# Research Papers — 2026-09 · bundle page 22 of 266 > Full-text markdown of papers #1051–1100 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/22 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/21.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/23.md) ### Beyond Semantic Narrowing: Robust and Efficient LLM Watermarking with Hamming Neighborhoods Authors: Zewen Sun, Tongyang Zhao, Liyao Xiang, Mingxuan Ma, Lingzhe Wang, Zhiyuan Li Published: 2026-09-29 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.37218v1 Semantic watermarking improves robustness against watermark removal attacks by embedding detectable signals into sentence-level representations. However, existing watermarking methods typically impose watermark-specific semantic preferences on generated sentences without explicitly accounting for the highly non-uniform and context-dependent semantic preference of LLM generation. When these two preferences are poorly aligned, many natural continuations become incompatible with the watermark, causing semantic narrowing: reduced semantic freedom, increased resampling cost, and potential degradation on tasks with strict semantic requirements. To alleviate this problem, we propose HammingMark, which uses the semantic hash of the preceding sentence as a dynamic center and accepts candidates whose hashes fall within its Hamming neighborhood. Defining watermark validity over a Hamming neighborhood in compact hash space retains a larger fraction of naturally likely semantic continuations. The coarse many-to-one hash mapping further allows diverse semantic realizations to remain watermark-valid. Experiments on C4 and BookSum show that HammingMark achieves strong robustness, high detectability, and near-unwatermarked generation quality, requiring only 2.2 sampled candidates per accepted sentence,a 72.8% reduction compared with the most sampling-efficient existing method. On more complex tasks with strict semantic constraints, HammingMark achieves the highest detection rates with the highest or tied-highest ROUGE-L scores, demonstrating its effectiveness in balancing watermark detectability and generation quality under constrained generation settings. --- ### CRJudgeBench: Can AI Detect Plausible but Invalid Code Reviews? Authors: Yue Pan, Jiawei Li, Ziyuan Zhang, Xiangxin Zhao, He Ye Published: 2026-09-29 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2609.37216v1 Large language models can generate plausible code-review comments, but such comments may contain technically incorrect claims that mislead developers. We study technical trustworthiness judgment: determining whether a review comment's core technical claims are correct and applicable to the reviewed code in its repository context. Existing code-review benchmarks primarily evaluate review generation, issue discovery, or general comment quality, but do not directly assess whether an agent can determine the technical trustworthy of an individual review comment. To fill this gap, we introduce CRJudgeBench, a benchmark of 1199 instances constructed from real pull requests and expert-verified perturbations, covering both trustworthy and plausible but untrustworthy comments. We further present Sentinel, a repository-grounded agentic judge that actively gathers code evidence to verify review comments before making judgments. Starting from Qwen3-Coder-30B-A3B-Instruct, Sentinel is trained on the CRJudgeBench training split through iterative action-level learning from a privileged teacher. On the 359-instance CRJudgeBench test set, Sentinel achieves 76.60\% accuracy, outperforming GLM-5.3 by 6.13 percentage points and its base model by 19.78 points. These results show that even state-of-the-art general-purpose LLMs struggle to identify untrustworthy comments, while iterative action-level learning substantially improves the accuracy of repository-grounded trustworthiness judgments. Our dataset is available at https://huggingface.co/datasets/dcloud347/CRJudgeBenchmark --- ### Explainable Machine Learning for Multilayer Planar Winding Inductance Estimation Authors: Spyros Rigas, Theofilos Papadopoulos, Georgios Alexandridis, Antonios Antonopoulos Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37211v1 Rapid and accurate self-inductance estimation for multilayer rectangle-shaped planar windings is essential for modern high-frequency power converters, yet traditional workflows rely on complex mathematical equations, rigid monomial formulas or unexplainable black-box machine learning (ML) models that degrade severely outside their training domain. This paper introduces an explainable ML framework unifying post-hoc feature attribution (SHAP and permutation importance) with Kolmogorov-Arnold Network-guided symbolic regression via the SR-KAN framework to discover closed-form analytical equations without prior structural assumptions. Evaluated on a new open-source dataset of over 10,000 Finite Element Analysis (FEA) simulations across seven out-of-distribution (OOD) classes, standard tree-based ensembles exhibit severe extrapolation errors (> 36%), whereas the unconstrained SR-KAN expression achieves a robust OOD relative error of 8.22%. Experimental verification across 55 physical printed circuit board prototypes (up to 8 layers, with inductances from 4.11 μH to 559.27 μH) confirms that the KAN-discovered expression translates effectively to real-world hardware, predicting inductance with a mean absolute relative error of 6.26%. To support reproducible research, the complete FEA simulation dataset and prototype measurements are released open-source. --- ### Pointwise or Pairwise: When Do Pairwise Losses Help Reward Learning, Provably? Authors: Junghyun Lee, Minsoo Ha, Sanghwa Kim, Yeongjong Kim, Eunjee Lee, Seiyun Shin, Kwang-Sung Jun Published: 2026-09-29 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2609.37209v1 Pairwise losses are increasingly used for reward learning even when pointwise rewards are observed, with mixed empirical results. When and why do pairwise losses outperform pointwise losses? We study this question in a grouped offline contextual-bandit setting allowing multiple actions per context, capturing many reward learning scenarios. We compare Value Regression (VR), which regresses observed rewards pointwise, with Value Difference Regression (VDR), which regresses reward differences between a pair of actions sampled under the same context. We consider a semiparametric model where the mean reward is the sum of a learnable action-dependent component and an arbitrary context-dependent yet action-independent nuisance, capturing context-specific disturbances. Using a unified localized analysis, we prove finite-sample regression guarantees for finite and linear function classes and translate them into offline-regret bounds. For finite classes, VDR eliminates the misspecification term in the VR bound and improves a reward-scale-dependent error term by averaging over actions within each context, a benefit absent from the corresponding VR term. For linear classes, neither method uniformly dominates: within-context differencing removes nuisance-induced bias but may increase estimation variance relative to using absolute rewards when the misspecification is sufficiently low. This yields a feature geometry-dependent bias-variance tradeoff, which we corroborate with numerical experiments. --- ### Learning to Prove, Not Just to Answer: Reinforcement Learning from Formal Verification for Natural-Language Logical Reasoning Authors: Qili Zhang, Qianren Mao, Hanze Cai, Kaiming Zhao, Yuening He, Xihan Lei, Yashuo Luo, Hanwen Hao, Yutong Gu, Likang Xiao, Zhijun Chen, Weifeng Jiang et al. Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37203v1 Large language models (LLMs) are increasingly deployed for natural-language logical reasoning, where the final answer is easy to check but the proof behind it is not. In natural-language logical reasoning, an intermediate conclusion should follow from its premises, and the resulting derivation should support the final answer. Existing methods lack machine-checkable verification of intermediate conclusions and answer-supporting proof dependencies, so they may assign credit to invalid or answer-irrelevant steps. We propose Proof-R1, an RL framework from formal verification that trains LLMs to construct verifiable proofs for natural-language logical reasoning. Proof-R1 admits a generated conclusion into the verified proof state only when the corresponding reasoning action satisfies the proof obligations through UNSAT-based machine-checkable formal verification. Proof-R1 also recovers the answer-supporting dependency closure to trace the proof structure of the final answer and align outcome credit with the proof dependencies. Experiments demonstrate that Proof-R1 improves answer accuracy across three logical reasoning benchmarks and four backbone models and outperforms training-free agents and training-based methods in terms of reasoning-process verifiability. --- ### Probabilistic Symbolic-Distillation Model of Droplet Collision for Spray Simulation at High Ambient Pressures Authors: Weiming Xu, Tao Yang, Peng Zhang Published: 2026-09-29 Categories: physics.flu-dyn, cs.LG arXiv: https://arxiv.org/abs/2609.37202v1 Droplet collision governs droplet population dynamics in many chemical engineering processes, such as spray drying, spray cooling, agricultural spraying, and combustion. Existing analytical models impose deterministic, pairwise boundaries between collision outcomes, whereas machine-learning classifiers lack the explicit functional form required of analytical collision submodels. In this study, we develop a probabilistic symbolic-distillation model using nearly forty thousand experimental events spanning eight regimes and five dimensionless parameters, including over five thousand data for ambient pressure up to 50 atm. A machine-learning teacher learns the joint outcome-probability landscape from these data, and symbolic regression subsequently distils it into eight class-specific expressions that jointly define a coupled analytical model. The resulting analytical field replaces abrupt regime switching with finite-width fuzzy boundaries. It outperforms the evaluated conventional analytical boundary models and reveals that their main limitation is the inability of zero-width boundaries to represent gradual probability transitions. The "biased-dice" sampling scheme provides a statistically consistent and practically convenient model implementation for Eulerian-Lagrangian spray simulation. --- ### Adaptive Reward Routing: Dynamic