibitive because bidirectional joint attention repeatedly processes large reference streams. While existing optimization schemes mitigate generic temporal redundancy, they typically rely on coarse-grained static reuse and overlook the distinct dynamics of references and targets. Specifically, we observe that reference representations often evolve slowly along the generation trajectory, while the target often assigns little attention mass to them; reference drift and this target-to-reference exposure jointly shape how strongly stale reference states affect the target. To exploit these patterns, we introduce \RefAdapt, a training-free framework for adaptive control of joint attention between references and targets. Instead of rigid static strategies, \RefAdapt combines consecutive target-Q change with previously observed target-to-reference attention mass to control reference computation adaptively at block granularity. Under ultra-few-step settings, \RefAdapt enables speedups of up to $2.097\times$ on 4-step MiniMax H3 and $3.54\times$ on 8-step Qwen Image Edit, while maintaining comparable visual quality. --- ### AECSF: Adaptive Ensemble Conditional Score Filtering for High-Dimensional Nonlinear Data Assimilation Authors: Yangwen Zhang, Shiwei Ni, Xiaoping Zhang, Xiaofei Guan, Lili Ju Published: 2026-09-26 Categories: stat.ML, cs.LG, math.NA arXiv: https://arxiv.org/abs/2609.32411v1 Bayesian state estimation for high-dimensional nonlinear dynamical systems entails a fundamental tension between statistical fidelity and computational tractability, as particle weights can collapse, while Gaussian ensemble updates can miss non-Gaussian posterior structure. Score-based diffusion filters offer a sampling-based alternative, but existing training-free score filters often rely on heuristic likelihood corrections, which can compromise posterior accuracy by neglecting uncertainty about the system state associated with each noisy reverse particle. To address these issues, we propose AECSF, a training-free adaptive ensemble conditional score filter. AECSF constructs an analytically tractable score estimator from the conditional Tweedie identity, which recasts noisy posterior score estimation as estimating the conditional mean of the system state given a noisy reverse particle and the observation. To estimate these conditional means efficiently, AECSF employs a shared adaptive weighted proposal ensemble, while particle-specific conditional weights yield an estimate for each noisy reverse particle without separate proposal sampling. The proposal ensemble is updated using reverse-particle information within the same reverse-diffusion run to improve conditional-mean estimation. Theoretically, we characterize when a fixed weighted proposal measure yields the exact noisy posterior score. Under stated assumptions, we establish a bound relating conditional-mean estimation errors to reverse-sampling endpoint error. Numerical experiments demonstrate that AECSF improves the accuracy of posterior sampling and nonlinear filtering in high-dimensional problems with limited forecast ensembles. --- ### Recovery-Directed Symbolic Distillation of Neural Likelihoods Authors: Kianté Fernandez, Xinwei Li Published: 2026-09-26 Categories: cs.LG, q-bio.NC, stat.CO, stat.ML arXiv: https://arxiv.org/abs/2609.32409v1 Amortized neural likelihoods enable computationally expensive inference for models with analytically intractable or unspecified likelihoods, but their black-box nature limits interpretability. We introduce a symbolic distillation pipeline that converts trained neural likelihoods into explicit, interpretable expressions optimized for efficient parameter estimation. Our approach uses a recovery-directed objective to guide symbolic regression toward expressions that preserve parameter-recovery accuracy rather than merely approximating the likelihood function. Candidate expressions are evaluated on held-out datasets and selected using a criterion that jointly accounts for expression complexity, parameter-recovery performance, and distributional distance from the learned likelihood. We evaluate the pipeline on the diffusion decision model, a classical cognitive model, whose analytically tractable likelihood provides ground truth for controlled evaluation. The proposed recovery-directed objective improves parameter recovery over standard symbolic-regression objectives. The resulting symbolic likelihoods enable over 100 times faster parameter evaluation than both neural likelihoods and, when available, the exact likelihood, while maintaining a manageable loss in precision. We further demonstrate these computational benefits in Bayesian hierarchical inference on empirical data. Our pipeline provides a lightweight interface for integrating symbolic distillation with existing neural-likelihood estimation methods and can be adapted to a range of