# Research Papers — 2026-09 · bundle page 219 of 266 > Full-text markdown of papers #10901–10950 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/219 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/218.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/220.md) ### Recompilation Is Not Enough: Test-Guided Decompiled-C Repair Authors: Yuhan Huang, Puzhuo Liu, Jianlei Chi Published: 2026-09-07 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2609.07201v1 Decompiled C often becomes recompilable only after repair, but recompilation alone does not establish test-observed behavior. A recompiled command-line binary can still parse options incorrectly, print different bytes, or return a different exit status. We present a few-step workflow for repairing decompiled C using compiler feedback and related official tests. Compiler and linker diagnostics first guide build repair. Once the repaired C recompiles into a binary, smoke checks and related official tests expose behavioral discrepancies for semantic repair. In a preliminary static-enriched evaluation on 104 Coreutils 9.5 binaries with available decompiler exports and deterministic exact-output smoke comparisons, 91 binaries (87.5%) recompile and pass the test gate; 9 do not recompile within the repair budget, and 4 recompile but still fail the test gate. The result suggests that test-gate feedback can make LLM-assisted repair of decompiled C more auditable than compile-only recovery. --- ### KABURI-TTS: Phoneme-Keyed Activity-conditioned Bi-channel Utterance Rendering for Interaction Authors: Ryuichiro Higashinaka, Shinnosuke Takamichi, Tetsuji Ogawa Published: 2026-09-07 Categories: cs.SD, cs.CL arXiv: https://arxiv.org/abs/2609.07200v1 Realizing full-duplex spoken dialogue requires large amounts of two-channel, one-speaker-per-channel conversational speech data. Although conversational text-to-speech (TTS) engines have been developed, they are not necessarily robust to two-party simultaneous phenomena such as backchannels, interruptions, and overlaps that occur while the interlocutor is speaking. In this work, aiming at conversational speech synthesis that reproduces human-like overlap, we propose KABURI-TTS. KABURI-TTS takes a per-speaker phoneme raster as input and renders the speech of the two speakers on separate channels, conditioned on the per-frame phonemes and the voice activity derived from them. Because the phoneme raster is supplied by a separate module, the proposed method enables controllable generation of one-speaker-per-channel, two-party spoken dialogue. A user evaluation shows that, compared with strong baselines, the proposed method attains higher naturalness at both the utterance and the interaction level. Furthermore, an analysis of voice activity confirms that the proposed method produces more overlap and more frequent turn-taking. --- ### Protocol effects on feature-based hardware-Trojan detection across Trust-Hub families Authors: Hang Xiao, Chuhong Xu, Kainan Zhou, Gangzhen Qian, Lu Yi Published: 2026-09-07 Categories: cs.LG, cs.AI, cs.CE, cs.DB, cs.GT arXiv: https://arxiv.org/abs/2609.07199v1 Trust-Hub reuses host circuits: several files differ mainly in the inserted Trojan. When gates from sibling variants enter both training and test folds, a detector can benefit from host logic it has already seen. We measure that effect instead of proposing another classifier. The corpus contains 49,124 gates from 16 netlists grouped into five host families. We left the parser, 36 gate features, class weighting, model settings, threshold, and family-level aggregation unchanged and altered one choice: the test boundary. The three settings draw test gates from the pooled corpus, withhold a complete netlist, or withhold every variant of one host. The choice matters. Random forest records F1/AP of 0.914/0.978 with pooled gates, 0.636/0.851 with one netlist held out, and 0.460/0.577 with a host family held out. XGBoost falls from 0.946/0.976 to 0.464/0.544 across the same comparison. Logistic regression loses AP, although its fixed-threshold F1 is not monotonic. Each family shows the same pooled-to-family direction. Feature removal, repeated model and simulator seeds, score normalization, parser-related exclusions, and a smaller sample change the size of the gap without reversing it. Aggregation also matters: a gate-weighted average is dominated by the larger ISCAS files, so the headline values give each host family one vote. Bootstrap and jackknife summaries keep the gap positive, but their folds reuse training families. We treat the five family rows as descriptive evidence rather than independent trials. Five host families are too few for a population claim, and the experiment says nothing about transfer to a new cell library or an