Multi-Reward Optimization for Joint Audio-Video Diffusion via Forward-Process RL Authors: Songlin Yang, Xiaotong Zhao, Jiacheng Zhang, Zhe Wang, Toyota Li, Eric Liu, Alan Zhao, Anyi Rao Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37200v1 Multi-reward guided reinforcement learning (i.e., RL) offers a promising way to improve joint audio-video diffusion models along several complementary objectives, including modality-specific quality, cross-modal semantic alignment, and temporal synchronization. Its effectiveness, however, depends on two quantities that change during training: where reward-driven updates should act, and how competing rewards should be combined. Existing methods tend to rely on fixed routing and reward weights, failing to track evolving model functions. To address these limitations, we propose Adaptive Reward Routing to jointly adapt update locations and reward coordination during forward-process RL (i.e., DiffusionNFT) of joint audio-video diffusion models. Our method consists of two components. (i) Cross-Modal Influence-Guided Routing (Localizing Updates): We use bidirectional cross-attention responses as an efficient proxy for evolving cross-modal influence, dynamically reweighting token-aware losses and scaling gradients across cross-modal layers without additional model interventions. (ii) Preference-Preserving Modality-Aware Reweighting (Coordinating Rewards): We preserve predefined weights as preference priors and use branch-specific reward-gradient interactions as residual corrections after warm-up. This resolves evolving conflicts without letting dominant rewards suppress weak but essential objectives. Extensive experiments demonstrate consistent improvements in modality quality, semantic consistency, and audio-video synchronization over strong RL baselines. Ablations and mechanism analyses further validate the complementary benefits of adaptive update routing and reward coordination. --- ### V-Engram: Trigger-Indexed External Memory for Modular Text-to-Image Personalization Authors: Haoran He, Runyuan Cai, Yiming Wang, Lin Yu, Xiaodong Zeng Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37198v1 Pretrained text-to-image models contain broad visual knowledge, yet they cannot reliably acquire or refine a specific visual identity from only a few references while preserving compositional control. Token-embedding methods are compact but often underfit identity, whereas adapter-based methods improve fidelity through persistent weight updates that can be costly to store and interfere when concepts are composed. We introduce V-Engram, a trigger-indexed external memory mechanism for Stable Diffusion 3.5. Each concept is assigned an explicit trigger that retrieves concept-specific memory, whose gated directions enter frozen text-encoder and MMDiT context states as relative residuals. Separating this memory from backbone adaptation enables prompt-selective and multi-concept access without merging model updates. Experiments show that V-Engram broadly matches DreamBooth-LoRA in overall subject fidelity while showing advantages in settings such as contextual subject preservation. Prompt-matched loading retrieves only matched entries, reducing most additional adaptation-state loading for a single-concept query. Qualitative results further demonstrate paired-trigger composition and same-class separation, while prompts without registered entries retain the frozen model's base behavior. Together, these results establish trigger-indexed memory as a modular interface for adding targeted visual evidence without rewriting the generator. --- ### ToolFence: Fine-Grained Authorization for Secure Tool-Using LLM Agents Authors: Yanjie Li, Xiangyu He, Xuelong Dai, Bin Xiao Published: 2026-09-29 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.37196v1 Tool-using LLM agents remain vulnerable to indirect prompt injection because trusted instructions and untrusted observations share one context, allowing malicious content to steer consequential input-filtering defenses. Multi-path consensus defenses still leave a high attack success rate because they examine content or aggregated outputs rather than authorizing effects, especially for the within-tool attack, which preserves the intended tool but manipulates its arguments. Data-Flow Control such as CaMeL provides stronger guarantees, but incurs substantial time latency that limits practical deployment. We introduce ToolFence, which compiles a typed authorization blueprint before execution, enforces it through a deterministic monitor, and when the blueprint is incomplete asks a judge to grant new capabilities rather than adjudicate each concrete call. ToolFence provides two key advantages. First, its fine-grained provenance-aware authorization enables the system to distinguish user-authorized values from untrusted observations, effectively addressing the within-tool attack. Second, its deterministic fast path and capability-level runtime grants substantially reduce the frequency of expensive judge calls, improving runtime efficiency. On AgentDojo with Qwen3-max, ToolFence reduces overall ASR to near zero with only a 3.80 percentage-point clean-utility drop and practical runtime overhead. --- ### Exploring In-Context Learning for Handwritten Text Recognition Authors: Eric Ayllon, Abel Gandia, Jorge Calvo-Zaragoza Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37195v1 Handwritten Text Recognition (HTR) systems have become an indispensable tool for the digitization of historical documents. Not only do they cut down time and cost, but they also allow democratizing access and processing of their contents by generating their transcripts. However, literature in HTR currently focuses mostly on specialized models that require large amounts of annotated samples to achieve satisfactory performance. We explore the use of In-Context Learning with pre-trained Vision-Language Models (VLMs) to create a transcription pipeline without updating the model's parameters. We then evaluate this pipeline across multiple collections and models, and demonstrate that general-purpose VLMs can be effectively taught how to transcribe handwritten text from images. To assess how our observations may translate to practical applications, we evaluate the performance in a Cross-Domain (CD) scenario, where context examples are drawn from a different collection than the query image. Results in both the controlled In-Domain (ID) scenario and the realistic CD scenario follow the same patterns. First, as context size grows, the error range is expected to narrow towards the average performance. Thus, larger context sizes sacrifice the performance of the oracle-best sampling for lower expected error rates. The results obtained show that, without any parameter updates, this methodology has strong potential to compete with traditional HTR in the presence of domain shift. Moreover, we show and argue that some context samplings work better than others and suggest more effort should be put into finding an ideal sampling method in future work. --- ### Adaptive Rotation for iSOMA: Geometry, Benchmarking, and Noise Robustness in Variational Quantum Objectives Authors: Vojtěch Novák, Ivan Zelinka Published: 2026-09-29 Categories: cs.NE, quant-ph arXiv: https://arxiv.org/abs/2609.37193v1 We study whether the coordinate dependence of the improved Self-Organizing Migrating Algorithm (iSOMA) can be reduced while retaining its inexpensive leader-directed migration mechanism. We introduce iSOMA-AR, which learns a basis from successful migration displacements and selectively applies the standard perturbation mask in that basis. On the complete noiseless BBOB suite, iSOMA- AR significantly outperformed baseline iSOMA across matched conditions, with the largest gains on geometrically difficult landscapes. A targeted ablation shows that the learned orientation is beneficial on a rotated ill-conditioned landscape and that moderate changes of the gate threshold and rotation cap preserve the qualitative result. On CEC 2011 Real World Optimization Problems, iSOMA-AR outperformed iL-SHADE on most problems, although its advantage over baseline iSOMA was not statistically significant. A canonical-jSO rerun is reported as a post-hoc sensitivity check alongside the original jSO-derived comparator. On frustrated-spin variational quantum objectives, adaptive rotation improved most transverse-field conditions, while gains on the diagonal and anisotropic models were absent or selective. Under strong effective sampling noise, the SOMA variants were the most robust population-based methods in the comparison, but iSOMA-AR was not significantly better than baseline iSOMA. Repairing all-zero PRT masks greatly reduced repeated-point evaluations without changing endpoint quality significantly, making this implementation detail unlikely to explain the noise result. Overall, adaptive rotation is most useful on coordinate-sensitive deterministic problems, while the observed noise robustness appears to arise mainly from the underlying SOMA migration mechanism. --- ### HaPRL: Human-Anchored Process Reinforcement Learning for Visual Search Agent Authors: Zhangquan Chen, Yaoxin Niu, Xiang An, Mingze Sun, Zhumei Wang, Chih-Ting Liao, Hongkun Cao, Ruqi Huang Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37190v1 Multi-turn visual search agents answer questions about high-resolution images by iteratively deciding where to look. Reinforcement learning for these agents rewards only the final answer, leaving the search process unsupervised. Consequently, faulty routes in which the reasoning process is erroneous yet the final result is correct arise frequently, which in turn leads to ineffective training, i.e., scaling along the wrong paths. In this paper, we introduce HaPRL, the first framework to reinforce the search process with human search behavior. We first build an annotation platform and collect 1K+ human-annotated data with fine-grained behavioral signals. During training, a carefully designed judge scores each rollout with task-adaptive weights, anchored on the distilled trace of how a human annotator actually searched the same image. Extensive experiments show that HaPRL consistently outperforms outcome-based RL, and early-stage process supervision yields 6.7x more improvement in subsequent outcome-based scaling. Our results also demonstrate the importance of aligning model behavior with human process annotation signals, which offer new insight into the training of foundation models. --- ### Corruption-Robust Sparse Linear Contextual Bandits with Knapsack