simulation-based inference settings. --- ### Automatic Speech Recognition for the Basaà Language: A Low-Resource Approach Authors: Sophie Gertrude Ngo Mock, Charles Moudina Varmantchaonala, Paul Dayang, Jean Michel Nlong, Christopher Gies Published: 2026-09-26 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.32408v1 The rapid advancement of Artificial Intelligence (AI) and Natural Language Processing (NLP) has revolutionized the way humans interact with machines. Among the most impactful developments is Automatic Speech Recognition (ASR), which enables computers to convert spoken language into text. Systems such as those built on deep neural networks, transformer architectures, and self-supervised learning have achieved near-human performance for well-resourced languages such as English and French. Yet, these advances have disproportionately benefited a small fraction of the world's languages. --- ### Opening LLM Judges: Recovering Preference Signals Beyond the Final Verdict Authors: Sourabrata Mukherjee, Sunayana Sitaram Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32407v1 LLM judges are widely used to evaluate model outputs, but their verdicts can be unreliable: a judge may favor the worse answer for its position, length, or other surface features. When a judge is wrong, is the information needed to judge correctly absent from the model, or present in its internal representations but not reflected in the output? We study this across 64 open-weight evaluators and 14 datasets, including causal interventions on 41 judges (editing activations mid-run to see whether the verdict changes). On LLMBar, built so the superficially better answer is the worse one, the verdicts of 50 judges agree with human labels only 0.456 of the time, even after averaging both answer orders. Yet a small probe on the same judges' activations, with no weight updates, reaches 0.846, and 0.686 once surface features such as length and position are residualized out (0.507 with shuffled labels). The gap holds across eight benchmarks and model families, but is not universal: a score of how well surface features alone predict the human label, computed before any probe is trained, predicts the size of the gain (Spearman rho = 0.90). On rubric tasks that score one answer at a time, leaving no surface cue to exploit, reading the internals gives no advantage. The interventions also show that editing activations mid-network already changes the verdict, before it can be read off directly, and locate the pathways carrying position and length bias. At the same label budget, the recovered signal lets a judge flag cases where it is likely wrong and yields better labels for preference learning. A wrong verdict, then, does not mean the judge lacks the information, and a simple diagnostic shows when it is worth recovering. --- ### Toward On-Chip Training of Spiking Neural Networks for Dense Event-Based Vision Authors: Maxime Vaillant, Axel Carlier, Lai Xing Ng, Christophe Hurter, Benoit R. Cottereau Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32405v1 Event cameras provide low-latency, asynchronous visual sensing for resource-constrained robotics. Spiking neural networks (SNNs) process event streams naturally, but training deep SNNs with backpropagation through time (BPTT) requires substantial memory and remains difficult on neuromorphic hardware. Local learning avoids this by restricting error propagation to local blocks, but existing methods mainly target classification rather than dense prediction. We introduce DELL (Dense Event-driven Local Learning), a block wise scheme for dense event-based vision that replaces global gradient propagation with local dense supervision. Learnable, spatially structured local heads supervise each block at its appropriate resolution while preserving temporal dynamics within blocks. We evaluate DELL on optical-flow regression and semantic segmentation with a fully spiking U-shaped architecture. On DSEC optical flow, DELL reduces peak training memory by 39.6% relative to end-to-end BPTT while improving accuracy, reaching 1.670 px endpoint error on the official test benchmark versus 1.941 px for the same backbone trained end-to-end. Block detachment behaves more like a regularizer than a constraint. DECOLLE, the existing local-learning baseline, relies on fixed random local read-outs poorly suited to dense regression, resulting in a 3.9x higher endpoint error; learnable local heads recover this loss and outperform end-to-end training across all optical-flow metrics. On segmentation, they recover most of the performance gap, although DELL remains a few mIoU points behind end-to-end training. With 2.3M parameters, 24x fewer than the strongest SNN baseline, the backbone remains competitive with the SNN state of the art on DSEC. These results extend local learning to dense event-based prediction while