# Research Papers — 2026-09 · bundle page 264 of 266 > Full-text markdown of papers #13151–13200 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/264 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/263.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/265.md) ### Towards Generalizable Visually Grounded Exploration of Household Devices Authors: Linhao Zheng, Zeming Liu, Wangke Chen, Li Zeng, Wanxiang Che, Heyan Huang, Yuhang Guo Published: 2026-09-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.00845v1 Recent advancements in Vision-Language Models (VLMs) have demonstrated impressive capabilities in static visual recognition and high-level semantic reasoning. However, current embodied exploration paradigms still heavily rely on imitation learning from human-annotated trajectories, which severely limits agents' generalization ability. The key bottleneck of realizing general autonomous embodied agents lies in Generalizable Visually Grounded Exploration: the ability to operate novel devices without manuals or specific training by actively grounding abstract world knowledge into fine-grained visual affordances. Yet, existing benchmarks fail to evaluate this capability: they generally rely on explicit documents and annotated trajectories, neglecting the dynamic Hypothesis-Interaction-Refinement process essential for functional device operation. To bridge this gap, we introduce VGEBench, a comprehensive benchmark designed to evaluate the generalizable visually grounded exploration capabilities of VLMs. Unlike static datasets, we construct a Logic-Driven State Machine framework. This framework simulates multi-turn interaction loops, compelling agents to achieve goals by active visual perception and feedback-driven correction. Experimental results demonstrate that existing VLMs face significant challenges in translating semantic knowledge into physical execution and maintaining long-horizon state tracking. --- ### Staged Linguistic Seeding: Grounded Query Expansion for Verified-Unit QA in AI Contact Centers Authors: Hyeonseop Yoon, Jeong-Eun Park Published: 2026-09-01 Categories: cs.CL, cs.IR arXiv: https://arxiv.org/abs/2609.00844v2 Customer-service QA in an AI contact center (AICC) runs under deployment constraints that benchmark QA misses: tight voice-hotline latency and a high cost for unsupported or wrong automatic answers. We deploy a system that answers only from a closed set of verified QA units: it returns a retrieved unit verbatim, or routes to clarify, abstain, or handoff. The index is enriched offline by staged linguistic seeding (SLS): a human authors a per-unit world-grounded slot recipe, gpt-4.1-mini renders it into variants, and a light human gate filters them. One methodology is reused across both domains, so inference stays a single retrieval pass with no query-time generation. On held-out query variants from two industrial domains, SLS lifts hybrid R@1 to 0.881/0.930 (+0.27/+0.34), with gains across all five retrievers tested. At the same gpt-4.1-mini generation budget, SLS beats doc2query by +0.20/+0.32, while cross-provenance evaluation provides additional evidence of transfer across generated-query distributions. Verified-unit answering also removes free-form generation's unsupported-content surface (7-13% versus approximately 0%). We report this as an application study, including negative results. --- ### Residual Kalman Dynamics for Event-Based UAV Forecasting Authors: Per Nyblom, Hannes Ovrén, David Gustafsson Published: 2026-09-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.00839v1 We study short- and mid-horizon UAV bounding-box forecasting on the FRED event-camera dataset. We use a constant-velocity Kalman filter over a full center-size box state as a strong physical baseline, and train a residual model to predict acceleration-like corrections from recent box history, filtered state features, and local event representations. This simple residual formulation consistently improves over the Kalman baseline, with event-conditioned models giving the strongest results among the evaluated methods. We further show that part of the residual target is predictable from anchor position and velocity alone, indicating that canonical FRED results can reflect both visual evidence and dataset-specific motion priors. To analyze this effect, we introduce decorrelated subsets as a diagnostic stress test, showing that event-conditioned residual models retain useful predictive signal even when measured position- and velocity-based shortcuts are weakened. --- ### Probabilistic Model Checking of Autoregressive Neural Sequence Models Authors: Helge Spieker, Dennis Gross, Arnaud Gotlieb Published: 2026-09-01 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2609.00838v1 Test-set accuracy is silent on two issues that matter when deploying autoregressive neural sequence models: how much probability mass the system under test (SUT) places on constraint-violating alternatives that are reachable under sampling and what fraction of the input population satisfies a domain requirement. We answer both with probabilistic model checking. The pipeline extracts a discrete-time Markov chain (DTMC) from the SUT's token-by-token generation, verifies formal PCTL specifications with the PRISM model checker, and aggregates the per-input verdicts into a coverage curve over the input space. A soundness theorem establishes the DTMC as an under-approximation, so every verdict yields a certified interval on the SUT's true reachability probability. The coverage built from those verdicts is, therefore, conservative by construction. A counterexample-guided abstraction refinement (CEGAR) loop adaptively tightens the interval, and a maximum-likelihood algorithm extracts the most probable falsifying trace. Two case studies exercise the pipeline. On a GPT-2 computer-aided process-planning (CAPP) model with 100% test accuracy, the pipeline quantifies the probability mass greedy decoding hides, but that is reachable with sampling; and identifies the smallest training fraction at which an ordering requirement holds population-wide, neither of which test accuracy can report. We then verify the SMILES molecular generator with a 50x larger vocabulary. The only change is an external chemical-validity oracle, and the pipeline identifies the gap between structural completeness and chemical validity. --- ### Replacing Training with Memory: Listwise Selection for Text-to-SQL Authors: Yeonseok Jeong, Soyoung Yoon, Seongjun Lee, Seung-won Hwang Published: 2026-09-01 Categories: cs.SE, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.00834v1 Modern Text-to-SQL systems often follow generate-execute-select pipelines, generating multiple candidate queries then selecting the best one. Listwise selection, by jointly comparing multiple candidates, has been widely adopted, but fine-tuning listwise selectors is costly. We thus propose a fine-tuning-free listwise selector. We replace two major fine-tuning objectives with inference-time strategies: (1) learning selection criteria as ordering and (2) mitigating positional bias. First, we build reusable structured memories instead of learning selection behavior as model parameters. Given a question, MaP-SQL retrieves memories distilled from training data that encode how natural language maps to schema elements, SQL operations, and expected outputs. These memories serve as explicit decision criteria for evaluating candidates in a listwise manner. Second, to mitigate ordering bias of listwise selectors, we aggregate rankings across multiple input permutations, with inference cost optimized by execution results and pointwise scoring. Our approach improves selection accuracy while maintaining efficiency and compatibility with existing large language models. Across Text-to-SQL benchmarks, it produces more stable selection without fine-tuning and fewer unnecessary comparisons than existing methods. On BIRD-dev, it outperforms the previous state-of-the-art selector-based method R^3-SQL by 2.02 execution accuracy points on average using the same candidate sets, with 2.92x fewer tokens. --- ### Dense Process Supervision for Search Agents via Fact Utility Estimation Authors: Rongzhi Zhu, Xiangyu Liu, Yi Liu, Shuo Zhang, Ruirui Zhang, Rui Wu, Tao Jiang, Zequn Sun, Wenhao Xu, Wei Hu Published: 2026-09-01 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.00833v1 Reinforcement learning (RL) for search agents typically relies on outcome rewards. However, it often fails to achieve effective credit assignment, due to the unclear value of intermediate steps. It is hard to separate their contributions from the final result. In this paper, we propose a dense process supervision method based on fact utility estimation, which models the reasoning process as the accumulation of discrete evidence facts. We first extract structured facts from raw observations and organize them into an explicit fact store. To support credit assignment, we then cluster semantically equivalent facts and infer the posterior utility of each fact cluster using Bayesian estimation over group rollouts. Finally, we convert the estimated fact utilities into dense step-level rewards to guide RL training. Experiments on seven single-hop and multi-hop QA benchmarks show that our method consistently outperforms existing baselines. Ablation studies validate clear relative improvements on multi-hop QA compared to outcome reward-only training. --- ### TWIX: a Two-Stage Approach for End-To-End Named Entity Recognition and Relation Extraction Authors: Marco Martinelli, Laura Menotti Published: 2026-09-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.00832v1 The exponential growth of scientific publications calls for automatic Information Extraction (IE) systems to support knowledge discovery. In this context, the GutBrainIE benchmark evaluates Named Entity Recognition (NER), Named Entity Recognition and Disambiguation (NERD), and Relation Extraction (RE) systems in the gut-brain axis domain. We propose Two-stage Workflow for Information eXtraction (TWIX), an end-to-end IE pipeline featuring three interconnected modules, each leveraging a two-stage framework to solve all four GutBrainIE subtasks. Evaluation on the development and test sets shows that our method substantially outperforms the baseline by a wide margin, while also ranking first among all participant