Constraints Authors: Yige Wang, Hanyang Li, Yiming Zong, Wanteng Ma, Jiashuo Jiang Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37189v1 We study sparse linear contextual bandits with knapsack constraints under joint reward and consumption corruption. Consumption corruption creates a challenge beyond corrupted rewards: it affects not only statistical estimates, but also the recorded budget, resource prices, and stopping decisions that govern future allocation. We develop Robust Optimistic Primal--Dual (ROPD), an estimator-modular framework that combines corruption-aware confidence widths with online resource prices and a budget-safety rule. With concrete sparse implementation, ROPD achieves regret against a clean population-LP benchmark of $\widetilde O(T^{2/3}+ΓT^{1/3})$ under forced exploration and population-design coverage, and $\widetilde O(\sqrt T+Γ)$ under on-policy realized-design coverage, for a supplied valid corruption bound $Γ$ under the stated proportional-budget scaling and fixed model/design parameters. When the corruption level is unknown, Shared-Grid adapts confidence radii around common point estimates fitted to a single realized history, incurring explicit initialization and master-comparison costs; its sharper on-policy guarantee additionally requires recommendation coverage. Both methods preserve observed budgets on every realization and bound clean resource violation by cumulative consumption corruption. These results connect corruption-robust sparse estimation with resource accounting, pricing, and stopping in high-dimensional online allocation. --- ### InsightMap: Structured Spatial Modeling for Embodied Multimodal Reasoning Authors: Hongpei Zheng, Hujun Yin Published: 2026-09-29 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2609.37187v1 Language-guided navigation requires connecting partial observations to a persistent spatial reference and learning how actions change that representation. We introduce InsightMap, a framework that uses top-down maps as both explicit spatial memory and action-conditioned prediction targets. Historical views are linked to labeled map locations, and a shared multimodal backbone jointly learns navigation action prediction and post-action map generation. Map prediction provides auxiliary training supervision, while navigation inference decodes actions from the observed spatial context. An aligned RGB-D data pipeline supports a common interface for navigation, visual question answering, situated reasoning, and 3D grounding. On the validation-unseen splits of R2R-CE and RxR-CE, InsightMap achieves success rates (SR) of 56.9% and 54.9%, respectively. Adding map-prediction supervision improves R2R-CE SR by 4.3 and success weighted by path length (SPL) by 3.2 percentage points. On static spatial tasks, InsightMap achieves 103.7 CIDEr on ScanQA, 60.1% exact-match accuracy on SQA3D, and 53.1% grounding accuracy at 0.5 IoU on ScanRefer with detected object proposals. On Unitree Go2, it outperforms NaVid and NaVILA in hallway, lab, and office environments. --- ### Accelerated surrogate dynamics for dynamical, stochastic system evolution Authors: Marco Jochum, Ioannis Kouroudis, Gohar Ali Siddiqui, Taher Amine Hamzaoui, Manuel Gößwein, Alessio Gagliardi Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37184v1 Dynamic simulations are an entrenched way of gaining insight into the evolution of system dynamics. Their computational cost however is often prohibitively high, especially in cases of stochastic frameworks. Machine learning algorithms are especially suited as simulation surrogates. Nevertheless, they face some very distinct limitations. Firstly, the sheer dimensionality of these systems, however, precludes the use of traditional time series models who struggle with high dimensional feature spaces. Additionally, traditional time series focus exclusively on either long or short range effects, causing local or global drift given enough time. In this paper, we propose a framework that addresses those limitations. Our framework combines a Variational Autoencoder, with a convolutional or graph basis that reduces the dimensionality of the system. This latent vector is propagated in time using a Temporal Fusion Transformer model, which includes both long range and short range effect encoding, as well as static covariate support. We test our framework on three distinct cases, to prove its robustness and in all three we have achieved practically identical to the simulation results at a fraction of the time. Further, our framework is flexible enough to be adapted to any new system and provides an inbuilt uncertainty quantification for targeted experiment design. --- ### EgoHumanoid-V2: Human-to-Humanoid Transfer of Coordinated Whole-Body Skills for Loco-Manipulation Authors: Jin Chen, Yiming Jiang, Chongyang Xu, Modi Shi, Shijia Peng, Li Chen, Tianyu Li, Mu Xu, Yilun Chen, Steven Hoi, Hongyang Li Published: 2026-09-29 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.37181v1 Human demonstrations capture diverse scenes and rich whole-body skills without requiring robot teleoperation. Prior work on egocentric transfer has emphasized scene generalization in loco-manipulation under decoupled control, leaving direct transfer of coordinated whole-body skills less explored. We present EgoHumanoid-V2, the first egocentric human-to-humanoid skill transfer framework for coordinated whole-body loco-manipulation. At its core, coarse-to-fine action alignment combines kinematic reference correction with dynamics-aware refinement. It improves end-effector pose accuracy while preserving whole-body coordination. We also use robot-arm rendering and training-time image augmentation to reduce the visual embodiment gap and improve viewpoint robustness. On four real-world tasks, vision-language-action (VLA) policies trained on aligned human data show zero-shot skill transfer without target-task robot demonstrations. Task scores are comparable to those of policies trained on teleoperation data at a lower collection cost. These results support human data as direct skill supervision. --- ### Sparse cubical complexes for efficient topology-preservation in image data Authors: Alexander H. Berger, Marco Fontana, Daniel Rueckert, Johannes C. Paetzold, Laurin Lux, Ulrich Bauer Published: 2026-09-29 Categories: cs.CV, cs.CG, cs.LG arXiv: https://arxiv.org/abs/2609.37177v1 Persistent homology (PH) is a frequently used tool for extracting and preserving topological information from image data, particularly in image segmentation, where preservation of topological structures is important. However, despite its general applicability across dimensionality, domains, and target structures, the runtime cost of PH-based methods often makes their practical use infeasible. In this work, we argue that this runtime cost is largely driven by processing information that is unimportant for downstream application (e.g. as optimization objective). We propose sparse cubical filtrations as an alternative foundation for PH computation, reducing subsequent computational costs by factors of up to 100 on real datasets. We show close agreement with the optimization signal of the dense counterpart and empirically evaluate our solution's effectiveness as an optimization objective in realistic training regimes where other PH-based objectives can practically not operate (i.e., 3D data with large patch sizes). We show how our solution improves topological accuracy by up to 80\% across six diverse datasets while maintaining pixel- and region-based accuracy. --- ### Absorbed in Inertia: Activation Analysis for Computer-Use Agents Authors: Giulio Segalini, Zhi Wen Soi, Jérémie Decouchant, Lydia Chen Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37176v1 Computer-use agents have become increasingly capable of executing tasks on live desktops through natural-language instructions, based on trajectories of screenshots, actions, and reasoning. We discover that they can stealthily exhibit inertia, in which they repeat fruitless actions despite recognizing that these actions are ineffective. We hypothesize that inertia is reflected in the agent's internal state, i.e., the activation values of the agent's underlying model, and propose a protocol to measure the relationship between the two. Extensive analysis of high-dimensional activation states shows that inertia corresponds to an absorbing region of activation space, where activation values become stale across actions and even after attempts to steer them. We conjecture that drastically changing the agents' activations by re-initializing them is necessary to escape inertia. Specifically, we propose R$^3$ (Reset, Reroute, Restore), which temporarily resets the agent's context trajectory to escape the absorbing region and then restores the historical context to effectively complete the task. Our approach yields 17-55% lower measured inertia across models relative to unmodified agents. These results suggest that changing the context can interrupt recurrence more effectively than directly steering the resulting activations. Our code is available at https://anonymous.4open.science/r/vlm-agent-defense-D076 --- ### VLM Fine-Tuning for End-to-End Combinatorial Optimization Authors: Qingsong Yan, Xia Jiang, Yaoxin Wu, Wen Song, Lu Zhang, Yingjie Zhou Published: 2026-09-29 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.37175v1 Large language models (LLMs) have provided a unified interface for end-to-end combinatorial optimization (CO), but textual serialization alone may obscure spatial and relational structures that are important for generating effective CO solutions. This paper presents a general-purpose vision-language solver that augments textual instance descriptions with input-derived visual representations. A single vision-language model (VLM) is applied across different CO tasks and trained using supervised fine-tuning followed by verifier-guided reinforcement learning. While the visual inputs contain no gold solutions or solution-derived information, our experiments show that the VLM generally improves solution quality over its text-only counterpart, with particularly clear gains on more complex CO problems such as CVRP and JSSP. The advantage of visual information is more pronounced at large problem scales. --- ### SimpleEvol: An Agent-Loop Framework for LLM-Driven Automated Heuristic Design with Minimal Human Priors Authors: Jianghan Zhu, Cong Zhang, Rongjie Zhu, Chi Zhang, Zhiguang Cao Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37172v1 Large language models (LLMs) have emerged as