substantially reducing training memory. --- ### Shared Worlds, Private Minds: Structured Memory for Long-Form Writing as World Creation Authors: Qiuyu Tian, Xiaowen Gu, Hang Su, Jianghan Chao, Haojie Yin, Fan Guo, Xin Zhang, Jinjing Shen, Ewing Luo, Youyong Kong, Yingce Xia, Zequn Liu Published: 2026-09-26 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.32401v1 LLM agents that write long-form fiction need an explicit memory of the evolving storyworld to keep new events consistent with established facts. Such memory must keep heterogeneous narrative information distinct, integrate story developments across granularities, and recover dependencies that a writing request leaves implicit. We present NarraWorld, a structured memory system for long-form writing that treats memory construction as world creation. From a shared evidence-grounded graph, NarraWorld derives four connected views: world facts, per-character beliefs, open developments, and hypothetical branches (possible-world continuations). Hierarchical aggregation with atomic closure consolidates events into scenes, plotlines, and plots, keeping each higher-level node traceable to its constituent source spans. For retrieval, planned reconstruction infers a query's dependencies from the current narrative situation and a preview of memory, then assembles the relevant records within a token budget. Across three writing benchmarks, NarraWorld achieves the strongest aggregate results. Its memory also transfers to situated role-playing and largely preserves recall on a general-purpose long-term memory benchmark, paving the way for agents that sustain coherent storyworlds across diverse narrative tasks. --- ### SkillDRE: Dual-Stage Red-Team Evolution of Agent Skills via Pre-Execution and Runtime Feedback Authors: Pengyu Zhu, Jingyi Yang, Yi Liu, Li Sun, Sen Su Published: 2026-09-26 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.32400v1 Agent skills package instructions, executable code, and task-specific resources into reusable artifacts that agents can improve using execution feedback. The same mechanism also enables attackers to evolve malicious skills, making them more effective and less detectable. However, a candidate skill may pass pre-execution scanning yet fail to realize its target under runtime defenses, while a revision that repairs execution may introduce new scanner findings. We introduce SkillDRE, a fully automated framework for evolving complete malicious skill packages through a dual-stage feedback loop. Given a benign task and its associated skills, SkillDRE autonomously constructs and validates a task-conditioned malicious objective and a verifiable judge rule. It then holds both fixed while evolving the skill implementation, with preservation of legitimate task capability. SkillDRE combines scanner-guided evolution with runtime-guided refinement informed by execution outcomes observed under runtime defense. Each runtime-guided revision returns to the pre-execution stage for rescanning and further optimization before re-execution, forming a cross-stage closed loop. Evaluated on SkillsBench across four victim models, SkillDRE achieves an average attack success rate of 45.28%, exceeding the strongest baseline by 40.3%, while its final submitted skills receive no SkillScan findings and largely preserve benign-task performance. These results show that two-stage defense feedback can serve as a useful learning signal for adaptive red teaming and that evaluating either defense stage in isolation can miss the resulting attack capability. Codes is available at https://github.com/whfeLingYu/SkillDRE --- ### Endo-TSR: Temporal Spectral Modeling of Appearance and Motion for Endoscopic Reconstruction Authors: Taoyu Wu, Yiyi Miao, Qi Shao, Zhuoxiao Li, Zhe Tang, Limin Yu, Baoru Huang Published: 2026-09-26 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2609.32399v1 Endoscopic scene reconstruction requires modeling tissue motion and temporal appearance while recovering fine surface detail. Deformable Gaussian models provide explicit trajectories, but their fixed colour coefficients lack a dedicated temporal representation for photometric changes. We propose Endo-TSR, which augments deformable Gaussian splatting with bounded Fourier colour residuals and independent translation residuals on shared temporal frequencies. The colour residuals capture local appearance changes, while a Matérn spectral prior regularises motion corrections. Multi-scale Laplacian supervision guides tissue-detail recovery during joint image fitting. Extensive experiments on the EndoNeRF and StereoMIS datasets demonstrate state-of-the-art rendering quality, with the highest PSNR across all evaluated sequences. Ablation studies show that temporal appearance yields the largest PSNR gain among the tested component additions, while appearance and detail supervision jointly