submissions across all subtasks. These results indicate that the proposed two-stage pipeline effectively improves both precision and recall in practical settings. --- ### FLaG: Frequency-Domain Latent-attention Gated Pooling for Token Aggregation Authors: Kewei Li, Rongying Zhang, Xueli Wang, Xiwen Gong, Zhongjian Wang, Qiuchen Zhao, Lan Huang, Ruochi Zhang, Fengfeng Zhou Published: 2026-09-01 Categories: cs.AI, q-bio.BM arXiv: https://arxiv.org/abs/2609.00831v1 Token aggregation converts token-level representations into fixed-dimensional sample representations, but most pooling methods operate only in the original token space. We introduce Frequency-Domain Latent-attention Gated Pooling (FLaG), a plug-in aggregation module that re-expresses encoder outputs in the Fourier domain before final pooling. FLaG represents the nonredundant rFFT spectrum through concatenated real and imaginary components, summarizes spectral tokens with learnable latent queries, derives a sample-conditioned channel gate, and reconstructs modulated token representations for downstream aggregation. We evaluate the same architecture across ESM2-based antimicrobial peptide (AMP) activity prediction, ResNet18 image classification on CIFAR-10 and CIFAR-100, and three RoBERTa-based language tasks. FLaG achieves the best macro-averaged Spearman correlation coefficient, RMSE, and Recall@50 across four AMP backbone-species settings and the highest top-1 accuracy on CIFAR 10. It also achieves the best mean results on five of seven language metrics, although mean pooling remains strongest on STSBenchmark. AMP-side mechanistic analyses reveal low-frequency prediction sensitivity across most encoder layers, with increased relative high-frequency sensitivity in the final layer, and pronounced peptide-specific positional responses. The residual gate broadly amplifies spectral channels while preserving the low-frequency-dominated energy profile, whereas latent cross-attention exhibits sample- and species-specific spectral allocation. Overall, FLaG provides a transferable frequency-domain aggregation bias across protein, visual, and textual representations, with benefits that depend on the backbone and downstream task. Supplementary materials, source code, and data are available at https://www.healthinformaticslab.org/supp/ and https://github.com/Kewei2023/AMPCliff/tree/FLaG. --- ### Visual Attention Faithfulness in Vision-Language Models is Heterogeneous Authors: Xurui Song, Weishi Wang, Zhongqi Yue, Kuluhan Binici, Tao Bai, Hongxin Shao, Daniel Dahlmeier, Jun Luo Published: 2026-09-01 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.00830v1 Whether attention weights faithfully reflect model reasoning has been actively debated in NLP, yet this question remains largely unexplored for the visual modality in Vision-Language Models (VLMs). We address this gap through causal perturbation analysis on current VLMs, evaluating both the comprehensiveness and sufficiency gap of attention-ranked visual tokens. Our analysis reveals that visual attention faithfulness is heterogeneous, manifesting in three distinct processing modes: Faithful-Sufficient, where top-$k$ attention tokens are both necessary and sufficient for prediction; Faithful-Distributed, where they are necessary but broader visual context remains required; and Non-Focal, where no localized attention region is individually necessary while visual information remains an essential trigger for prediction. Furthermore, human-annotated ground-truth regions satisfy comprehensiveness in only $\sim 60$% of cases compared with model attention rankings, revealing systematic divergence between model visual reliance and human intuition. We demonstrate these patterns across both general VQA on VQAv2 and document tasks on VRDU and ChartQA, showing that visual attention faithfulness varies systematically with processing demands and model architectures rather than being uniformly faithful or unfaithful. --- ### HarnessEvolve: Learning from Reference Trajectories for Reliable Agent Self-Evolution Authors: Wen Jiang, Mingmin Chu, Yimeng Tian, Qianxin Zhang, Haofei Yang, Rui Yang, Yang Liu, Tao Lv, Fangming Li Published: 2026-09-01 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.00829v1 Self-evolving agents advance toward autonomy by optimizing their harness---prompts, skills, tools, and execution logic---based on environmental feedback. This paradigm, however, is hampered by three challenges: \textit{credit assignment failure}, where terminal success/failure feedback makes it ambiguous which step caused the error; \textit{shortcut learning}, where agents memorize task-specific patterns rather than acquire generalizable capabilities; and \textit{catastrophic forgetting}, where unguarded updates degrade previously acquired competence. In this paper, we introduce HarnessEvolve, a self-evolving framework that learns from reference trajectories to achieve reliable agent self-evolution. HarnessEvolve decouples the execution agent from the evolutionary pipeline, assigning execution, evaluation, optimization, and gating to independent agent modules, enabling generalizable and stable harness improvements. Specifically, HarnessEvolve overcomes credit assignment failure by generating reference trajectories (execution paths produced when given the ground-truth answers) and aligning failed executions against them to extract error signals, which are clustered to reveal systematic failure patterns. To prevent shortcut learning and catastrophic forgetting, candidate harness updates must pass two gates: a quality gate that filters data leakage and prompt bloat, and a performance gate that accepts each update if it improves on the current batch without degrading recent batches, with epoch-end validation on a held-out set selecting the best-performing accepted agent snapshot. We conduct extensive experiments on several benchmarks spanning open-domain and enterprise scenarios, using different models and agent frameworks. Results demonstrate that HarnessEvolve consistently outperforms state-of-the-art baselines across all benchmarks and settings, confirming reliability across task domains. --- ### Polished but Unresolved: Identifying Late-Stage Pressure States in Long-Horizon Tool-Use Agents Authors: Haoyang Chen, Yi Liu, Jianzhi Shao, Xiaozhou Xu, Zhe Sun, Wei Hu Published: 2026-09-01 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.00823v1 Long-horizon tool-use agents need not only to search and plan, but also to decide when to finalize. We study late-stage pressure states, in which an agent is biased toward submitting a final answer that appears complete and polished while key constraints remain unresolved. We first train a linear probe to show that this pressure state is identifiable from the agent's hidden states. Then, we use activation interventions along this pressure direction and find that shifting the hidden states changes both the pressure score and whether the agent continues tool use or submits early. Through controlled context manipulations, we further see that the pressure is mitigated by constraint clarity and action mapping. Based on these findings, we propose Probe-Sensed Pressure Relief (PSPR), a plugin that applies lightweight pressure relief direction under moderate pressure and moves to structured organization under high pressure risk. Experiments on multiple long-horizon benchmarks show that our method consistently strengthens existing agent methods. --- ### AnalysisBank: An Expert Analysis Pattern Library for Financial Report Generation Authors: Yajing Yang, Yunshan Ma, Kelvin J. L. Koa, Min-Yen Kan Published: 2026-09-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.00818v1 We argue that financial report generation should operate at the analytical rather than structural level, composing content from data-derived insights rather than high-level topics or sections. To this end, we propose AnalysisBank, which distills expert reports into a reusable library of Analyses, each pairing a data signal, an analytical move, and the expert span it was derived from. At inference time, AnalysisBank matches input signals to library entries and applies the retrieved moves to compose the report. A study of Analyses distilled from 550 expert reports reveals a heavy-tailed distribution of 47-52 signal types spanning 13 move types. On two financial benchmarks across four LLM backbones, AnalysisBank increases the proportion of novel, data-grounded insights by 1.7-3.7x over structural-level baselines. Transfer to scientific writing suggests that the distinction generalizes beyond finance. Code and the distilled Analysis library are available at https://github.com/yajingyang/AnalysisBank. --- ### Can Scene Text Recognition Read Rare Compositions? Authors: Genpei Zhang Published: 2026-09-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.00816v1 Scene text recognition is reported as 89--97% accurate on the six standard benchmarks, and the problem is widely treated as saturated. We present an alternative reading. When the same test images are stratified jointly by ground-truth word rarity and character n-gram novelty against a reference corpus, accuracy at the rare-word x rare-trigram corner of the resulting 5x5 grid drops 10--18 pt below the q3/q3 centre across nine English specialised recognisers, and the same direction (corner below centre) holds on all 13 of 13 (language, model) pairs we test across four writing systems (Latin, Han, Han+kana, Arabic). The drop is not a capacity bottleneck. A 6x vision-backbone scale-up (CLIP4STR-Base 158M -> CLIP4STR-Huge 1.0B, OpenCLIP ViT-H/14 LAION-2B) leads every benchmark in aggregate accuracy yet leaves the stress corner unchanged (86.9 -> 86.5, within paired-bootstrap noise). Four converging probes--layer-wise probing, confidence-when-wrong, attention re-balancing, and a cross-script commit-vs-abstain error split--localise the failure to the autoregressive decoder's lexical prior. We then ask how much of the gap existing techniques recover. Of 16 non-architectural mitigations, the largest mean q5/q5 gain is +1.3 pt and none clears the paired-bootstrap noise floor; the only intervention that does is the architectural shift from autoregressive to CTC decoding (SVTRv2, +2.5 pt, p=0.02, n=474). A confidence-routed AR-CTC ensemble adds a directionally consistent +0.6 pt that stays within noise, and