powerful tools for automated heuristic design (AHD), enabling iterative generation and refinement of heuristics. However, the dominant paradigm embeds LLMs as narrow, fixed components, such as crossover or mutation, within heavily hand-engineered evolutionary frameworks. We argue this misapprehends LLMs. It treats them as specialized tools rather than general reasoners, constrains them to low-level operations, and underutilizes their autonomy. Moreover, the extensive human priors in these frameworks violate the bitter lesson principle that general methods scaling with computation surpass hand-crafted solutions. This raises a key question: which AHD framework designs best convert stronger LLM capabilities into better heuristics? To address this, we propose metrics for LLM-driven AHD framework handcraftedness (AHI) and intelligence conversion efficiency (ICE). Evaluating ten LLMs across three challenging combinatorial optimization problems, we obtain a notable finding that frameworks with fewer human priors consistently yield higher ICE. Based on this finding, we propose SimpleEvol, an agent-loop framework for AHD which removes nearly all human priors and allows the LLM to operate autonomously. SimpleEvol consistently achieves the highest ICE, often by a large margin. Our results challenge the trend toward complex AHD pipelines and point to a lighter and more model-centric alternative, suggesting that reducing human priors is a more effective strategy to scale up with model intelligence. The source code is available at https://github.com/HenryZhu1029/SimpleEvol-Master. --- ### Bridging Semantic Gaps in RAG through Generated Context Knowledge Fusion Authors: Xinkai Du, Chao Lv, Yalin Sun, Quanjie Han, Lei Yao, Maosong Sun Published: 2026-09-29 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.37171v1 Retrieval-Augmented Generation has established itself as a fundamental framework in natural language processing, seamlessly integrating information retrieval with the generative capabilities of large language models. However, this process is fundamentally constrained by a critical challenge: semantic space mismatch between queries and retrieved contexts. We propose Knowledge-Aware Semantic Bridging (KASB), a novel framework that improves passage selection quality through semantic space alignment between queries and retrieved documents through intelligent knowledge fusion. Our approach leverages the complementary strengths of generative and retrieval-based knowledge through a multistage process that enhances both relevance and accuracy. We evaluate KASB on three popular open-domain Question Answering datasets to demonstrate the effectiveness of our approach. --- ### Interpolated Policy Distillation: A Controllable Continuum Between Off-Policy and On-Policy Distillation Authors: Youxu Shi, Yifan Sun, Dacheng Yin, Haomiao Tang, Guangting Wang, Fengyun Rao, Jing Lyu, Dong Liu Published: 2026-09-29 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.37170v1 Off-policy and on-policy distillation have traditionally been formulated as separate paradigms, each favoring a different property of distillation trajectories. Teacher-generated (off-policy) traces are typically high-quality but lie far from the student's distribution, whereas student-generated (on-policy) rollouts are more learnable but often contain erroneous reasoning. We view these paradigms as the endpoints of a policy continuum and posit that a more effective rollout policy may lie in between. We introduce \textbf{Interpolated Policy Distillation (IPD)}, which defines the next-token distribution at every decoding step as an explicit linear interpolation between the student and teacher distributions. The interpolation operates at the distribution level, token by token, and its coefficient provides direct control over the balance between trajectory quality and student learnability. Naively sampling from this policy would require sequentially querying the teacher at every token and is thus expensive. To make IPD practical, we accelerate it with a new speculative-decoding rule while exactly preserving the interpolated next-token distribution.At the trajectory level, the resulting rollouts naturally interleave student- and teacher-generated segments. Unlike recent heuristic segment-interleaving methods, however, this interleaving is induced by an exactly realized token-level interpolated policy rather than by hand-designed switching rules. Across text-only and multimodal reasoning benchmarks, IPD consistently outperforms both endpoint policies (SFT and OPD), their conventional two-stage combination (SFT-then-OPD), and recent heuristic segment-interleaving methods, demonstrating that token-level policy interpolation better balances trajectory quality and student learnability. --- ### Trajectory Soup: Pushing the Compute-Scaling Frontier of LLM Mid-training via Diverse Trajectories Authors: Zhehao Huang, Changxin Tian, Qingyuan Yang, Kunlong Chen, Ziqi Liu, Zhiqiang Zhang, Xiaolin Huang, Jun Zhou Published: 2026-09-29 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.37169v1 Mid-training equips pretrained large language models with specialized and reasoning capabilities, but the returns of this stage are bounded since additional serial compute yields little further downstream improvement and can even degrade some capabilities, which places a practical ceiling on how much compute mid-training absorbs. We revisit how this compute should be allocated to a single run or multiple similar optimizations. We find that branches forked from a shared checkpoint under various controlled recipe reaches measurably different regions of parameter space, and establish a form of compatible diversity that extending one run cannot supply. Therefore, we introduce Trajectory Soup, which distributes a mid-training budget over several independent branches, and consolidates strongest checkpoints selected on validation through intra- and inter-trajectory averaging into a single model. A local bias and variance analysis separates the two averaging levels, showing that inter-trajectory averaging removes residual error beyond the reach of averaging within a trajectory, while checkpoint selection carries a bias that bounds how many checkpoints are worth merging. Across model scales, learning-rate schedules, token budgets, and trajectory counts, Trajectory Soup improves aggregate downstream performance over the strongest single-trajectory average under matched budgets and keeps improving as budgets expand, with the advantage preserved after an identical post-training pipeline. These results position trajectory allocation and merging as a practical way to extend the compute-scaling frontier of mid-training beyond serial saturation. --- ### Disentangling Spurious Correlations in Vision-Language-Action Models via Predicting Domain-Invariant Latent Lookahead Authors: Junghyun Kim, Ngseo Kim, ChungWoo Lee, Seoyeon Lee, Woo-Jeong Baek, Adam Zhou, Chip Huyen, Jun-Ki Lee, Gi-Cheon Kang, Byoung-Tak Zhang Published: 2026-09-29 Categories: cs.RO, cs.LG arXiv: https://arxiv.org/abs/2609.37165v1 Vision-Language-Action (VLA) models remain brittle under visual distribution shifts, often relying on spurious correlations tied to domain-specific factors rather than task-relevant structure. We propose Domain-Invariant Latent Lookahead (DILL), a representation-learning framework that mitigates shortcut learning in VLA policies. Our key idea is to supervise policies with domain-invariant future latents learned from domain-transformed trajectory data. A Task-Domain Encoder is trained with contrastive objectives and Gaussian disentanglement regularization to separate task-relevant structure from domain-specific visual variation. The learned encoder then provides future latents for VLA policy learning through lookahead prediction and domain disentanglement, encouraging the policy to focus on task-relevant structure rather than incidental visual factors. Counterfactual task-view evaluations show that DILL reduces shortcut reliance, while LIBERO-Plus evaluations demonstrate improved visual robustness, with 69.1% average success, 11.4 percentage points above the strongest baseline. Real-world manipulation experiments further support DILL's applicability beyond controlled simulation. Complementary latent-space diagnostics show that these behavioral gains are accompanied by representations that better preserve task-consistent structure while suppressing domain-specific variation. Our project page is available at https://dill-vla.github.io/. --- ### End-to-End Self-Supervised RGB-T Tracking without Modality Misleading Authors: Shenglan Li, Rui Yao, Kunyang Sun, Hong Jia, Yong Zhou, Javen Qinfeng Shi, Xinyu Zhang Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37162v1 RGB-T object tracking leverages the complementary characteristics of visible and thermal infrared modalities to improve robustness under adverse conditions. Existing supervised methods typically rely on costly modality-aligned bounding box annotations, while most self-supervised approaches follow a two-stage pseudo-labeling paradigm, making tracker training sensitive to pseudo-label quality and preventing joint end-to-end optimization. In this paper, we propose ESMTrack, a fully end-to-end self-supervised RGB-T tracking framework without offline pseudo-label generation or dense frame-level bounding box annotations. Given only the standard initial-frame annotation used in visual tracking, ESMTrack learns discriminative and temporally consistent representations through two complementary objectives: a grounding triplet loss on annotated initial frames and a cross-frame temporal triplet loss on unlabeled search frames, with reliable samples selected by forward-backward consistency. To address modality dominance bias, ESMTrack employs a three-branch architecture consisting of a fusion branch and two unimodal branches for RGB and thermal inputs. We quantify modality contributions using the Average Peak-to-Correlation Energy by measuring response discrepancies between the fusion and unimodal branches. The resulting reliability estimates guide a training-time modality decoupling mechanism that suppresses dominant-modality shortcuts and adaptively weights cross-modal contrastive learning for task-level alignment. Extensive experiments on five RGB-T tracking benchmarks show that ESMTrack achieves competitive state-of-the-art performance, strong cross-dataset generalization, and real-time inference speed. The source code is available at https://github.com/LiShenglana/ESMTrack. --- ### The Vote