improve rendering with fixed Gaussian counts. --- ### Function Over Form: Distributional Orthogonalization in Mixture-of-Experts with Replica Expert Mechanism Authors: Jinfan He, Yunzhuo Liu, Kai Zhang, Weidong Han, Key, Rayying Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32398v1 The scaling of LLMs increasingly relies on MoE architectures to decouple active computation from total parameter count. However, the efficacy of MoE is often constrained by expert collapse and representation redundancy, both leading to underutilization of model capacity. To address these challenges, this paper proposes Distributional Orthogonalization Loss (DO-loss), an auxiliary regularization that shifts the focus from static weight diversity to dynamic routing behavior. By representing each expert's token assignment history as a high-dimensional binary load signature, DO-loss penalizes signature overlap to prevent expert collapse while encouraging functional specialization. To align this algorithmic design with system efficiency, we further introduce the Replica Expert Mechanism (REM), which improves load balancing through a two-tiered strategy: adjusting replica expert placement at the global-batch level and performing real-time token dispatching at the micro-batch level. Empirical evaluations demonstrate that our method outperforms the evaluated routing algorithms on downstream tasks for both 4.8BA0.5B and 30BA3B MoE models, while maintaining comparable training efficiency. --- ### SinBrief: A Hybrid Framework for Abstractive Text Summarisation of Sinhala Legal Documents Authors: Minduli Lasandi, Nevidu Jayatilleke Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32397v1 Legal document summarisation in low-resource languages presents significant challenges due to the scarcity of annotated data and the complexity of domain-specific terminology. This paper presents SinBrief, a hybrid abstractive summarisation framework for Sinhala legal documents that does not require human-annotated training data. The proposed framework combines domain-aware word graph construction with neural sentence scoring to generate abstractive summaries from Sinhala legal text. Five sentence scoring models are evaluated within the framework: mBert, Llama 3.1, Falcon 7B, Laser, and a continually pre-trained Llama model domain-adapted to Sinhala legal text. The framework is evaluated on a Sinhala legal corpus using reference-free metrics, including Coverage, Density, Compression Ratio, SummaC, and Self-BertScore. Experimental results demonstrate that SinBrief produces summaries with lower lexical overlap than extractive baselines while maintaining factual consistency, demonstrating the viability of hybrid, largely annotation-free abstractive summarisation for low-resource legal NLP tasks. --- ### FA-Bench: A Benchmark for Word-Level and Phone-Level Forced-Alignment and ASR Timestamps Under Clean and Noisy Conditions Authors: Wei Chu, Yuanzhe Dong, Ke Tan, Dong Han, Yichao Zhou, Ruchao Fan, Bingshen Mu, Jingbei Li, Vishwas Shetty, Sarthak Bisht, Ziyue Qiu, Massa Baali et al. Published: 2026-09-26 Categories: cs.CL, cs.PF arXiv: https://arxiv.org/abs/2609.32396v1 Forced alignment aligns speech audio with a text transcript to generate word and phone timestamps. Published comparisons normalize transcripts, split the data and match boundaries differently, so their numbers cannot be read together. We present FA-Bench, an open framework that fixes those choices once and releases the code, splits, phone mapping, text normalization and scoring script, with results published periodically. Track 1 gives every aligner the reference transcript and Track 2 gives it a recognizer's output, on the same audio, clean and degraded four ways, with 21 open models and 9 commercial APIs under a unified protocol. We score every boundary of an utterance and check the two labels beside it, so a word the recognizer missed or invented is charged. Using a tolerance-based F1 as our primary metric eliminates the 9% to 14% score inflation that standard MAE causes on recognition-dependent systems in conversational speech. We then group boundaries by their position and how many adjacent words were recognized correctly, which shows where a system lost the score. We discovered systematic bias in how current systems time words, with Whisper about 150 ms early and several commercial ASR APIs over 50 ms late. Code and results are at https://github.com/olewave/fa-bench --- ### Memory as a cache: Exact context reuse and deletion by construction Authors: Shengyao Wang, Jiang Liu Published: 2026-09-26 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32395v1 The KV cache of a transformer entangles every token's representation with its entire prefix: a passage encoded once cannot be reused under a different prefix or removed without recomputing everything after it, so exact cache reuse is