industrial design. It supports a narrower conclusion: sibling benchmark variants can inflate apparent transfer. Benchmarks with several variants of one host circuit should report family-aware holdouts and all five family results beside pooled scores. --- ### EmoMed: An Emotionally-Aware Agent for Multimodal Medical Support with Real-Time Information Retrieval Authors: Ivan Nasonov, Nikita Glazkov, Ivan Makovetskiy, Mikhail Mozikov, Daniil Sukhorukov, Andrey Savchenko, Ilya Makarov Published: 2026-09-07 Categories: cs.AI, cs.CY, cs.HC arXiv: https://arxiv.org/abs/2609.07194v1 We present EmoMed - a multimodal medical consultation agent that adapts its responses based on users' emotional states while maintaining clinical accuracy. The system processes text and medical images, detects affect indicators (anxiety, confusion, urgency) from user input, and adjusts response tone, structure, and detail level accordingly. To ensure factual reliability, the agent grounds clinical information through a dual retrieval mechanism: web-based fact-checking and an API-connected, continuously updated medical knowledge base. We evaluate our approach across seven state-of-the-art language models (GPT-4/5, Qwen3, Llama 4, Gemini 2.5, Grok4, Claude3) using comprehensive metrics including LLM-as-judge assessments, MedQA style accuracy tests, BERT Score, safety/helpfulness ratings, and multimodal medical benchmarks. The results demonstrate that emotionally adaptive responses consistently outperform neutral baseline across evaluation dimensions, without compromising clinical accuracy. A controlled user study validated these findings, with participants reporting improved perceived empathy and communication clarity, while maintaining trust in factual accuracy. Source code: https://github.com/NasonovIvan/EmoMed-Agent --- ### FedRAW: Preserving Rare-Label Influence in Asynchronous Federated Learning Authors: Prashant Bajpai, Divya Saxena, Philippe Lalanda, German Vega Published: 2026-09-07 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.07192v1 Asynchronous federated learning improves scalability by updating the global model from a server-side buffer of client updates as they arrive, rather than waiting for all selected clients to finish. While efficient, this arrival-driven aggregation can silently distort representation learning under heterogeneous participation. We identify silent rarity failure, a hidden failure mode in which clients holding rare labels contribute too weakly to the global model even though its overall accuracy appears largely unaffected. This failure arises from two coupled effects: rare-label clients may submit updates less frequently when they are slower or less available, creating participation bias; and once their updates enter the buffer, standard asynchronous aggregation assigns them no compensating influence, creating aggregation bias. We propose FedRAW, a fully server-side aggregation method that preserves rare-label influence without changing local training, client objectives, or communication protocols. FedRAW combines client-level update deduplication, which prevents frequently arriving clients from repeatedly dominating the update buffer, with rare-label-aware weighting, which increases the influence of clients carrying low-coverage labels. We formalize silent rarity failure through participation and aggregation bias, and show that FedRAW increases rare-label client influence over uniform aggregation while preserving convergence. Across EMNIST Balanced, CIFAR-10, HAM10000, and ISIC-2019, FedRAW improves rarelabel accuracy while preserving comparable global accuracy and adding negligible server-side computation. --- ### CHILD: Human-in-the-Loop OOD Detection for Safe Clinical Deployment Authors: Jinlun Ye, Kaiyue Lu, Runhe Lai, Xinhua Lu, Jia-Xin Zhuang, Ruixuan Wang Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07188v1 Out-of-distribution (OOD) detection is critical for safe deployment of medical AI systems. Recently, test-time adaptation (TTA) has emerged as a new paradigm for OOD detection, automatically adjusting detector behavior during deployment. However, such automatic adaptation mechanisms may raise safety concerns in safety-critical clinical environments. While physician oversight can mitigate these risks, it is resource-intensive and must be judiciously allocated. To reconcile safety with efficiency, we propose CHILD, a training-free framework designed to enhance streaming OOD detection via sparse human feedback. Operating under strict budget constraints, CHILD employs an adaptive risk-aware sample selection mechanism to pinpoint only the most decision-uncertain samples for review. Crucially, it maximizes the utility of this sparse feedback through a retrieval-based score calibration