Hides the Failure: Aggregation Choice and Noise Robustness in Heart Murmur Detection Authors: Nicholaus Dismas Ladislaus, Olatunji Damilare Emmanuel, Samuel Chol Buol Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37161v1 Noise robustness in automated phonocardiogram (PCG) murmur detection, and how it is measured, remains underexamined despite growing interest in low-resource screening. We evaluate two independently reimplemented pipelines, Hierarchical Multi-Scale Convolutional Network (HMS-Net)--CNN, and Bidirectional Long Short-Term Memory (BiLSTM)--LSTM, under controlled, multi-severity noise with noise-augmented fine-tuning and held-out generalization testing. Under matched aggregation, the complete BiLSTM pipeline outperforms the complete HMS-Net pipeline across all conditions in accuracy and Weighted Accuracy. A stable aggregate accuracy score can misrepresent what individual predictions show: HMS-Net's native aggregation degrades under salt-and-pepper noise far less than majority-vote (MV) aggregation at the same severity, a gap reflecting window-level disagreement its native rule absorbs, while BiLSTM's MV accuracy rises after noise-augmented training even though its individual predictions do not improve. HMS-Net's training effect is significant under one accuracy metric but not another. Noise-robustness conclusions can depend as much on evaluation choices as on the models themselves. --- ### GLASS: Global Latent Aggregation with Slot-based Set Decoding for Scalable All-Atom Crystal Generation Authors: Hendrik Kraß, Seyed Mohamad Moosavi, Mathias Niepert Published: 2026-09-29 Categories: cs.LG, cond-mat.mtrl-sci arXiv: https://arxiv.org/abs/2609.37158v1 Generative models for crystals enable the discovery of novel structures, but scaling all-atom generation to larger systems such as metal--organic frameworks remains challenging. We connect this difficulty to the correspondence problem of particle-space generation. Even on a single fixed target set, index-free permutation-equivariant particle flows require substantially more training for reliable generation as set size and density increase, under both independent and optimal-transport couplings. To resolve this challenge, we introduce GLASS---Global Latent Aggregation with Slot-based Set Decoding, which encodes structures in a permutation-invariant global latent space and learns their distribution via flow matching. A learned-slot decoder constructs all atoms in parallel, removing atom-wise correspondence from generative transport. On MP20, GLASS is competitive with particle-space models, and flow training can reach the validity of the training data at every structure size. On a QMOF subset, GLASS generates MOFs with up to 150 atoms per unit cell without conditioning on building blocks, topology, or composition, and approaches the structural validity of the training data. On both datasets, flow training exposes a validity--novelty tradeoff, and MOF novelty remains limited by autoencoder generalization on the available data. These results show that separating correspondence assignment from generative transport provides a simple route toward high-validity generation of larger atomistic systems. --- ### From Learner Behavior to Reusable Skills for Effective and Efficient Learner Simulation Authors: Zijian Chen, Zheng Zhang, Miao Jia, Xingchen Hu, Weibo Gao, Linan Yue Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37157v1 Learner simulation aims to reproduce how a particular learner behaves on new tasks. Although Large Language Models (LLMs) can generate increasingly fine-grained learning behaviors, existing approaches often need to repeatedly process a growing interaction history to reconstruct the learner. This introduces additional context and inference costs and makes the acquired learner-specific simulation capability difficult to reuse across different LLMs. We therefore propose Learner2Skill, which externalizes the simulation capability acquired from historical interactions into a persistent and reusable Simulation Skill. The Skill captures the learner's current learning state and recurring response patterns, evolves as new real interactions arrive, and can be adapted to a new LLM through lightweight executor calibration without reconstructing the learner from scratch. Experiments show that Learner2Skill more faithfully reproduces fine-grained learner behavior while reducing overall token cost, and that the same constructed Skills can be effectively reused across different LLM executors. --- ### Lucid Dreaming for World Models: Learning to Doubt Imagination and Decide by Trust Authors: Ziqi Wen, Ting Xu, Lianyu Wang, Xian Lin, Yanda Meng, Huazhu Fu, Meng Wang, Ching-Yu Cheng Published: 2026-09-29 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.37156v1 World models enable agents to learn and plan in imagination, but predictions beyond their experience can become unreliable and mislead decisions. Existing uncertainty estimates derived from predictions can remain overconfident on unfamiliar state-action pairs. We propose the Lucid World Model (LucidWM), which learns doubt from experience and propagates trust through imagination. By integrating Subjective Logic into categorical latent transitions, LucidWM distinguishes predicted outcomes from their evidential support and assigns each transition a degree of doubt. The complement of this doubt defines transition-level trust, which accumulates multiplicatively along imagined trajectories to reweight returns for policy learning and guide action selection. Uncertainty estimation requires no additional parameters or forward passes. Evaluated on four base world models against seventeen uncertainty readouts, LucidWM detects environmental changes and signals uncertainty during action-corrupted rollouts. In a controlled navigation case study, acting on trust reduces the number of steps required to reach the goal from 362 to 190. Fifteen demonstration videos show how LucidWM doubts its dreams and acts on that doubt. Videos are available at https://lucidwm.github.io. --- ### When Tools Silently Lie: Evaluating and Mitigating Blind Compliance in Tool-Augmented Data Agents Authors: Zifu Tao, Changqing Yin Published: 2026-09-29 Categories: cs.AI, cs.CR arXiv: https://arxiv.org/abs/2609.37153v1 Tool-augmented data agents rely on tool outputs for analytical decisions. Yet successful execution can return plausible but incorrect evidence, requiring agents to decide whether to trust or verify it. Understanding this failure requires examining both the evidence obtained through checking and the answer ultimately adopted. We introduce ToxicBench to measure checking and adoption under numerical, label, schema, and retrieval errors, pairing clean and poisoned observations over fixed source data. In the 118-task GPT evaluation across three adapters, poisoning lowers task success by 26 to 39 percentage points. Ordinary retries help under one-shot poisoning, whereas repeated poisoning reveals wrong-answer adoption after checking. Controls on three public tables isolate how supplied evidence affects recovery. After freezing the scorer, we compare its judgments with human annotations on 200 trajectories, finding 96% task-success agreement. Human judgments support retry gains over Base and confirm adoption after checking on audited tasks. We release trajectories, versioned scoring, and reference and delivery audits. These findings highlight evidence availability and answer selection as complementary dimensions of agent reliability. --- ### Multimodal Detection of Higher-Order Behavioral Constructs: Self-Compassion in Structured Reflective Interaction Authors: Siddhant Jain, Dimitra Tsovaltzi Published: 2026-09-29 Categories: cs.LG, cs.CL, cs.CV, cs.HC arXiv: https://arxiv.org/abs/2609.37148v1 Many of the qualities that matter most in how people learn and grow, how someone regulates their emotions, reflects on a setback, or stays aware of others during a difficult conversation, are not directly observable. They have to be inferred from how someone speaks, moves, and sounds over time, and they resist the kind of clean labeling that most machine learning pipelines are built around. We study this challenge through a case that is well grounded in psychological theory but rarely modeled computationally: self-compassion, the tendency to respond to one's own setbacks with patience rather than harsh self-criticism. We examine how it appears during structured reflective interviews in a technology-mediated training setting, where people naturally talk through socio-emotionally demanding situations. Since no existing dataset captures this kind of construct in this kind of setting, we collected and annotated 51 reflective dialog sessions using an independent, temporally overlapping annotation scheme grounded in established theory. We consolidate the underlying six-component psychological model into a three-class supervision space, balancing self-kindness and mindfulness against self-critical or overwhelmed states, and build a reproducible window-based pipeline that aligns video, audio, and text on a shared timeline. Unimodal models trained on each modality separately are compared against a simple probability-level fusion strategy, which yields modest but consistent gains over the best single modality. We close by discussing where each modality succeeds or struggles, what this suggests about how this kind of construct is actually expressed in reflective speech, and what would be needed to model it, and constructs like it, more effectively. --- ### Improved Distributional Diffusion Models Authors: Tommaso Martorella, Alexandre Galashov, Felix Krause, Stefan Andreas Baumann, Valentin De Bortoli, Arthur Gretton, Björn Ommer Published: 2026-09-29 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2609.37147v1 Distributional Diffusion Models (DDMs) replace the standard mean-prediction denoiser with a \emph{distributional} denoiser trained via a scoring rule objective, learning a stochastic approximation to $p(x_1 \mid x_t)$ rather than its conditional mean. However, scaling DDMs to modern image-generation settings faces two obstacles: (i) multi-particle training incurs overhead that scales with the number of particles, (ii) DDMs use globally fixed scoring rule hyperparameters, forcing a single trade-off across sampling budgets. We mitigate these limitations by deferring particle expansion to late transformer layers, and the hyperparameter trade-off by introducing time-dependent scoring rule schedules informed by the dynamical regimes of~\citet{Biroli2024}. Combined with