its dominant learned coefficient is each model's own minimum-softmax confidence--independently echoing the mechanism above. No configuration we test improves both the compositional corner and aggregate accuracy. The rare-input long tail thus points to architectural change rather than added capacity. --- ### RingMoClaw: An Experience-Inspired Multi-Agent Framework for Self-Evolving Research in Remote Sensing Authors: Kaiyue Kang, Qixuan He, Peijin Wang, Yingchao Feng, Chao Ren, Kangxin Wang, Wenhui Diao, Yixiao Wang, Liangjin Zhao, Kaiwen Wei, Nayu Liu, Xian Sun Published: 2026-09-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.00814v2 Remote sensing visual models have continuously advanced various interpretation tasks. However, the research process behind model improvement still heavily relies on manual expertise, requiring extensive trial-and-error iterations in model design, data processing, and performance diagnosis. Existing agent-based approaches mainly focus on task execution and workflow orchestration, while lacking the capability of autonomous research iteration for continuous performance optimization. To address this issue, we propose RingMoClaw, an experience-inspired self-evolving multi-agent framework for remote sensing visual interpretation. RingMoClaw integrates a research branch, a quality-control branch, and a dual-stream dynamic experience bus to establish a closed-loop optimization process covering strategy generation, experiment execution, independent review, and experience accumulation. The heterogeneous Critic mechanism provides stage-wise diagnosis and feedback, while the dual-stream experience bus incorporates external knowledge and internal experimental experience to guide strategy evolution and eliminate ineffective searches. Extensive experiments on four remote sensing downstream tasks, including object detection, scene classification, semantic segmentation, and change detection, demonstrate the effectiveness and generalization of RingMoClaw. Compared with the corresponding baseline models, RingMoClaw improves performance by 1.84\% mAP$_{50}$ on object detection and achieves consistent gains across the other three tasks, while reducing the required evolution steps by over 40\% compared with existing research automation frameworks. These results suggest that RingMoClaw offers a feasible route from task execution toward continuous research driven model evolution in remote sensing. --- ### One Policy, Any Budget: Internalizing Budget-Aware Search via Reinforcement Learning Authors: Xiaowei Sun, Jin Li, Yili Hong, Yikun Fu, Yanghua Xiao Published: 2026-09-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.00813v1 While reinforcement learning has enabled LLM-based search agents to invoke external tools, existing methods train under fixed budgets and cannot adapt when constraints vary at deployment. We propose AnySearch, a framework that enables a single policy to perform budget-aware search under any budget constraint through a training scaffold and curriculum reinforcement learning. In the first phase, we train the agent with explicit budget state injection and structured reasoning prompts that guide efficient allocation under linearly decaying budgets. In the second phase, the scaffold is removed and the agent learns to operate autonomously under adaptively sampled budget constraints, matching inference conditions. Both phases are optimized with a composite reward that couples answer accuracy with budget efficiency through absolute and relative signals, where an adaptive weight amplifies the efficiency signal for high-accuracy queries and attenuates it for low-accuracy ones. Extensive experiments on seven general and multi-hop QA benchmarks show that our method outperforms baselines across all budget scales, generalizes to unseen constraints beyond the training range, and achieves superior tool productivity without excessive token overhead. Our code is available at https://github.com/xwsun01/AnySearch. --- ### ReBridge-Flow: Re-Coupling Posterior Bridges in Flow Matching for Image Restoration Authors: Jiaqi Zhang, Yiqi Wang, Hongjie Wu, Bohan Guo, Xinan Wang, Zichen Luo, Taotao Cai, Zhi Chen, Mingkai Zheng Published: 2026-09-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.00811v1 Flow Matching provides an efficient generative prior for image restoration by learning continuous transport between source and data distributions. However, existing methods typically incorporate measurement constraints through local corrections. Such corrections may disrupt the source-clean endpoint coupling implicitly encoded by the pretrained flow, making the corrected endpoint pair incompatible with the current state. To address this issue, we propose ReBridge-Flow, a posterior bridge re-coupling method. Specifically, given the current state, ReBridge-Flow first decodes the corresponding local source and clean endpoints. It then incorporates measurement information through clean-side anchoring and synchronously re-couples the source endpoint, yielding a measurement-aware endpoint pair with improved local bridge compatibility. The re-coupled endpoints further define a posterior-informed transport direction for advancing the sampling process. We also introduce the Posterior Bridge Defect, which jointly characterizes measurement error, deviation from the flow prior, and bridge mismatch, and leads to explicit updates for clean-side anchoring and source-side re-coupling. Extensive experiments on multiple natural and medical image restoration tasks demonstrate that ReBridge-Flow effectively alleviates bridge mismatch and improves the structural consistency of restored images. --- ### Ctrl-F-Resist. Practices, Challenges, and Technical Needs of Civil Society Organizations Monitoring the Far-Right Online Authors: Elisabeth Steffen, Helena Mihaljević Published: 2026-09-01 Categories: cs.HC, cs.AI, cs.CL, cs.IR arXiv: https://arxiv.org/abs/2609.00808v1 As far-right actors increasingly exploit online platforms to disseminate ideology and mobilize supporters, civil society organizations (CSOs) play a vital yet underrecognized role in monitoring antidemocratic dynamics online. Unlike fact-checkers or content moderators, CSOs engage in long-term, contextualized analysis, often in resource-constrained settings and under precarious conditions. Despite their critical societal role, CSOs face significant barriers to adopting or co-developing technical solutions, including legal uncertainty, limited platform access, and chronic underfunding. Existing research and tool development efforts have largely overlooked these actors in favor of more institutionally embedded stakeholders. This paper addresses this gap through a qualitative study with 15 practitioners from 12 Germany-based CSOs engaged in online monitoring, positioning them as key yet overlooked stakeholders in the governance of digital spaces. We explore their current practices, challenges, and expectations regarding technological support. Our findings show that monitoring remains largely manual due to the lack of tailored tools, with enhanced search capabilities emerging as the most pressing technical need. While participants express openness to AI-supported features such as media processing and content discovery, many remain skeptical of automated classification, citing concerns around trust, legal usability, and professional credibility. Grounded in these findings, we introduce a conceptual monitoring workflow and describe its implementation in an open-source Telegram monitoring prototype designed to flexibly support diverse monitoring goals. We outline concrete design, policy, and research recommendatios, and introduce the manual labor trap as an empirically grounded concept that explains why monitoring CSOs tend to remain locked into labor-intensive, low-capacity arrangements. --- ### Towards a Reliable and Practical Eval Pipeline Authors: Emma Thuong Nguyen, Abhishek Ghose Published: 2026-09-01 Categories: cs.AI, cs.SE arXiv: https://arxiv.org/abs/2609.00805v1 LLM-based software systems increasingly require effective "evals" as quality gates in the development lifecycle. However, existing work typically addresses individual aspects of eval reliability rather than the full set of practical requirements. We present an end-to-end eval pipeline that combines eval checklist creation, with learned aggregation for checklist responses, to improve agreement across LLM judges and accuracy against human judgments. The framework additionally pro- vides self-consistency, explanations, and prediction uncertainty, and we empirically demonstrate its effectiveness. --- ### TEIDAN: A Multilingual Multiparty Dialogue Corpus Authors: Taiga Mori, Koji Inoue, Mikey Elmers, Divesh Lala, Tatsuya Kawahara Published: 2026-09-01 Categories: cs.CL, cs.HC arXiv: https://arxiv.org/abs/2609.00802v1 Multi-party interaction is a central setting for human communication and a necessary target for human-agent interaction systems that must participate in group conversation. Yet available corpora often focus on meetings, task-oriented interaction, text-based interaction, or acted scenarios, and fewer resources support cross-linguistic comparison of spontaneous face-to-face triadic discussion. This paper presents TEIDAN, a multilingual multimodal corpus that currently consists of Japanese and English three-party conversations. TEIDAN records groups of three participants discussing open-ended topics with individual pin microphones, a microphone array, and participant-facing cameras, and provides IPU-based transcripts for both language portions. Earlier studies used subsets of the Japanese portion for task-specific benchmarks in multi-party dialogue modeling; in contrast, this paper presents TEIDAN as a corpus resource spanning both Japanese and English, with planned expansion to additional languages. We describe the collection design, participants, recording setup, transcription format, and corpus statistics, and provide preliminary analyses to illustrate how TEIDAN can support research on turn-taking, addressee recognition, and multimodal grounding in human-human and human-agent interaction. --- ### No Pixel Left Behind: Filling Gaps in Anime Colorization Authors: Masahiro Kono, Akinobu Maejima, Yuki Koyama, Yotam Sechayk, Takeo Igarashi Published: 2026-09-01 Categories: cs.HC, cs.CV, cs.GR arXiv: https://arxiv.org/abs/2609.00800v1 Animation production workflows often involve digital colorization of line art, where small unpainted regions ("gaps") frequently occur and remain an underexplored challenge. We conducted a formative study in Japanese animation (anime) pipelines and found that while the paint bucket tool is widely used for base coloring, tiny enclosed areas are frequently overlooked, resulting in time-consuming manual detection and filling. We introduce GapFill, a tool grounded in professional practices that reduces the effort of gap detection, zooming, and color selection. Our deep-learning method suggests appropriate fill colors by referencing surrounding regions, leveraging the flat-color nature of anime-style images. In a user study with 13 professional colorists, our system improved performance and usability in gap-filling tasks over conventional methods. The study also suggested that prediction accuracy alone is not the primary factor for usability, that appropriate colors can be contextually ambiguous, and that GapFill can complement existing tools depending on users' trust in new AI-powered assistance. --- ### Advanced Pixel Diffusion Model with Guided Sparse Global Refinement Authors: Weiyi You, Jinhua Zhang, Xingyu Zhou, Wei Long, Junyu Lou, Shuhang Gu Published: 2026-09-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.00798v1 Pixel-space diffusion has recently emerged as a promising direction for high-fidelity image generation by modeling images directly in the original pixel domain. However, pixel-space diffusion is computationally demanding due to the extremely high dimensionality of natural images. For efficiency, existing pixel diffusion models either compromise fine details with large-patch tokenization or confine subsequent refinement within individual patches. Such intra-patch refinement inevitably restricts structural continuity across patch boundaries and long-range token interactions, limiting refinement quality. To address these issues, we propose PixSGR, a novel Pixel diffusion framework with Sparse Global Refinement tailored for modeling the distribution of natural images directly in pixel space. PixSGR starts from a supervised low-channel bottleneck to efficiently capture the low-dimensional manifold of natural images. It then progressively expands the channel dimensionality and spatial resolution to recover increasingly fine-grained structures. At the spatial refinement stage, coarse-scale attention maps preselect globally relevant interactions to pre-sparsify fine-scale attention, enabling non-local refinement beyond isolated patches without the quadratic cost of dense attention. Extensive experiments on ImageNet validate the effectiveness of PixSGR. It achieves an FID of 1.51 at 256$\times$256 and maintains performance when scaled to 512$\times$512, attaining an FID of 1.60. --- ### SFAD: Speculative Factuality-Aware Decoding Authors: Guanqiao Chen, Di Wang, Lijie Hu Published: 2026-09-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.00796v2 As one of the most critical challenges in large language models, contextual faithfulness directly determines their reliability in knowledge-intensive applications. This task is particularly challenging as it requires balancing factual consistency with generation efficiency. Contrastive decoding methods require dual forward passes (with and without context) to compare model outputs, doubling inference computational overhead, while post-training alignment demands extensive reinforcement learning with substantial computational overhead. To address this challenge, we present SFAD, a speculative decoding framework that enhances contextual faithfulness without inference degradation. We first construct ConFide, a preference dataset with fine-grained atomic perturbations, to train a context-faithful draft model via Direct Preference Optimization. During inference, Epistemic Friction detects potential hallucinations by quantifying distributional tension weighted by specialist certainty. When friction exceeds the threshold, Asymmetric Logit Steering refines the target distribution through residual-based logit injection; otherwise, standard speculation proceeds. Extensive experiments demonstrate that SFAD substantially improves faithfulness while achieving $2.48\times$ speedup, offering a practical solution for efficient LLMs. --- ### Agentic programs: an emerging form of scientific software in computational materials science Authors: Yunsung Lim, Haekwan Jeon, Jaesun Kim, Jisu Kim, Seungwu Han Published: 2026-09-01 Categories: cond-mat.mtrl-sci, cs.AI arXiv: https://arxiv.org/abs/2609.00795v1 Computational materials science has traditionally delegated algorithmic tasks to computers while leaving scientific judgments to humans. We argue that recent LLM-based agent harnesses enable an emerging form of scientific software, agentic programs, that combine deterministic algorithms with bounded LLM-based judgment, task-specific verification, episodic maturation, and complete delegation in production. We illustrate this concept with DeMARS, an agentic program for constructing atomistic models from experimentally measured disordered crystal structures. --- ### MADS: A Multiview Acoustic Descriptor Set Beyond Standard Spectral Summaries Authors: Utsab Ghosh, Roshni Chakraborty Published: 2026-09-01 Categories: cs.SD, cs.AI, eess.AS arXiv: https://arxiv.org/abs/2609.00792v1 Dominant audio classification pipelines rely either on compact handcrafted summaries or on fixed time-frequency frontends such as log-mel representations prior to deep modeling. While highly successful, these representations do not explicitly expose the physical dynamics of the underlying sound-generating event. We introduce MADS (Multi-view Acoustic Descriptor Set), a compact 19-dimensional physics-informed descriptor set de- signed to capture complementary spectral, temporal, mechanical, and stochastic structure in audio signals. Rather than treating sound only as a spectral pattern, MADS encodes properties related to excitation, damping, periodicity, impulsiveness, and structural consistency within a unified multi-view representation. We evaluate MADS using standard classical machine learning models on ESC-10, ESC-50, and MSoS, and compare it against two conventional handcrafted baselines: a compact 26D MFCC- based baseline and an expanded 38D spectral-summary baseline. Across ESC-10 and ESC-50, MADS achieves the strongest peak results overall, reaching 81.00% and 52.78%, respectively, while using roughly half the dimensionality of the 38D baseline. On MSoS, MADS again delivers the strongest top-end performance, reaching 67.48%. These results establish MADS not merely as a competitive standalone descriptor set, but as the foundational descriptor layer of a broader acoustically grounded representation program for future frame-level and deep-learning-compatible audio modeling. --- ### Instella-MoE Technical Report Authors: Jiang Liu, Sudhanshu Ranjan, Prakamya Mishra, Yonatan Dukler, Gowtham Ramesh, Jialian Wu, Ximeng Sun, Wen Xie, Chaojun Hou, Vikram Appia, Zhenyu Gu, Zicheng Liu et al. Published: 2026-09-01 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.00791v1 In this work, we introduce Instella-MoE, a fully open Mixture-of-Experts (MoE) language model with 16 billion total parameters and 2.8 billion active parameters per token, trained entirely from scratch on AMD Instinct MI300X and MI325X GPUs. Instella-MoE combines a sparsely activated MoE design with architectural and system-level innovations, including Gated Multi-head Latent Attention (Gated MLA) and FarSkip-Collective connectivity, enabling efficient large-scale training and inference. The model is developed through a multi-stage pipeline comprising pre-training, mid-training, long-context extension, supervised fine-tuning with feedback-driven data curation, direct preference optimization, and reinforcement learning with Multi-Teacher On-Policy Distillation. Instella-MoE achieves an average score of 76.7 across standard pre-training benchmarks, outperforming prior fully open models including OLMo-3-7B, SmolLM3-3B, and OLMoE-1B-7B, while remaining competitive with open-weight MoE and dense baselines at comparable active-parameter scales, including Moonlight-16B-A3B and Qwen3.5-4B. After post-training, our final Think checkpoint achieves an average score of 73.2 across instruction-following, reasoning, math, coding, and chat benchmarks, outperforming both fully open and open-weight models with comparable or larger active parameter counts in our evaluation. To support transparent and reproducible research, we release the complete Instella-MoE model flow, including model weights, training configurations, data mixtures, and training code. Together, these contributions establish Instella-MoE a strong, fully open foundation for efficient, high-performing MoE models and reproducible research. --- ### Subspace Levenberg Marquardt Algorithms in Training Neural Networks Authors: M. Duc Hoang Published: 2026-09-01 Categories: cs.LG, math.OC arXiv: https://arxiv.org/abs/2609.00789v1 The Levenberg-Marquardt (LM) algorithm is a well-known second-order method for rapid convergence and strong robustness when training small- to medium-sized neural networks (NNs). However, its computational and memory costs increase significantly as the number of parameters in an NN grows. To address this limitation, subspace methods have been proposed, such as the Krylov subspace LM (KSLM) and the hybrid subspace LM (HSLM), making second-order algorithms more efficient. In this work, we evaluate the subspace Levenberg-Marquardt algorithms for regression and classification tasks in neural networks. We compare the performance of subspace LM variants with the classical LM method, as well as other popular first-order algorithms, such as stochastic gradient descent (SGD) and Adam. --- ### Forbid Your Attention: Fooling Multimodal Large Language Models by Selectively Removing Intrinsic Focus in Spectral Domain Authors: Daizong Liu, Junhao Dong, Zhiyuan Ma, Xiaoye Qu, Xiang Fang, Runwei Guan, Keke Tang, Jianfeng Dong, Yew-Soon Ong Published: 2026-09-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.00788v1 Multimodal large language models (MLLMs) have extended the capability of large language models (LLMs) to process more contextual multimodal information, showing remarkable