limited to shared prefixes. We present SMem, an architecture whose context representation is a cache by construction. A block-local encoder maps each block to memory rows independently of other blocks, and a reader conditions generation on their union through cross-attention. For every parameter setting, memory composes exactly at fixed block indices, deleting a block is an exact $O(b)$ update for $b$-token blocks, and the memory state is independent of the edit path. At $4\times$ the training context, under the shared recipe, SMem retrieves planted needles beyond any trained-length window (exact match 0.14-0.28 at distances of 31 and 63 blocks), where learned-position, RoPE, and Block-Attention-style transformers all score at most 0.02. A fully cached context is served by computing one block alone at a near-constant 3.1-6.2 ms, whereas cold prefill grows with context; batched decode stores 34-38% fewer KV rows and runs 1.4-1.7$\times$ faster when bandwidth-bound; and deletion beats suffix recomputation by 8.5$\times$ at 512 blocks and 452$\times$ at 4096 blocks (32-256$\times$ the trained length, probing the cost model rather than a served regime). The cost is a perplexity gap of -4.7% to +2.8% (negative favors SMem) against a parameter-matched transformer with the same positional scheme, at 160M-1.5B on FineWeb-Edu across two recipes and a learning-rate search. SMem also composes with RoPE: at 160M and 410M the composite matches or leads the matched transformer and closes 29-59% of SMem's gap to a RoPE transformer. Dropping prefix entanglement thus keeps perplexity comparable while making the cache exactly composable and editable. --- ### Beyond Scripted Search: Sample-Efficient Reward Discovery via Agentic Black-box Optimization Authors: Minghao Li, Rui Tan, Ruihang Wang Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32394v1 Designing dense reward functions for low-level reinforcement learning (RL) control remains difficult. Recent work uses large language models (LLMs) to iteratively generate and refine reward functions using policy-training feedback within scripted search algorithms. However, evaluating each candidate requires a full RL training run, making sample efficiency a central challenge for reward search on complex control tasks. To address this limitation, we propose an Agentic Reward Black-box Optimization (ARBO) framework, in which an LLM agent builds the search strategy at run time from an evaluation history maintained as its persistent workspace. The evaluation history comprises two components: observations maintained by the evaluation oracle, including candidate scores, per-term training curves, and error tracebacks; and an agent-maintained belief that records diagnoses and intended next steps. The agent queries both with tools and generates the next batch of reward candidates, rather than generating them in a single pass from a fixed prompt. Across four control domains, ARBO achieves gains of 29.9% in manipulation success rate and 192.8% in power-grid score over baseline means under a shared evaluation budget. Ablations examine each component's contribution and sensitivity to backbone choice. --- ### Convergent Plug-and-Play Image Restoration with Annealed Noise Levels Authors: Samuel Hurault Published: 2026-09-26 Categories: math.OC, cs.LG arXiv: https://arxiv.org/abs/2609.32393v1 Plug-and-Play (PnP) methods solve imaging inverse problems by incorporating deep denoisers into iterative optimization algorithms. Although practical implementations often decrease the denoiser noise level $σ$ along iterations, most existing convergence analyses assume a fixed denoiser. In this work, we establish convergence guarantees for a broad family of Plug-and-Play algorithms with annealed noise level, spanning deterministic methods (RED--GD and PnP--PGD) and stochastic methods (SNORE, equivariant RED, and a variant of PnP--Flow). For each method, we identify an explicit, nonconvex objective associated with the terminal denoising level and prove asymptotic stationarity of the iterates with respect to this objective. Our analysis does not prescribe any decay rate for the noise schedule, and our assumptions cover both learned gradient-step denoisers and exact MMSE denoisers. Overall, our theoretical results bridge the gap between existing PnP convergence theory and the decreasing-denoising practices used by state-of-the-art image restoration methods. We empirically demonstrate the benefits of such schedules and illustrate the predicted convergence behavior on several imaging inverse problems, including inpainting, super-resolution, demosaicing and tomography. --- ### SCLATE: a Substrate for Continual-Learning Agent Training and Evaluation Authors: Youngmok Jung, Sirajul Salekin, Henry Tran, Javier Movellan, Zhao Huang, Manjot Bilkhu Published: 2026-09-26 Categories: cs.AI, cs.DC, cs.LG arXiv: https://arxiv.org/abs/2609.32391v1 Continual-learning agents are systems of models, harnesses, and memory operating over long multi-session horizons. Evaluating and training them requires interleaving tasks with agent-side events such as session stop and start, crons, and memory consolidation. Yet existing benchmarks and training frameworks schedule only the benchmark's own events, leaving each benchmark and agent pair to build a custom scheduling loop. We present SCLATE, an execution substrate where benchmarks and unmodified agents each add their events to one open event scheduler through an adapter. A hybrid simulated clock runs these events on a shared timeline, flowing in real time while the agent works and skipping idle gaps, which compresses a month-long scenario into hours. SCLATE also serves as a rollout engine that runs any agent's harness and memory unmodified, recording the tokens and log probabilities of every model call through an in-container proxy. We port seven benchmarks to SCLATE and compare ten unmodified harness and memory configurations head to head on ten models. The comparison shows that an added memory system does not reliably beat the harness's native memory and that models differ widely in how they use the same harness and memory. We then post-train Qwen3.5-4B through unmodified harnesses and memory systems. The model learns to use both, reading 6.8x fewer file lines with a 16.7-point higher SWE-bench Verified pass rate, and writing richer memory records, while its held-out MetaClaw accuracy rises by up to 11.8 points. --- ### Reward Hacking and Agent Containment Failure: A Monte Carlo Study Based on the 2026 Hugging Face Incident Authors: Murat Ozer, Bulent Erenay, Ibrahim Berber Published: 2026-09-26 Categories: cs.AI, cs.CR arXiv: https://arxiv.org/abs/2609.32390v1 The July 2026 intrusion into Hugging Face production infrastructure showed how reward hacking can become an external cybersecurity incident when a capable agent encounters weak containment boundaries. This study develops a probabilistic risk model linking five stages: reward hacking, containment escape, usable access, persistence, and failure of detection. A Monte Carlo simulation evaluates 100,000 runs under each of four control configurations. Input distributions represent explicit uncertainty and are used for comparative analysis rather than real-world frequency prediction. Under the stated assumptions, layered controls reduce simulated external-incident probability substantially more than network isolation or monitoring used alone, an ordering that holds under independent plus/minus 25% perturbation of every coefficient in the model across 300 draws. Sensitivity analysis shows that agent capability and weaknesses in monitoring, authorization, and credential control exert the greatest influence on modeled risk. Human temporal discounting and metric gaming provide a behavioral analogy for short-horizon optimization, but the study does not infer that AI agents experience gratification or human motivation. The results support treating cyber-capable agent evaluations as hostile security zones in which indirect egress, shared infrastructure, credentials, and evaluation artifacts must remain outside the agent's effective authority. --- ### RefCompose: Multi-Reference Image Generation via LoRA-Conditioned Diffusion Authors: Sai Sri Teja Kuppa, Parth Shinde, Priyadharsan Balaji S, Jinka Harshavardhan, Sriprabha Ramanarayanan Published: 2026-09-26 Categories: cs.CV, eess.IV arXiv: https://arxiv.org/abs/2609.32389v1 Filmmakers and visual artists routinely need to compose multiple references, actors, locations, props, cultural elements, into a single coherent shot, but existing tools either fail to scale past a handful of references or destroy fine grained subject identity in the process, since per reference tokenization scales memory linearly with reference count $N$ and generated content often departs from the given references rather than reproducing them. We propose \textbf{RefCompose}, a pixel space compositional conditioning framework that decouples \emph{where} things go from \emph{what} they look like, via a single fixed resolution reference canvas that keeps conditioning size constant regardless of reference count. Spatial layout is induced at inference time from a frozen diffusion transformer and extracted via Grounding DINO, requiring no LLM or dedicated layout model, while dual stream LoRA adapters inject a layout derived depth map and the encoded canvas through separate low rank streams, disentangling geometric scaffolding from localized appearance. On the Dense Layout protocol, RefCompose consistently outperforms layout based and state of the art multi reference baselines on color, texture, shape, spatial accuracy, and identity/content preservation at higher reference counts, all with constant inference memory, making it