a DiT-based latent setup, these changes make DDM training practical on class-conditional ImageNet-$256^2$, achieving 4.48 FID at 4 steps and 2.38 at 50 steps with DiT-XL/2, from a single model trained from scratch in one stage, without a teacher, self-distillation or JVPs. The result is a stochastic few-step generator whose FID does not degrade as the sampling budget grows from 4 to 50 NFE, and the same recipe transfers to text-to-image generation. Code and pre-trained models available at https://github.com/CompVis/iDDM. --- ### From Judgment Quality to Downstream Utility: Rethinking LLM-as-a-Judge for Open-Ended Tasks Authors: Zheng Zhang, Lufei Li, Xinyue Tan, Yuanhao Zeng, Ziwei Shan, Yexin Li, Kan Ren Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37145v1 LLM-as-a-Judge is increasingly used to evaluate policy responses on open-ended tasks that lack ground-truth answers. Existing work often directly converts the resulting judgments into reward signals for policy training, paying limited attention to intrinsic judgment quality and largely restricting the use of Judges to training-time supervision. We systematically investigate judgment quality and downstream utility by examining both how judgments are elicited and how they are used. For judgment elicitation, we vary the Judge protocol along three dimensions: verdict granularity, critique usage, and evaluation batching. For judgment usage, beyond policy training, we extend Judge to test-time inference through Best-of-N selection, Judge-guided revision, and beam search. We find that, (i) Surprisingly, judgment quality and downstream utility do not always align. (ii) Judge protocol design substantially affects both intrinsic judgment quality and downstream utility. (iii) Judge guidance effectively converts test-time compute into performance gains, with benefits varying across inference strategies. Our results call for a multifaceted evaluation of LLM Judges on open-ended tasks, encompassing intrinsic judgment quality, and downstream utility. --- ### LoLBench: Evaluating Coding Agents with Long-Horizon Proposals on Large Software Systems Authors: Yun Peng, Zihan Wu, Zeyang Zhuang, Xin Zhou, Rui Shu, Xu Han, Chun Yong Chong, Yuan Wang, Jiakun Liu Published: 2026-09-29 Categories: cs.SE, cs.CL arXiv: https://arxiv.org/abs/2609.37143v1 Modern coding agents can deliver increasingly large repository-level changes, and recent benchmarks reflect this by emphasizing long-horizon tasks with large reference implementations. Many benchmarks evaluate coding agents' implementation capability to produce correct code edits from detailed specifications. However, practical modular development tasks also require the perception capability of grounding user intent and high-level design to derive a specification. We introduce LoLBench to evaluate both capabilities through the entire proposal-to-implementation process on large software systems. It is a multilingual benchmark of 100 tasks across 29 software systems in five domains. Each task provides a human-written enhancement proposal with user intent and high-level design. On average, proposals contain about 5,000 words, software systems contain 2.4 million source lines of code (LoC), and implementation pull requests (PRs) change approximately 5,500 LoC. Across 28 agents we evaluated, the best agent resolves only 14% of tasks and achieves a 52.7% Fail-to-Pass (F2P) pass rate. Failure analysis identifies incomplete code localization as a major bottleneck, while providing reference-derived file trees alongside API specifications improves resolved rates by 16--22 percentage points (2.4--17$\times$), reaching at most 34%. These results show that both perception and implementation remain central challenges for coding agents in practical modular development on large software systems. LoLBench is available at https://huggingface.co/datasets/lolbench26/LoLBench. --- ### MSTypography: Multi-character Semantic Typography via Balancing Word Legibility and Object Recognizability Authors: Xinye Yang, Xinding Zhu, Kai Fang, Xinyi Ren, Mengjian Li, Bin Cao, Jiazhou Chen Published: 2026-09-29 Categories: cs.CV, cs.GR arXiv: https://arxiv.org/abs/2609.37141v1 Semantic typography is a design technique where the visual representation of a word conveys its semantic meaning, while maintaining its legibility. Existing digital typography methods mainly focus on single-character scenarios. They suffer from a lack of legibility constraints and insufficient local deformation when extended to multi-character words, as the intricate structures among multiple characters are hardly preserved during the typography process. In this paper, we propose a global-to-local typography framework for multi-character scenarios. It performs mask-driven silhouette approximation at the global level, while semantic-guided refinement at the local level, with a culling step in between to improve efficiency. To preserve word legibility, we designed structural losses (including explicit collision constraints and implicit Jacobian singular value constraints) and an OCR constraint for character-level readability. To enhance the object recognizability, we leverage semantic guidance with diffusion priors, which drives the character glyph toward the target concept while preserving its structural integrity. To the best of our knowledge, this is the first multi-character semantic typography method that effectively balances word legibility and object recognizability. Evaluations on five representative languages (English, Chinese, Japanese, Korean, Arabic) demonstrate superiority over SOTA methods. Codes will be open-sourced. --- ### TaoFlowForge: Progressive Native Mesh Generation via Cascaded Flow Matching Authors: Xianze Fang, Qiyuan Feng, Dongfang Sun, Yan Zhang, Xiuchao Wu, Jingnan Gao, Jiangjing Lyu, Chengfei Lyu, Gang Yu Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37139v1 3D content generation technology has significantly advanced the work of designers, as well as the 3D printing and gaming industries. However, it remains difficult to produce lightweight, editable, and topologically clean artistic content that is directly production-ready. To achieve this, we present TaoFlowForge, an artistic mesh foundation model that generates production-ready meshes. Specifically, TaoFlowForge decomposes the mesh generation process into vertices generation and their connectivity prediction, i.e., edges. We formulate vertices generation as a two-stage coarse-to-fine process and incorporate several effective loss functions to further enhance its performance. In the connectivity prediction stage, we propose a simple yet effective method for estimating the connectivity affinity between vertices and additionally predict per-vertex normals, which determines the correct orientation of faces. Besides, we construct a large-scale dataset combining hand-crafted 3D assets with public high-quality topology datasets. Based on this, a carefully designed data curation pipeline is employed to filter the raw dataset, retaining only high-quality topology data for model training. Our model is trained on the combined dataset and tested on both out-of-distribution hand-crafted set of 3D assets and public datasets. Under image-conditioned generation, TaoFlowForge outperforms autoregressive methods and achieves state-of-the-art results among open-source mesh topology generators. We will release all the code and weights together with a portion of our test dataset. --- ### Multi-Granularity Language-Guided Imitation Learning via Instruction Decomposition Authors: Yi-Pei Chiu, Wei-Ta Chu Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37135v1 Using language instructions as conditions to guide robot policy learning has recently become an important research domain. However, existing language-guided policy learning methods typically use an overall task description to guide the entire demonstration trajectory. For manipulation tasks involving multiple execution stages, these methods assign the same language description to different subtasks, making it difficult to distinguish the behaviors required at different stages. In this work, we propose a multi-granularity language guidance method based on instruction decomposition. The proposed method decomposes an overall task description into more fine-grained, concrete subtask-level language instructions, thereby enhancing learning efficiency and improving performance. We evaluate the proposed method in the setting of multi-task imitation learning and validate its effectiveness. --- ### SQUARE: Structured Quantum Representation Adapters as Compact Quadratic Feature Maps for Frozen Language Models Authors: Emily Jimin Roh, Hyojun Ahn, Hoyeong Lee, Soohyun Park, Sung Whan Yoon, Vaneet Aggarwal, Joongheon Kim Published: 2026-09-29 Categories: quant-ph, cs.AI arXiv: https://arxiv.org/abs/2609.37134v1 Frozen language models (LMs) are increasingly used as fixed feature extractors for downstream reranking, scoring, and preference modeling, raising a practical question: how should a compact module represent interactions among features in a fixed low-dimensional bottleneck? Common linear and low-rank adapters remain linear at the adaptation module itself, whereas explicit second-order alternatives introduce pairwise interactions through direct parameterization or predefined factorizations. We propose SQUARE, a Structured QUAntum REpresentation adapter that amplitude-encodes the bottleneck vector, applies a parameterized quantum circuit, and measures the resulting state. We show that each basis-probability feature is exactly a normalized quadratic form in the bottleneck coordinates, while the additional Pauli-$Z$ readouts are signed linear combinations of these probabilities. The measured map can therefore parameterize interactions over $O(d^2)$ coordinate pairs through a small set of shared circuit parameters, where $d$ is the bottleneck dimension. It provides a structured parameterization within, rather than beyond, the classical normalized-quadratic feature class. In a disjoint same-pipeline evaluation over eight GLUE-derived controlled interaction tasks and five shared seeds, SQUARE achieves an average test accuracy of $0.7565$, compared with $0.7355$ for an affine normalized-quadratic predictor, $0.7271$ for the evaluated parameter-matched Givens mixing model, $0.6817$ for an MLP, and $0.6155$ for a frozen-circuit control. Under reduced supervision, it also shows consistent gains over the strongest evaluated classical comparator, with the same qualitative pattern across multiple frozen LM backbones. All