progress in diverse realistic multimodal applications. Despite their strong perception and reasoning abilities, recent studies reveal that MLLMs remain highly vulnerable to adversarial inputs, especially those targeting visual components. However, existing attacks mainly focus on global perturbations, lacking an understanding of how MLLMs internally interpret visual structures. In this paper, we make the attempt to investigate the intrinsic focus of MLLMs in the frequency domain and discover that their predictions are particularly sensitive to phase information, which encodes essential structural and semantic cues. Based on this observation, we propose a novel phase-aware adversarial attack framework that explicitly restricts adversarial perturbations to structure-relevant phase regions to suppress the MLLMs' focus for effective and imperceptible attacks. To further amplify the structural influence, we also introduce an auxiliary adversarial prompt learning module to guide multimodal misalignment around phase-sensitive regions, misleading the MLLM's attention toward targeted structural patterns. Extensive experiments on multiple representative MLLM models and datasets demonstrate the superior effectiveness of our method compared to existing attacks. --- ### StudyBench: Can Self-Evolution Squeeze Textbooks for Olympiad Capability? Authors: Yinghao Chen, Zixi Chen, Bingxiang He, Ziqing Qiao, Huan-ang Gao, Yinuo Xu, Yuxin Zuo, Zeyuan Liu, Yuhao Zhan, Chaojun Xiao Published: 2026-09-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.00787v2 Humans need to study only a handful of well-written textbooks to master a discipline and attempt its hardest problems. We argue that an ideal self-evolution method should share the same property, that is autonomously learning from raw training material for transferable problem-solving capability. However, we still lack a direct measurement for it. We introduce StudyBench, a controlled physics benchmark that directly measures how efficiently a self-evolution method converts training material into capability. We organise the test set into an Application Set, consisting of difficult textbook problems and evaluating absorption ability, and a Transfer Set, consisting of olympiad-level problems and evaluating transfer ability. Benchmarking representative self-evolution methods across three base models, we find that improvements on the Application Set rarely translate to the harder Transfer Set. A guidance ablation exposes a Guidance Gap: even the strongest method closes only a small fraction of what the same material unlocks when supplied as in-context guidance. Besides, every method hits a Compute Plateau, saturating well before exhausting its compute budget. The remaining gap is therefore a method problem rather than a data or compute problem. By offering a clean and controlled benchmark, StudyBench turns self-evolution progress from an open-ended pursuit into a measurable target for future research. Our code is released at https://github.com/thunlp/StudyBench. --- ### When Features Become Instances: Inverted Contrastive Learning for Unsupervised Feature Selection Authors: Utsab Ghosh, Roshni Chakraborty Published: 2026-09-01 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.00782v1 Unsupervised feature selection seeks a compact subset of informative features without access to class labels, making feature utility difficult to define. Existing UFS methods therefore rely on indirect structural criteria, such as similarity preservation, locality, sparsity, cluster geometry, or reconstruction quality. In this paper, we instead study UFS through representation consistency and propose Inverted Contrastive Learning for Unsupervised Feature Selection (ICLFS), a feature-wise contrastive framework that reformulates UFS as a representation learning problem over features rather than samples. ICLFS first inverts the data matrix so that each feature is represented by its sample-profile vector, then constructs multiple masked positive views together with a shuffled negative view, and learns projector-space representations that remain consistent across these structured perturbations under an InfoNCE-based objective. Motivated by recent findings that cosine-based and InfoNCE-based training affect embedding norms, we use projector-space embedding magnitude as the saliency signal for ranking features. The resulting norm-based ranking is subsequently refined through Laplacian-Gated Ranking Correction, which suppresses locally redundant candidates while preserving salient ones. Extensive experiments on 12 benchmark datasets show that ICLFS achieves the best clustering accuracy on 10 datasets against both classical and neural baselines under the standard clustering-based UFS evaluation protocol, while remaining competitive on the other two. These results show that feature-wise contrastive representation consistency provides a strong and effective alternative to neighborhood, cluster, and reconstruction-based UFS formulations. --- ### Solaris: Towards Interfaces That Are Generated, Not Coded Authors: Yuval Alaluf, Omri Avrahami, Guy Bukchin Leshem, Michal Geyer, Kfir Goldberg, Elad Richardson, Diego Alarcón, Alejandro Alvarez, Cole Garry, Anastasis Germanidis, Tenaya Goldsen, Corina Gurau et al. Published: 2026-09-01 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.00776v1 Digital interfaces are traditionally implemented through intermediate representations such as code, requiring their appearance and behavior to be specified in advance. We introduce Solaris, an interface world model that instead generates an interactive UI directly, frame by frame, in response to user actions. Solaris treats mouse interactions as conditioning signals and autoregressively synthesizes the resulting visual state at interactive speeds. To enable real-time generation while maintaining visual coherence over extended interactions, we combine autoregressive frame generation with few-step distillation and training on the model's own outputs. A language model complements the visual world model by interpreting user intent and specifying how interactions should affect the generated environment, separating high-level reasoning from visual rendering. By generating both the appearance and behavior of an interface dynamically, Solaris enables open-ended interactions that need not be explicitly programmed in advance. We view interface world models as a step toward a new paradigm for software, where interfaces are generated and adapted continuously around user intent rather than implemented as fixed collections of predefined states and --- ### VOIM: Training-Free Open-Vocabulary 3D Instance Mapping for RGB-D and Monocular SLAM Authors: Sangmin Song, Sarath Kodagoda, Marc G. Carmichael, Karthick Thiyagarajan, Amal Gunatilake, Kelly Prentice, Jodi Martin Published: 2026-09-01 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.00775v1 We present Voxel-Grounded Online Instance Manager (VOIM), a training-free voxel-grounded instance manager that builds open-vocabulary 3D instance maps from RGB-D or from monocular RGB alone, a regime no prior training-free system addresses. Online systems typically segment object instances and label them at first detection, committing when evidence is weakest. VOIM instead defers label and instance decisions until soft evidence from unmodified, off-the-shelf perception has accumulated per voxel across views. We show that the mapping stage, rather than the particular perception models, carries the result: across four perception configurations on ScanNet++, varying the region descriptor, the detector label prior and the mask source, the map exceeds the strongest online RGB-D system, OVO-SLAM, by between 4.8 and 11.7 mIoU. Perception is not neutral, and substituting that baseline's own descriptor family costs 4.1 of the margin, yet the baseline carries the marginally better 2D descriptor (33.7 vs. 31.5 mIoU over three scenes) and still realizes the weaker map. Under a like-for-like protocol VOIM reaches 44.07 mIoU on ScanNet++ against 32.37, winning all ten scenes and both aggregations (pooled 33.31 vs. 25.97), and the same system runs unchanged to fully monocular RGB, matching that baseline pooled on Replica (27.80 vs. 27.50). The advantage is regime-specific: under Replica's all-classes scoring, matched inputs give a split result, 28.60 vs. 27.50 pooled against 24.59 vs. 30.11 on the per-scene mean. Room scale is label-limited and building scale drift-limited. Labeling does not run in real time, dominated by per-class detection over the full vocabulary. The maps export occupancy grids and resolve free-form queries to object instances. --- ### Semi-Supervised Classification with Informative Missing Labels in Weibull Mixture Models Authors: Jinran Wu, You-Gan Wang, Geoffrey J. McLachlan Published: 2026-09-01 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2609.00774v1 We consider semi-supervised classification from a partially classified sample arising from a two-component Weibull mixture. The feature is observed for all data, whereas some class labels are missing. The probability of a missing label is modelled as a function of classification uncertainty, giving a feature-dependent missing-at-random (MAR) mechanism that shares parameters with the Weibull-mixture classifier. The missing-label indicators can therefore provide information about the classifier in addition to the observed features and available class labels. Under a common Weibull shape, a Bayes' rule has at most one positive decision boundary, which is unique when the rule is nonconstant; under unequal shapes, it can have two. We characterise these decision regions, derive the Fisher information for the classifier after adjustment for nuisance parameters in the missingness model, and obtain a decision-boundary expansion of the expected error rate of the plug-in sample rule relative to the Bayes error. The expansion yields classification-specific asymptotic relative efficiency formulas for the one- and two-boundary cases and shows that a positive-definite increase in Fisher information is sufficient, but not necessary, for a smaller first-order expected error rate. Numerical studies and a semi-synthetic analysis based on hard-drive failure data illustrate potential reductions in expected error rate and improvements in decision-boundary estimation from