module, which refines model predictions using a compact feature cache without any parameter updates. Extensive experiments on four medical benchmarks demonstrate that CHILD turns limited supervision into significant reliability gains: with a sparse feedback budget of only 5%, it reduces the average FPR95 from 72.63% to 60.26% and improves AUROC from 75.53% to 81.85%, consistently outperforming state-of-the-art baselines. Our code is publicly available at https://github.com/figec/CHILD. --- ### SIFTING: A Novel LLM-Based Framework for Structured and Transparent Information Extraction from Clinical Free-Text Reports, with Application to Tumor Staging in Lung Cancer Authors: Mirco Hess, Gerben van Veenendaal, Joris Wakkie, Yiwen Soo, Malcolm H. Lawson, John D. Maclay, Arjun Nair, Neal Navani Published: 2026-09-07 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.07185v1 Background: Large language models (LLMs) show promise for extracting information from clinical free-text documents, but their outputs are often unstructured and lack traceability, complicating validation and adoption in clinical workflows. In this work we introduce SIFTING, an LLM-based framework designed to address these shortcomings. Methods: SIFTING combines the language comprehension capabilities of LLMs with segment-level processing and structured prompts with strict output control, linking findings to the source text to enable both accurate and transparent information extraction. To demonstrate its capabilities, we applied the framework to the task of extracting tumor T-stage information from 130 lung cancer radiology reports (SIFTING-T-stage). A compact 4-bit quantized version of the open-source LLM Llama-3.3-70B (35 GB) was used in a fully self-hosted setup, providing full control over data and model. Performance was evaluated against a reference standard created by four clinical experts and compared with a range of LLMs as used in a conventional single-prompt approach, using bootstrap resampling to estimate confidence intervals. Results: SIFTING-T-stage achieved an accuracy of 90% (95% CI: 84-95) against the reference standard. We found its performance to be comparable to even the largest state-of-the-art LLMs with reasoning capabilities and to be interchangeable with clinical experts (p < 0.001), while at the same time offering full traceability through source text references. Conclusion: SIFTING enables accurate, structured, and traceable information extraction from clinical free-text documents. It ensures data control, reproducibility, and verifiable outputs that can support clinical validation and workflow integration. --- ### Tensor network representations of discrete maximum entropy distributions via mean polytopes Authors: Alex Goessmann, Martin Eigel Published: 2026-09-07 Categories: math.ST, cs.AI, cs.LG, math.CO arXiv: https://arxiv.org/abs/2609.07184v1 We present tensor network representations for discrete maximum entropy distributions under expectation constraints. To this end, we introduce Computation-Activation Networks (CompActNets), a tensor network architecture that subsumes exponential families. By leveraging the geometry of the convex polytope of realizable expectation vectors, we represent any maximum entropy distribution in the same architecture. We exploit the fact that proper faces of this polytope correspond to the boundary closure of exponential families, which restricts the distribution's support. We then derive explicit representations for the support within the CompActNet architecture. The proposed framework suggests tensor network ranks as complexity measures for faces. Finally, a case study on Boolean statistics links the geometry of 0/1-polytopes directly to propositional formulas. --- ### CircuitLens: Reasoning Circuits as Data Selection Signals for Reinforcement Learning with Verifiable Rewards Authors: Zhuofan Chen, Ziqian Jiao, Yikai Cui, Zhixin Cai, Jun Bai, Wenge Rong Published: 2026-09-07 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.07183v1 Reinforcement learning with verifiable rewards (RLVR) is sensitive to which problems a model trains on, yet existing selection criteria--difficulty filtering, hand-curation, reward-trajectory scoring--assess data value as an intrinsic property of problems, independent of the model that will learn from them. We introduce Circuit Reasoning Score (CRS), a selection signal derived from 46 reasoning-sensitive attention heads identified via contrastive ablation, computed in a single forward pass on the frozen base model without reward labels or rollouts. CRS runs against the intuitive hypothesis that stronger reasoning-circuit engagement produces better training data: on Qwen2.5-Math-7B, the lowest-engagement decile improves over random selection on three