circuit experiments use simulation, while the learned feature map can be evaluated exactly in batched PyTorch without quantum hardware. --- ### Train Ahead, Distill Back: Bootstrapping On-Policy Self-Distillation for Large Language Models Authors: Zheng Zhang, Xinyue Tan, Lufei Li, Xinyi Zhang, Yexin Li, Kan Ren Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37132v1 On-policy self-distillation (OPSD) improves large language models by letting a self-teacher with privileged information provide dense token-level supervision on the model's own trajectories. Yet existing methods typically construct the self-teacher from the current, initial, or slowly averaged policy state, leaving the quality of supervision constrained by the teacher's ability to exploit privileged information. We ask whether the model's own optimization progress can instead be recycled into a stronger self-teacher. In this paper, we introduce Bootstrapped On-Policy Self-Distillation (B-OPSD), which temporarily trains the policy ahead to obtain a future teacher, restores the student to the original policy state, and then uses the future teacher to supervise the restarted student. The future teacher improves supervision in two complementary ways, it can generate more reliable privileged trajectories and, conditioned on them, provide more informative token-level targets along the restarted student's on-policy trajectories. Experiments on mathematical reasoning with Qwen3-4B and Qwen3-8B show consistent improvements over standard OPSD in both settings, including gains from 27.50 to 41.30 and from 48.80 to 64.44 in the rollout-privileged setting. Our findings point to a broader principle for self-improving models that future learning progress can be distilled backward, preserving acquired knowledge while bootstrapping beyond the optimization state that produced it. --- ### SkillCome: Group Contrast Skill Optimization with Dual Memory Authors: Haolin Li, Feng Hong, Ang Li, Chilin Fu, Weichang Wu, Ya Zhang, Yanfeng Wang, Xiaolu Zhang, Jiangchao Yao Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37128v1 Skill evolution improves the capabilities of large language models by analyzing trajectories generated under a given skill and modifying the skill accordingly. Existing approaches typically generate a single trajectory per question. However, this provides insufficient optimization signals since it requires inferring effective skill edits from a solitary path. It is difficult to pinpoint which actions caused the failure in a failed trajectory, or to determine which actions in a successful one should be incorporated into the skill. Furthermore, they rely on a local batch of trajectories for analysis, making the optimization direction susceptible to noisy evidence. To address these, we propose SkillCome, a Skill-evolution method based on group Contrast optimization with dual memory. For each question, SkillCome generates trajectories and performs group contrast analysis to precisely identify key behavioral divergences between successful and failed trajectories, offering reliable optimization signals. The dual memory system further accumulates evidence from historical steps to track patterns shared across different groups, leading to more generalized optimization directions. Together, SkillCome builds a systematic optimization process that transforms experience from observed successful trajectories into reusable skills. Extensive experiments on six benchmarks spanning question answering, reasoning, and agentic tasks demonstrate the effectiveness of our method. SkillCome consistently outperforms baselines across five models of varying families and scales, with gains up to +5.69 points. --- ### LLM unbranding: Erasing Commercial Identity while Preserving Generic Utility Authors: Kajetan Ożóg, Alicja Wojciechowska, Dawid Malarz, Paweł Batorski, Artur Kasymov, Przemysław Spurek Published: 2026-09-29 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.37127v1 Establishing unbranding as a critical practice to prevent visual logos from acquiring negative connotations is standard in image generation. Large Language Models (LLMs) now face a parallel and emerging challenge. These models frequently generate brand descriptions within diverse contexts. This frequency introduces significant risks, such as trademark dilution, false attribution, and brand defamation. In response, we formally define the novel task of LLM Unbranding. We specifically address the complex challenge of managing trade dress within textual outputs. This involves neutralizing characteristic language, slogans, and stylistic markers that define brand identity. Crucially, these elements are less evident than explicit visual logos. To benchmark this task, we introduce a comprehensive evaluation dataset incorporating prominent brands from multiple commercial domains. We rigorously evaluate existing state-of-the-art machine unlearning models using this benchmark. This evaluation identifies their limitations in selective textual unbranding. Finally, we propose MUTE, a novel inference-time method that effectively neutralizes textual trade dress while preserving the LLM's general capabilities and utility. By leveraging an iterative refinement loop, MUTE systematically optimizes system instructions to safely eliminate brand leakage without requiring fragile parameter updates. Code and dataset: The evaluation dataset and code for LLM Unbranding are available at https://github.com/KajetanOzog/LLM_unbranding. The implementation of MUTE is available at https://github.com/KajetanOzog/MUTE. --- ### When Should Agents Check External State? Budgeting Observations for Stored Intentions Authors: Zhengkun Di, Bin Shi, Kai Sun, Yiming Xu, Bo Dong Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37125v1 Prospective memory allows an agent to retain an intention tied to a future condition, but the stored intention does not reveal whether that condition currently holds. Checking it may require web access, multi-step tool use, and paid calls. Existing systems decide when intentions require attention, but do not allocate the resulting observations under a shared budget. We introduce the first resource-allocation formulation for the external observations required by stored intentions under a shared episode budget. BudgetPM offers two policy variants that share a hard-budget executor. BudgetPM-Static uses a lightweight Logistic scorer to learn whether a check improves the current decision. BudgetPM-Sequential distills full-episode hindsight schedules into a lightweight policy that decides when to spend or reserve capacity using only pre-query information at deployment. We evaluate BudgetPM against two public memory-agent systems, five matched controls, and four hand-designed monitoring or budget-adaptation rules. Across two benchmarks and three backbones, BudgetPM-Static outperforms adapted Mem0 and PMA workflows. On PM-Bench, its Logistic scorer reaches competitive quality--cost operating points alongside higher-capacity scorers and retains 99.9--100\% of unconstrained quality with 42--54\% fewer observations. Under severe scarcity and the same hard caps, BudgetPM-Sequential exceeds the strongest tested natural monitoring schedule by 1.92--2.58 Set F1 points. It reaches the same Set F1 and on-time recall with 16--33\% fewer observations. Matched attribution, exact-cost analysis, and a fixed-budget load intervention link this gain to competition between present and future opportunities. These results yield a demand--capacity design rule: local gating works when capacity covers demand, while future-aware supervision adds value when observations compete across time. --- ### EviViT: Evidence-Adaptive Vision Transformers for Fine-Grained Perception Authors: Yaoxin Niu, Zhangquan Chen, Yang Zhang, Xiang An, Zhumei Wang, Chih-Ting Liao, Hongkun Cao, Ruqi Huang Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37123v1 Fine-grained visual perception enables vision-language models to distinguish subtle attributes and ground their answers in visual evidence. In high-resolution scenes, processing the whole image at greater resolution spends visual tokens on irrelevant content, while isolated crops can lose the context needed to interpret the selected evidence. We introduce EviViT, a lightweight attachment that learns where a pretrained vision transformer should acquire detail. Human visual-search traces supervise a question-conditioned evidence density, which guides regional re-reading from the original pixels and the allocation of visual tokens. A sparse, coordinate-aware bridge then connects the regional features to the global scene, allowing the host to interpret precise evidence in context. Learned with the host backbone frozen, the attachment serves both the base model and compatible post-trained descendants without refitting. Experiments across nine hosts show consistent gains in average fine-grained accuracy. Matched-budget comparisons further show that EviViT outperforms global-only processing at every tested token ceiling while using fewer visual tokens. --- ### Learning the Structure of Triangular Transport Maps Authors: Morten Blørstad, Pekka Parviainen, Berent Ånund Strømnes Lunde Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37122v1 Triangular transport maps provide a flexible approach to sampling-based probabilistic modeling, including density estimation, generative modeling, and Bayesian inference. They transform an unknown target distribution into a simpler reference through a monotone triangular map. The map structure is defined by a variable ordering and sparsity pattern, which together encode a directed acyclic graph. Map quality can depend strongly on this structure, yet finding a good structure is computationally expensive because each candidate generally requires fitting a different map. A central challenge is therefore to learn density and structure jointly, while keeping computation manageable as dimension grows. We introduce Self-Structuring Transport Maps (SSTM), which learn the map, ordering, and sparsity jointly. We use SoftSort to learn the variable ordering and $L_0$ gates to learn the sparsity, while preserving a triangular structure. To keep the map scalable, we use a monotone BatchEnsemble that shares one weight matrix across all map components through rank-one adapters. Across synthetic and real data, jointly learning the structure and map gives better density estimates than estimating the structure first. When the structure is identifiable from the density, SSTM matches the density performance of a map fitted with