modelling feature-dependent label missingness. --- ### Deep Skew-t Mixture Models Authors: Jinran Wu, You-Gan Wang, Geoffrey J. McLachlan Published: 2026-09-01 Categories: stat.ME, stat.CO, stat.ML arXiv: https://arxiv.org/abs/2609.00773v1 High-dimensional clustering is challenging when component distributions are both heavy-tailed and directionally asymmetric. We propose a deep skew-$t$ mixture model (DStMM), a hierarchical factor-analytic mixture based on the generalised-hyperbolic skew-$t$ normal mean--variance representation. A shared inverse-gamma mixing variable is propagated along each complete latent pathway, allowing heavy tails and directional asymmetry to be modelled jointly while preserving conditional Gaussianity. Each complete pathway therefore admits an exact GHST marginal representation. We formalise the reductions to symmetric deep $t$, Gaussian deep-mixture, and single-layer GHST factor-analytic models, discuss local non-identifiability and the implementation-level parameter-counting convention, and derive the conditional generalised inverse Gaussian law used for estimation. Estimation is carried out by a stochastic/Monte Carlo EM algorithm, with an explicit implementation-based parameter count for BIC architecture comparison. Simulation studies show that DStMM performs similarly to the symmetric robust model when skewness is absent but provides increasing gains as directional asymmetry becomes stronger, particularly under heavier tails; the same qualitative behaviour persists under smaller samples and unequal mixture proportions. Two real-data applications provide complementary evidence. On the UCI handwritten-digit benchmark, DStMM gives the strongest clustering performance under a common deep architecture, while on the Gas Sensor Array Drift data, DStMM improves on both deep Gaussian and deep $t$ alternatives and, under the implemented BIC criterion, selects a non-trivial second mixture layer. Together, these results support the value of propagating skewness and heavy-tail variation through a deep latent mixture while retaining an exact pathway-level likelihood. --- ### DiagEvo: Diagnosis-Guided Self-Evolution via Hierarchical Error Memory Authors: Xincheng Wei, Yifan Ding, Yoshua Li, Dongsheng Ma, Rongxiang Weng, Xunliang Cai, Wenjian Ding, Yao Zhang Published: 2026-09-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.00768v1 Self-play is an effective paradigm for language-model self-evolution, but without guidance, solver performance can plateau or decline across rounds. Unguided methods steer question generation with signals such as difficulty, learnability, or diversity. These signals keep questions challenging and varied but do not specify which unresolved reasoning weaknesses later rounds should target. Guided methods obtain direction from external task resources, including human examples, document corpora, or specified difficulty targets, and therefore rely on task information supplied outside the self-play loop. We show that the needed direction can instead be derived from the solver's own failure history. We introduce DiagEvo, whose diagnostician extracts recurring error causes from this history and stores them in a hierarchical error-cause memory. The memory groups related causes under skill nodes and tracks each as Active or Mastered according to self-consistency on targeted questions. The challenger uses these states and recurrence counts to balance cause-targeted generation with free exploration. Double-confidence filtering retains intermediate-difficulty questions only when the most common solver answer has a clear vote lead. DiagEvo derives its curriculum from information produced during self-play, without external task resources. With the default 4B diagnostician, DiagEvo outperforms every baseline in mean accuracy across all nine benchmarks for each of the three solvers: Qwen3-4B, Qwen3-8B, and OctoThinker-8B. On Qwen3-8B, it reaches 72.3% mean accuracy across five mathematical reasoning benchmarks, 4.5 percentage points above R-Zero. Its mean accuracy across all nine benchmarks is 57.4%, 1.1 percentage points above DARC. Ablations show that the hierarchical error-cause memory and double-confidence filtering both contribute to these gains. --- ### Are You Thinking What I am Thinking? : Examining Conceptual Separation in Neural Architectures Authors: Jaee Ponde, Roshni Agarwal, Subhashis Banerjee Published: 2026-09-01 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.00764v1 Neural networks are increasingly employed to identify both well-defined and ambiguous concepts, yet output-level metrics reveal little about how those concepts are represented internally. Our study asks if these networks exhibit \textit{conceptual separation}: if examples of the same concept form coherent representations, and whether related concepts lie closer together in the representation space. We examine this conceptual organisation in Convolutional Neural Networks (CNNs) and Large Language Models (LLMs) through geometric and distributional analysis of their internal activations. In CNNs, familiar ImageNet concepts form coherent and semantically ordered representations, while this coherence weakens for unseen concepts and suffers within-class domain shift. In LLMs, clearly distinct domains remain well separated, related subdomains move closer together, and the distinction between ambiguous topics collapses at both the mean and covariance level. These results suggest that conceptual separation can reveal structure that output accuracy alone cannot, and may serve as a useful diagnostic of how robustly a model represents the concepts it is asked to identify. Code and data available on \href{https://github.com/JaeeRoshniCapstoneProject/Are-You-Thinking-What-I-m-Thinking-Examining-Conceptual-Separation-in-Neural-Architectures}{GitHub}. --- ### Automated Tree Knowledge Graph Construction using Ontology Expansion and Retrieval from Vietnamese History Textbooks Authors: Ket Doan Nguyen, Minh N. H. Nguyen Published: 2026-09-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.00763v1 Hierarchical Knowledge graph (KG)-based retrieval augmented generation (RAG) has emerged as a powerful approach for supporting large language models with structured knowledge. However, there are primary challenges: (i) the lack of methods for automatic KG construction using ontology expansion for low-resource languages such as Vietnamese, (ii) the absence of systematic evaluation for knowledge retrieval strategies leveraging the hierarchical structures. In this paper, we propose an end-to-end pipeline for KG construction and retrieval strategies evaluation. In the KG construction, we employ a three-phase hybrid relation extraction pipeline: intra-batch deduplication via Union-Find, approximate cross-batch search, and LLM extraction with a centroid filter that reduces prompts combined with a five-step dual-LLM validator to prevent bloated ontology. A two-tier architecture consists of unmergeable structural nodes to preserve the document structure and mergeable content nodes. The retrieval evaluation consists of three graph traversal strategies: Top-Down, Horizontal, and Bottom-Up, which are evaluated on a synthetically generated benchmark of 1,210 Vietnamese queries from 109 subgraphs, categorized by five query directions. In this paper, we construct the tree knowledge graph from Vietnamese high school History textbooks (nearly 400 pages) to produce 750 nodes and 4,341 semantic edges with controlled ontology growth from 40 to 41 types. Among experimental graph traversal strategies, the Top-Down strategy with structure surpasses the vector baseline by 4.7 percentage points in NDCG@10. As a result, tree-structural information provides valuable information beyond flat cosine similarity but degrades performance when the query does not require structural context. --- ### Frozen Cores Need Task Signal: Fisher-Whitened Cross-Covariance for Low-Resource LLM Adaptation Authors: Wentao Ye, Zhanming Shen, Zhiqing Xiao, Yao Ding, Haobo Wang, Gang Chen Published: 2026-09-01 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.00762v1 Parameter-efficient fine-tuning is usually framed as a question of how many parameters to update. Under a severe trainable-state budget, however, where those coefficients act is equally consequential. We study this choice through frozen-core adaptation: a calibration pass fixes left and right bases for each weight matrix, and fine-tuning optimizes only an $r\times r$ core. This removes the ability of trainable factors to repair a poor initial span and makes subspace quality directly observable. We introduce FCCA, which estimates the signed input--error cross-covariance, whitens it with diagonal Fisher moments, truncates it in the resulting local metric, maps the selected directions back, and applies thin QR to obtain stable core coordinates. Under a matched $r^2$ budget, we compare eight basis constructors on 11 tasks, four model settings, and three seeds. On Qwen2.5-3B, FCCA reaches an 83.0 macro-average, 2.3 points above the next-best matched-budget constructor, and exceeds its unwhitened RawGrad control on all 11 tasks. It ranks first at all three Qwen scales and finishes within 0.13 points of the best method on Llama-3.2-1B. Controlled ablations show gains of 2.7--17.2 points from whitening and identify QR as necessary for stable core optimization in the tested regime. Finally, FCCA comes within 0.32 and 0.23 average points of LoRA and DoRA while optimizing 36.9K rather than roughly 7.4M parameters. These results show that a carefully selected fixed span can recover most of the benefit of movable low-rank factors at a much smaller trainable and optimizer-state cost. --- ### A Unified Mechanistic Analysis of Knowledge- and Safety-Based Refusals Authors: Yuri Son, Seunghee Kim, Hyuhng Joon Kim, Taeuk Kim Published: 2026-09-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.00760v1 Large language models (LLMs) are increasingly trained to decline queries that fall outside their knowledge (knowledge-based refusal, KR) or violate safety policies (safety-based refusal, SR). Although KR and SR result in superficially similar responses, they have largely been studied in isolation, leaving open whether they share an underlying mechanism. We address this gap with a systematic study on a new dataset of 213 contrastive quadruples