a practical building block for multi subject cinematic composition at production scale. --- ### Using Machine Learning to Investigate Predictors of Fasting Blood Glucose: Insights into Circadian Timing and Age Interactions Authors: Viktoriya Bu-Dager, Silvia Cirstea Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32386v1 Impaired glucose regulation is a major contributor to metabolic dysfunction and type 2 diabetes. This study developed an interpretable machine-learning framework to predict log-transformed fasting blood glucose using metabolic, hormonal, lifestyle, demographic, nutritional, and circadian variables from the National Health and Nutrition Examination Survey 2017--2020 pre-pandemic dataset. After merging multiple NHANES sub-datasets, data processing used a leakage-resistant pipeline in which imputation, scaling, and one-hot encoding were performed only after dataset splitting and within training folds. Elastic Net, LASSO, and XGBoost models were evaluated using 94 candidate predictors and engineered circadian interaction terms. Performance was assessed using mean absolute error, root mean squared error, coefficient of determination, calibration, and Shapley Additive Explanations. The final interaction-augmented XGBoost model achieved strong performance on the independent test set, with a mean absolute error of 0.0804, a root mean squared error of 0.1148, and a coefficient of determination of 0.7761, using 10 predictors. Glycohemoglobin was the dominant predictor, followed by insulin, diabetes diagnosis, gamma-glutamyl transferase, age, race, and gender. Among the engineered interaction terms, sleep midpoint multiplied by age was consistently retained in repeated random-split analyses, although its contribution remained modest relative to dominant glycaemic predictors. These findings support further investigation of circadian-age interactions in metabolic health. --- ### DashAct: A Progressive Diagnostic Benchmark for GUI Agents in Interactive Dashboard Analysis Authors: Chuhan Zhang, Qi Xie, Ziyue Wang, Jianing Yin, Yunfan Zhou, Dazhen Deng, Yingcai Wu Published: 2026-09-26 Categories: cs.HC, cs.AI arXiv: https://arxiv.org/abs/2609.32385v1 Interactive dashboards require users to reveal and connect evidence across stateful interactions. Although graphical user interface (GUI) agents could automate this process, existing dashboard benchmarks primarily report final answers or task success. They provide limited insight into whether failures arise from maintaining the analytical process, selecting actions, or grounding visual targets. We introduce DashAct, to our knowledge the first benchmark to diagnose these failures at a fine-grained level within the same dashboard task. DashAct contains 357 human-verified interaction trajectories with milestone dependencies and hierarchical target annotations. Its progressive diagnostic cascade evaluates end-to-end execution, restores verified context for next-action prediction, and provides target semantics and a local view for visual grounding. By progressively restoring the conditions for success, DashAct measures the minimum support an agent needs to recover rather than scoring isolated skills. Experiments show that current models struggle even as support is added. The cascade outcomes reveal bottlenecks hidden by end-to-end scores and provide actionable guidance for improving GUI agents. --- ### TimeES: Probabilistic and Deterministic Time Series Forecasting via Evolutionary Spectra Authors: Weiwei Ye, Renhe Jiang, Hangchen Liu, Dongyuan Li, Yoshihide Sekimoto Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32384v1 Real-world time series are inherently non-stationary, with trends, periodic patterns, and uncertainty evolving over time. While the Fourier domain offers a natural lens to model time series, current deep learning approaches do not explicitly model evolution and randomness in the Fourier spectra, which limits their ability to accurately predict both the expected trajectory and its uncertainty in non-stationary time series. Motivated by Evolutionary Spectra (ES) theory, we propose TimeES, a general framework that enables probabilistic and deterministic forecasting via the evolutionary spectra theory. Specifically, we derive a parameterizable evolutionary spectra formulation, recasting non-stationary random process modeling as learning an evolving representation modulated by random variables. Furthermore, we reduce the complexity of the estimated spectra from O(NM) to O(NK), where K << M/2, by exploiting Hermitian symmetry and spectral energy sparsity for frequency selection. Based on a simple linear backbone, our proposed TimeES achieves consistent state-of-the-art performance across both deterministic and probabilistic forecasting tasks, with high efficiency and interpretability. Code is available at: https://github.com/wwy155/TimeES. ---