medium-difficulty benchmarks (GSM8K +2.0 pp, OlympiadBench +1.6 pp, Minerva +2.9 pp), while the highest-engagement decile gains less and is indistinguishable from the middle decile. The advantage has boundary conditions: on a domain-curated pool no selection method separates from the others; at 1.5B scale the useful direction differs; and the lowest-reward training condition produces the strongest downstream generalization. Within the Qwen2.5-Math settings tested, RLVR data selection appears regime-dependent rather than reducible to a static ranking of problem quality. --- ### Deep Learning for Biopsy-Free Subtyping of Basal Cell Carcinoma from Dermatoscopic Images Authors: Alexandros Papadopoulos, Chrysa Episkopou, Ioannis Sarafis, Aimilios Lallas, Anastasios Delopoulos Published: 2026-09-07 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.07180v1 Basal Cell Carcinoma (BCC) is the most common type of skin cancer, accounting for nearly 80% of skin cancer di- agnoses. Its optimal clinical management is guided by the distinct histopathologic subtype, with aggressive variants requiring more drastic measures. In current clinical practice, subtyping relies on skin biopsies, a procedure both costly and invasive. In this paper, we conduct a preliminary investigation into using deep learning for BCC subtyping, solely from a single dermatoscopic image of the lesion. Given the limited data at our disposal, we employ pre-trained vision transformers (ViTs), a state-of-the-art family of models highly effective for challenging downstream tasks with limited labeled data. Through repeated stratified k-fold cross-validation, we demonstrate that ViTs can achieve superior performance (AUC 0.784 on a dataset of 1271 dermatoscopic images of various BCC subtypes) over standard CNN-based baselines as well as previously-reported human reader perfor- mance, on the task of differentiating aggressive BCCs from other subtype families. These initial findings highlight the potential of combining deep learning and dermatoscopy to provide a biopsy- free alternative for BCC subtyping, thus aiding in improving treatment planning and patient outcomes. --- ### Weakly-supervised Kidney Tumor Classification from CT Scans with Multi-Instance Learning and Anatomical Filtering Authors: Joonas Ariva, Dmytro Fishman Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07178v1 Deep learning models for CT scan analysis are often limited by the scarcity of precise pixel-level annotations, which require significant radiologist effort to produce. Training on scan-level labels alone reduces annotation requirements but introduces challenges: low supervision ratios and large input volumes make models prone to overfitting and shortcut learning. In this work, we investigate two complementary methods to address these challenges: multi-instance learning (MIL) and anatomical filtering. MIL divides CT volumes into 2D slice instances, enabling efficient 2D architectures with ImageNet pretraining rather than computationally demanding 3D models. Anatomical filtering uses Compass, our self-supervised body part regression model, to crop scans to pathology-relevant subregions without requiring segmentation masks. We evaluate two MIL frameworks - Attention-based MIL (ABMIL) and FocusMIL - on kidney tumor classification across one internal dataset (TUH) and two external datasets (KiTS23 and TCGA-KiRC). Our best models achieve F1 = 0.83 on the internal test set using only scan-level labels. We further show that anatomical filtering with the Compass model is critical for the out-of-distribution generalization of embedding-based ABMIL, while instance-based FocusMIL demonstrates greater inherent robustness to distribution shift. While evaluated on kidney tumors, we consider this a proof-of-concept for a broader weakly supervised CT classification pipeline applicable to other organs and pathologies. --- ### PhysMAS: Physics-Grounded Multi-Agent Synthesis of Compositional 4D Gaussians Authors: Jiang Qin, Chunji Lv, Yangguang Wei, Yang Gao, Ming Liu, Lizhong Ding, Ye Yuan, Yinjie Lei, Changsheng Li Published: 2026-09-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.07174v1 Efficient, fully automatic, and physically plausible 4D Gaussian synthesis is an important goal for dynamic scene generation. Recent physics-based methods couple 3D Gaussians with the Material Point Method (MPM) to generate physically driven motion, but extending this paradigm to heterogeneous multi-part objects and interacting multi-object scenes remains challenging. Object-level physical assignment collapses distinct parts into a single material state, while one-shot predictions from large language models, vision-language models, or agents neither reliably bind different materials to identified parts