that jointly probe both refusal types. We find that KR and SR are governed by overlapping yet distinguishable mechanisms. Both share a refusal direction, yet the overlap is asymmetric: SR signals transfer more strongly to KR than the reverse. Type-specific specialization emerges mainly in upper layers, with KR aligning with uncertainty- and knowledge-related representations and SR with safety- and policy-related ones. We thus characterize refusal as a commit-then-specify process: a shared initial mechanism commits to refusing, then type-specific features in later layers specify whether the grounds are epistemic or normative. --- ### Compile, Don't Memorize: A Context Compilation Architecture (CCA) for In-Context Learning Authors: Jinhu Qi, Minda Hu, Wentao Zhang, Weiqiang Jin, Yanyu Chen, Junli Wang, Irwin King Published: 2026-09-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.00759v1 Large language models (LLMs) increasingly handle in-context learning (ICL) tasks where a long, novel context defines the rules, knowledge, and output schema for a series of questions. On benchmarks that grade against every detail of the context, even strong open-weights models pass only 12-16% of tasks: a single overlooked rule fails the whole response. We argue this brittleness is structural: the dominant "read-and-reason" paradigm asks the model to extract, plan, generate, and self-verify in one forward pass. We therefore ask whether explicit context compilation can fix it, how it compares to existing long-context strategies (gist retrieval, multi-agent self-play), and where the resulting harness benefit holds across task structure and model scale. We propose the Context Compilation Architecture (CCA), whose central novelty is a typed intermediate representation (IR) with fixed slots (rules.{must_do, must_not, conditional}, output_spec, available_tools, data_profile) into which any prose context is compiled once; executable verifiers and a violation-gated correction loop follow as downstream consequences. On CL-bench (1,899 tasks across 4 open base models), CCA outperforms vanilla prompting and two long-context baselines (ReadAgent-P, Ctx2Skill) on every base model, lifting Kimi K2.5 from 15.4% to 21.4% with gains concentrated on rule-dense sub-categories. Code and cached completions are available at https://github.com/TonyQJH/cca-emnlp2026. --- ### Joint Training Is Not Enough: Conditioned Cross-Granularity Training for Multimodal Document Understanding Authors: Chengguang Gan, Yunhao Liang, Hanjun Wei, Qinghao Zhang, Shiwen Ni Published: 2026-09-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.00756v1 The Mutual Reinforcement Effect (MRE) asks whether a fine, span-level and a coarse, document-level task help each other when one model handles both. We test it in multimodal document understanding on three corpora, two of receipts and one of scanned business forms, comparing single-task, joint and conditioned training, which puts one granularity's gold output in the other's prompt during training only. We build Doc-MRE, an annotation layer pairing gold field extraction (point) with four document-level facets (line), from a three-judge LLM committee under a pre-registration, validated by blind re-annotation. One predicate, fixed in advance: at a shared recipe, a regime reinforces if it beats the matched single-task model on both granularities. Mixed joint training, the arrangement prior MRE work assumes, reinforces on no corpus at the main scale: it is below both single-task models on CORD and trades one granularity for the other on the two others, as single-task tuning does. Conditioned training reinforces on two of the three, CORD (+0.5 point, +4.8 line) and the forms corpus (+7.2 point, +11.0 line), resolvably on the coarse side and directionally on the fine one, and trades on WildReceipt; at that recipe no alternative measurably beats it on either side anywhere. Two byte-identical-prompt controls separate content from format: shuffled conditioning destroys the coarse-side skill but costs the fine side far less, and a neutral-content control reproduces the whole fine-side gain on WildReceipt, which is therefore prompt structure but buys nothing resolvable on the other two. On the forms corpus conditioning buys collapse avoidance: mixed training and the neutral control both assign the majority semantic label to all 50 test documents; only conditioning recovers the gold distribution. Probes find the information decodable under every regime with no resolvable increase under conditioning. --- ### S^3martCirc: Self-supervised Smart Circuit Discovery Authors: Wendy Zheng, Yinhan He, Liang Wu, Jundong Li Published: 2026-09-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.00755v1 Large Language Models (LLMs) have demonstrated remarkable performance across diverse tasks, from text summarization to question answering. Despite these capabilities, their black-box nature obscures internal decision-making processes. Mechanistic interpretability (MI) aims to address this by reverse-engineering neural networks into human-understandable algorithms. Current MI approaches for LLMs typically follow a two-stage paradigm: first identifying important components (circuit discovery), where components are typically individual nodes such as an attention head or feedforward neuron, and second determining the role they play in a certain task (functional interpretation). However, this sequential approach overlooks a fundamental insight: a component's importance and its functional role are inherently codependent. Unifying these stages presents two key challenges: (1) functional roles are often tied to specific nodes or components, limiting generalization, and (2) their identification relies on subjective interpretation rather than quantifiable metrics. To address these challenges, we propose S^3martCirc (Self-supervised Smart Circuit Discovery), a unified framework that simultaneously discovers circuits and interprets functionality. S^3martCirc abstracts node behavior into two general computational roles that generalize across tasks and defines a quantitative metric for assigning them, enabling importance and functional role to be discovered jointly rather than in sequence. Extensive experiments show that our framework outperforms existing methods in circuit discovery. --- ### How Do Language Models Choose Between Context and Memory? Authors: Benjamin Shih, John Winnicki, Arianna Cao Published: 2026-09-01 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2609.00753v2 When contextual information conflicts with knowledge stored in model parameters, activation directions can be used to decode and steer which source the model follows. However, successful steering does not establish that the unedited model uses those directions to choose between sources, or that they remain effective across tasks. To test these possibilities, we vary the stated authority of contextual claims while holding their content fixed. We first estimate authority directions from prompts in which context and parametric knowledge agree, then test their causal contribution when the two sources conflict. Interchanging naturally occurring activation values along these directions between matched high- and low-authority prompts reproduces 30--68% of the authority-induced shift in source choice across Qwen, Llama, and OLMo models, whereas matched controls reproduce almost none. We next ask what transfers across tasks: the learned direction versus the activation values exchanged along it. Using a direction learned on another task closed 9% of the source-choice gap, compared with 57% when learned on the task being evaluated. Both interventions exchanged activation values from the evaluated task. In a separate experiment, we kept its learned direction but exchanged values taken from another task, which closed 68% of the gap. These results show that authority-related activation values can causally influence source choice across tasks when inserted along directions learned for the task being evaluated. --- ### ContextPipe: Database-Inspired Context Assembly for Long-Horizon Agents Authors: Peng Xu, Zuyu Zhang, Yuze Sun, Feng Tian, Long Wang, Chen Zhang Published: 2026-09-01 Categories: cs.AI, cs.DB arXiv: https://arxiv.org/abs/2609.00749v1 Long-horizon large language model (LLM) agents require context assembly: the runtime must decide what to include in each prompt, in what order, and when to compact history under a hard context-window budget and a byte-sensitive prompt cache. In production agentic systems, this logic is scattered across prompt builders, ad hoc compaction routines, cache-break workarounds, and per-provider shims. We argue that context assembly is structurally isomorphic to query execution in a relational database: both execute under a hard budget, exploit a tiered cache, and leverage statistics. We adopt this discipline in ContextPipe: a five-phase pipeline (Plan Bind Optimize Execute Feedback) backed by a structured data-source catalog, a deterministic cache-aware optimizer, and an EXPLAIN ANALYZE trace. We show that context in ContextPipe is auditable, replayable, and failure-isolated. A preliminary evaluation using the SWE-bench Pro Qutebrowser subset shows that, compared with the append-only context construction policy, ContextPipe reduces total token volume by 31%, LLM calls by 23%, and response time by 9%, at the cost of a lower KV cache-hit ratio. --- ### Measuring Optimal Transport in Transformer Depth Authors: Alexandre Quemy Published: 2026-09-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.00748v1 A transformer carries each token's state from layer to layer, and the whole vocabulary carried together forms a cloud that moves with depth. We ask whether a trained network moves this cloud the way optimal transport would: at the cheapest cost, and along the map that pairs each token with its optimal destination. We measure both on Pythia-160m and Pythia-410m, with an exact assignment between consecutive layer clouds, a measured sampling floor, calibration on couplings known to be optimal, and a split of the cost into the common shift of the cloud and the token-specific moves. At the last layer, both models move their tokens where the optimal-transport map sends them, at the optimal cost for Pythia-410m and slightly above it for Pythia-160m. At the first layer they do not. In