the true structure and outperforms autoregressive flows. On large datasets, SSTM is competitive with autoregressive flows. --- ### Cross-Linguistic Effects in Bilingual Phoneme BabyLMs Authors: Nikitas Theodoropoulos, Maria Lymperaiou, Giorgos Filandrianos Published: 2026-09-29 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.37121v1 Cross-linguistic effects are a central topic in bilingual first-language acquisition. Artificial learners can help investigate L1-L2 interactions by enabling controlled comparisons across language combinations and learning conditions. Recent work explores this direction by training bilingual language models under developmentally plausible constraints. However, human and model learners still diverge in fundamental ways, with one major difference being input modality: children learn primarily from spoken input, whereas language models are typically trained on orthographic text. To reduce this gap, researchers have trained models on phonemic representations of speech. In this work, we combine these research directions to train bilingual BabyLMs with phonemic input. We keep English fixed as the L2 and vary the L1 across German, Swedish, Persian, and Basque, selected to represent contrasting combinations of syntactic and phoneme-inventory distance from English. Our results show stronger L1-related variation in grammatical learning trajectories under phonemic than orthographic input, while early lexical differences align with phoneme-inventory similarity. --- ### Unlocking the Critic: Reward-Free Policy Optimization for LLM Post-Training Authors: Hongyang Li, Xiao Li, Caesar Wu, Said Mammar, Grégoire Danoy, Pascal Bouvry Published: 2026-09-29 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.37119v1 Recent approaches to reinforcement learning (RL) post-training for large language models increasingly remove the critic to reduce training instability and memory overhead. Even where a critic is trained, it is discarded once training ends, although it has learned to predict outcomes. We revisit this trend and show that a pretrained critic's ability to predict future outcomes can make it a valuable asset for efficient long-horizon reasoning. First, we find that instability in critic-based RL for long chain-of-thought reasoning is largely an optimization artifact: keeping policy updates small and low in variance restores stable convergence. Second, a well-pretrained critic estimates the posterior probability of eventual success from later trajectory states and unfinished prefixes. Its predictions provide outcome-derived, dense, per-prefix learning signals that, during policy optimization, require neither completed rollouts, step-level annotations, nor external reward labels. Building on this insight, we introduce Reward-Free Policy Optimization (RFPO), which repurposes a single calibrated, frozen critic as a rollout-level reward, a value baseline for generalized advantage estimation, and a success forecaster for unfinished prefixes. We further show that binarizing the debiased score stops the policy from exploiting the critic's length bias. Binarized, RFPO matches supervised PPO without a single label in the training loop, while cutting compute and memory overhead. This makes RFPO well suited to long-horizon reasoning tasks, where outcomes arrive late and generation dominates cost: because rollouts can be rewarded before they finish, training no longer has to pay for waiting on every trajectory to complete. Our findings challenge the prevailing critic-free paradigm and establish critic-based, reward-free optimization as a scalable and computationally efficient path for LLM post-training. --- ### Multichannel Audio Quality Assessment: Extending Pretrained Perceptual Models to Spatial Audio Authors: Gouthaman KV, Shiv Gehlot, Vishnu Raj, Lars Villemoes, Arijit Biswas Published: 2026-09-29 Categories: cs.SD, cs.AI arXiv: https://arxiv.org/abs/2609.37116v1 Accurate perceptual quality assessment is essential for evaluating and optimizing spatial audio, where perceived quality depends on both signal fidelity and inter-channel spatial relationships. However, subjective evaluation is costly, while existing perceptual models are often trained for limited channel configurations and cannot be directly applied to higher-channel-count audio. This raises the question: how can pretrained perceptual knowledge be effectively reused for multichannel spatial audio? Using 5.1-channel audio, we study four levels of multichannel integration: signal, prediction, latent, and feature and propose two learned approaches: latent-level aggregation of spatial-group representations and the feature-level Feature-Band Group Attention (FGAtt), which adaptively fuses spatial groups at the feature level before perceptual processing. Across five 5.1-channel test sets, FGAtt achieves the strongest over- all performance, demonstrating the effectiveness of feature-level adaptation for reusing pretrained perceptual knowledge --- ### NRF-GS: Neural Residual Fields for Expressive and Compact Gaussian Splatting Authors: Pratik Singh Bisht, Andreas Kolb Published: 2026-09-29 Categories: cs.CV, cs.GR arXiv: https://arxiv.org/abs/2609.37115v1 We revisit the role of appearance modeling in 3D Gaussian Splatting (3DGS) and show that limited expressiveness in view-dependent reflectance is a key driver of representation redundancy. In standard 3DGS, low-order spherical harmonics (SH) are used, restricting the splats' ability to model high-frequency directional effects, which is typically compensated by increasing the number of splats. We propose \emph{NRF-GS: Neural Residual Fields for Gaussian Splatting}, a hybrid representation that replaces per-splat SH-bases with a shared neural residual field. Each Gaussian encodes a compact set of appearance features and a lambertian base color, while a lightweight \emph{global scene-level MLP} predicts view-dependent residuals conditioned on viewing direction, distance, and per-splat features. This formulation enhances directional reflectance modeling by combining diffuse per-splat reflectance representations with a shared global function for high-frequency details, enabling both higher expressiveness and parameter sharing across splats. Our key insight is that by accurately capturing high-frequency directional reflectance, especially in specular regions, the GS-representation becomes more expressive, reducing the need for geometrically redundant splats. As a result, NRF-GS achieves comparable or better rendering quality while reducing the number of Gaussians by up to 50\%, and produces visibly improved specular and high-frequency details. --- ### Interpretable intrinsic dimension estimation through componentwise calibration of distance and angle Authors: Chih-Hsuan Huang, Chih-Wei Chen, Szu-Chi Chung Published: 2026-09-29 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2609.37114v1 DANCo (Dimensionality from Angle and Norm Concentration) jointly calibrates nearest-neighbor distance and angular statistics and consistently reaches state-of-the-art accuracy on clean intrinsic-dimension (ID) benchmarks. Practical data, however, introduce neighborhood-relative noise and sample-amplitude heterogeneity that can distort these geometric signals. We reformulate DANCo componentwise, retaining separate distance and angular discrepancy curves so that the source of an estimate can be identified and interpreted. For the distance component, we derive a closed-form Kullback-Leibler divergence for the generic-order ratios of the generalized ratios ID estimator (Gride); when both angular parameters are matched (Full), Gride reduces mean percentage error from $27.7\%$ to $17.6\%$ at noise equal to $40\%$ of typical neighbor spacing on 24 manifolds. For the angular component, two sampling regimes motivate aligning mean direction while retaining concentration matching (Profiled). On a Gaussian scale mixture with generating dimension 70 embedded in 100 dimensions, profiling raises the Minimum Neighbor Distance (MiND) estimate from $22.8$ to $66.7$, while removing the known amplitudes restores MiND-Full to $71.9$; the control thus attributes the Full shortfall to amplitude heterogeneity. On CIFAR-10 and ImageNet, amplitude-reducing normalizations move angular location toward the references and narrow the Full-Profiled gap, an observational counterpart to the controlled mixture. Across four pretrained convolutional neural networks, Gride-Profiled, the two-nearest-neighbor estimator (TWO-NN), and the maximum-likelihood estimator (MLE) exhibit similar rise-and-fall profiles, while Full-Profiled differences identify the layers most sensitive to angular calibration. --- ### Neural Constitutive Learning for Generalized Reaction-Diffusion Systems Authors: Shang-Ke Chen, Yu-Peng Wang, Shih-Hsuan Hung, Wei-Fang Sun, Chao-Shun Zhan, Simon See, Min-Jhe Lu Published: 2026-09-29 Categories: cs.LG, math-ph arXiv: https://arxiv.org/abs/2609.37113v1 Generalized reaction-diffusion systems encompass diverse transport mechanisms and coupled reaction kinetics. A central question for neural PDE solvers is what should be learned so that a common interface can accommodate phase-field and degenerate transport, local reactions, and multispecies coupling. We propose the Neural Constitutive Laws--Mass-Compression-Transport (NCL-MCT) Solver, which learns PDE-specific constitutive responses while retaining temporal evolution in a shared MCT integrator. Transport is represented through mobility and thermodynamic driving force, and reaction through relative reaction rates. These constitutive responses depend on the current density rather than explicitly on the initial condition or elapsed time, motivating their reuse across different initial conditions and time horizons. The same interface supports velocity-data supervision and known-law supervision, neither of which requires time integration during training. When constitutive laws are known, supervision can be evaluated on independently sampled density fields, enabling trajectory-free constitutive learning without generating solution trajectories. Across seven systems, separately trained constitutive modules share the same interface and MCT integrator and achieve relative rollout $L^2$ errors of $10^{-4}$ to $10^{-2}$. Tests with unseen initial-condition families and an extended time horizon assess reuse beyond training conditions, while separate experiments demonstrate trajectory-free constitutive learning. These results support constitutive responses as an effective learning target for a shared neural PDE framework. ---