between, single layers can be judged on cost at only two of ten transitions, and blocks of several layers move the cloud at close to the optimal cost. The agreement at the last layer is much weaker at initialisation (0.64 against 0.86) and grows with training. --- ### Can Large Language Models Forecast What Researchers Study Next? Authors: Fenghai Li, Zihan Tang, Haofei Yu, Yining Zhao, Jiaxuan You Published: 2026-09-01 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.00747v1 Large language models increasingly generate research ideas, yet judging their novelty or feasibility at generation time does not establish whether they anticipate subsequent work. We introduce IdeaForecastBench to evaluate research idea forecasting. Given a community's literature up to a cutoff, a system produces up to five ranked ideas, which are evaluated against later papers. The benchmark comprises 624 rolling episodes across 52 topics, with a fixed retrieve-then-judge protocol and separately reported results from two judges. We compare five history-compression strategies across GPT-4.1, Qwen2.5-7B/14B, and Qwen3.5-9B, together with a learned Mode-Decomposition Forecaster (MDF). Under the primary GPT-4.1-mini judge, Summary improves on Direct in Hit@5 and Precision@5 across all four backbones. Qwen2.5 scores above GPT-4.1, whereas Qwen3.5 scores below it. An outcome-blind assessment finds that Qwen2.5 produces broader forecasts, but does not identify how much breadth contributes to its advantage. Threshold and judge diagnostics further clarify the limits of interpreting realization as precise anticipation. IdeaForecastBench provides a common task for studying which research ideas a community subsequently pursues and how reliably this outcome can be measured. --- ### Text Capability Loss in Vision-Language Adaptation: An Attention-Sink Diagnosis Authors: Minsik Choi, Geewook Kim, Young Geun Kim Published: 2026-09-01 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.00746v1 Fine-tuning a pretrained LLM into a vision-language model (VLM) can erode the backbone's text capability, with the damage concentrated on tasks that require following exact output rules, such as instruction following, chain-of-thought reasoning graded on a strictly parsed final answer, and similar evaluations with strict graders. We trace this gap to attention-sink corruption: VL fine-tuning perturbs the early sink position that anchors a large fraction of attention probability, and how well the base LLM preserves its sink tracks how much of the affected capability survives adaptation. Building on this view, we introduce Sink Strength, a single scalar computed on the base LLM in a few seconds on a single GPU that predicts post-VL degradation without any VL training. It consistently tracks relative degradation across the six VLM-LLM pairs and multiple format-sensitive tasks. Complementing this diagnostic, we find that post-pretraining QK-RMSNorm injection fails to reproduce the protection of native QK-RMSNorm, while several off-the-shelf weight-merging settings fail to recover the lost capability after VL training. These negative results underscore the value of screening backbones with Sink Strength before VL training and narrow the intervention space toward head-selective training-time protection. --- ### Feed-Forward Multi-view Multi-person Reconstruction with Contrastive Human-Aware 3D Representation Authors: Yuanwang Yang, Buzhen Huang, Zongxuan Ren, Jing Huang, Kun Li Published: 2026-09-01 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.00745v1 Multi-view human reconstruction has been extensively studied under simplified settings, yet robust and efficient multi-person reconstruction in unconstrained environments remains challenging. Existing bottom-up methods often rely on accurate camera calibration and explicit cross-view matching, and therefore struggle with severe occlusions and ambiguities. We propose a new top-down paradigm that maintains a unified, instance-centric human-aware 3D space, enabling simultaneous camera calibration, cross-view association, and human reconstruction via cross-modal contrastive learning. Observations from multiple views are lifted and fused into this shared 3D space, where geometric structure, visual appearance, and human-centric semantic cues are jointly encoded at the instance level. We further introduce a spatial contrastive learning strategy that aligns 3D features corresponding to the same human instance across different views and modalities while separating different instances. This enables correspondence reasoning, semantic aggregation, and instance discrimination to be performed natively in 3D, improving cross-view consistency and robustness under severe occlusions. Finally, structured human body models are recovered in a feed-forward manner by regressing SMPL parameters from instance-level 3D human tokens. Extensive experiments demonstrate robust, accurate, and efficient multi-view human reconstruction in challenging real-world scenarios. --- ### Mind the Rift: Cross-Scale Coupling Mismatch for AI-Generated Video Detection Authors: Siyu Li, Jin Yang, Weiheng Liang Published: 2026-09-01 Categories: cs.CV, cs.MM arXiv: https://arxiv.org/abs/2609.00742v1 As AI video generators achieve cinematic realism, reliable detection becomes essential for safeguarding digital trust. We identify cross-scale coupling mismatch as a new forensic signal, where scale refers to the level of abstraction (semantic dynamics vs. pixel-level residuals): in natural videos, macro-level temporal dynamics and micro-level residual patterns are intrinsically coupled by the unified imaging physics pipeline, whereas AI generators, whose training objectives do not explicitly preserve this joint distribution, systematically violate this coupling. Detecting such mismatch is challenging because it requires independently extracting information at both scales while simultaneously quantifying their cross-scale relationship. We propose RIFT (Representation Inconsistency Forensics on Trajectories), an orthogonal forensic framework that addresses this through three interlocking components: a macro stream that builds a dynamic baseline of expected temporal evolution via differential geometry and persistent homology on learned manifold trajectories, a micro stream that acts as a sensitive forensic probe via steganalytic filtering and temporal modeling, and a coupling divergence module that measures the conditional dependency between the two streams. Gram-Schmidt orthogonality guarantees the information-theoretic validity of this measurement. Experiments on two benchmarks (VidProM, 120K videos, 7 generators; GenVidBench, 68K videos, 4 generators) demonstrate that RIFT achieves 99.33% and 99.72% F1-score respectively, with 97.87% unseen-generator detection rate in leave-one-out evaluation, while exhibiting encoder agnosticism: scaling from ViT-S/14 (22M) to ViT-L/14 (300M) changes F1 by less than 0.1%, and switching to a different encoder family (DINOv1) reduces F1 by only 0.73 pp. Code is available at https://github.com/Litsay/RIFT --- ### Escaping Reasoning Basin Collapse with History-Biased Search Authors: Lu Cheng Published: 2026-09-01 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.00738v2 Inference-time search with large language models (LLMs) often concentrates on a small set of structurally or semantically similar trajectories, leaving alternative reasoning strategies underexplored---a failure mode we call \textit{reasoning basin collapse}. We introduce \textsc{BASIN}, a training-free, history-biased search method that groups reasoning states into basins and accumulates a revisit penalty on repeatedly selected basins, reallocating a fixed inference budget toward underexplored reasoning strategies. Under matched inference budgets, \textsc{BASIN} improves over Tree of Thoughts (ToT) by up to $+22$pp on Game of 24 and $+6.7$pp on MuSR. Because indiscriminate diversification can over-explore once search has found a promising basin, we further introduce \textsc{QA-BASIN}, a quality-aware variant that weakens the revisit penalty for high-quality basins and yields more robust gains. To characterize when basin-aware search helps, we introduce the \emph{redundancy gap} $Δ$, which measures the difference in search concentration between correct and incorrect predictions: standard ToT often operates near $Δ\approx 0$, whereas \textsc{BASIN} consistently shifts $Δ$ positive. Together, these results identify reasoning basin collapse as a failure mode of inference-time search and show that history-dependent bias provides a simple, training-free mechanism for escaping redundant reasoning under fixed compute. Code is available at https://github.com/GitHubLuCheng/basin --- ### Online Self-Weighted Fine-Tuning Authors: Haiquan Wen, Yiwei He, Bei Peng, Guangliang Cheng Published: 2026-09-01 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.00734v1 Standard supervised fine-tuning (SFT) assigns the same explicit loss weight to every expert demonstration, regardless of the model's changing competence over training queries. Reinforcement learning (RL) based methods adapt update strength using model-generated rollouts, but often require substantially more sampling and can be unstable on hard tasks. We propose \textbf{Online Self-Weighted Fine-Tuning (OSW-FT)}, a simple method that augments SFT with online, trajectory-level weighting. For each query, OSW-FT estimates the model's current success rate using a small number of inference-only rollouts and rescales the standard SFT loss accordingly. The optimization direction remains anchored to the expert trajectory, while the update magnitude adapts online. For binary-verifiable reasoning, we connect this weighting to SFT and RL at the gradient level, inspired by variance-reduction principles. The resulting estimator is unbiased for the exact OSW-FT surrogate update for any finite rollout count, and we analyze convergence with respect to the corresponding surrogate objective. Evaluated across Qwen3 series ranging from 0.6B to 4B on multiple challenging benchmarks (e.g., AIME), OSW-FT consistently improves over SFT on small-to-medium scale models. OSW-FT offers a favorable compute-performance trade-off as a practical approach for fine-tuning small-to-medium LLMs on binary-verifiable reasoning tasks with only \textbf{2 online rollouts}. ---