nor verify that the resulting MPM configuration is executable. Score Distillation Sampling (SDS)-based parameter optimization, meanwhile, requires repeated per-scene score evaluations and gradient backpropagation, incurring lengthy optimization and potentially yielding suboptimal or unstable solutions. We therefore present PhysMAS, a physics-grounded multi-agent framework. From a motion prompt and four scene views, an Object-Part Scene Agent establishes persistent identities and calls a Material Reasoning Agent for part-wise profiles. It invokes solver-aware skills to bind these identities and profiles to per-particle MPM fields and execute all objects in a shared domain; the framework then screens candidate forward-simulation results. This supports heterogeneous multi-part and interacting multi-object scenes without per-scene diffusion-score backpropagation. Extensive experiments demonstrate that, compared with recent physics-based 4D Gaussian baselines that rely on SDS, PhysMAS achieves better semantic alignment and perceived physical plausibility while requiring less runtime. --- ### The Oversight Gap: What LLM Safety Monitors Miss, and Why It Is Not Capability Authors: Xin Xu Published: 2026-09-07 Categories: cs.LG, cs.CR arXiv: https://arxiv.org/abs/2609.07162v1 Several properties safety monitors are asked to certify, among them cross-tenant noninterference, sandbagging and evaluation awareness, are 2-safety hyperproperties, witnessed only by two executions. The standard consequence is a binary impossibility: one trace cannot decide them. We replace the binary with a measurement. A tight bound puts the balanced accuracy of any single-trace monitor at $\tfrac12+\tfrac12\,TV(P_0,P_1)$, turning undecidability into a graded detectability frontier and defining an oversight gap: a monitor's shortfall below it. On a leak family with closed-form $TV$, nine LLM monitors are optimal at $TV=0$ but capture little signal as $TV$ grows; at $TV=1$, where a 20-line membership check scores $100\%$, they average $60.9\%$. That shortfall is mostly not capability: naming what to check closes $61\%$ of it while leaving the $TV=0$ control at chance. The same split runs through a $2{\times}2$ factorial: an imagined second run leaves monitors at chance ($50.4\%$) while the same rule on an executed second run reaches $90.0\%$, and a stored oracle without a comparison procedure yields only $68.2\%$. Information and procedure are each necessary and neither is capability. Under nondeterminism, replay tracks a closed-form $k$-replay curve only under the right projection, and a projection frontier shows the resulting dilemma is forced: narrow misses $98.6\%$ of off-channel leaks, broad flags $75.7\%$ of clean traffic, and attainable accuracy decays like $1/(qm)$ in the benign-variation rate and the channel count. Finally, two frontier LLM judges certified an earlier version of our own benchmark as sound while a sign test found a directional bias ($p=2.7\times10^{-5}$) that invalidated three of our findings. Construction validity for hyperproperty benchmarks should be proved mechanically, not audited by models. --- ### In-Place Instruction Following in Diffusion Language Models Authors: Zheng Nie, Zherui Li, Jiaming Zhang, Kun Wang, Zhenhong Zhou, Yufei Guo Published: 2026-09-07 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.07160v1 Diffusion Large Language Models (dLLMs) generate text via bidirectional iterative denoising, naturally supporting user-specified constraints anchored at arbitrary output positions, a paradigm known as In-place Prompting (IPP). We formalize this as the In-place Instruction Following (IIF) task and construct IIF-Bench, a hierarchical benchmark spanning literal, style, and discourse-function constraints, paired with a rubric-based local-global evaluation protocol. An inference-time attention-bias probe suggests that vanilla dLLMs often under-prioritize constraint spans during denoising. We then propose GRAFT, an IPP-oriented post-training framework combining constraint-aware SFT and preference optimization. On four representative dLLMs, GRAFT raises the average IIF score from 57.75 to 73.10 (+15.35 points), with absolute gains of 15.91 and 15.57 points on literal and discourse-function constraints, while preserving general generation ability. --- ### Unsupervised Domain Adaptation for Symbol Spotting in Historical Encrypted Manuscripts Authors: Giuseppe De Gregorio, Alicia Fornés, Lei Kang, Beáta Megyesi Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07159v1 The decipherment of historical encrypted manuscripts poses a fundamental challenge in Digital Humanities: before any transcription can begin, the symbol inventory of the underlying cipher alphabet must first be identified and characterized. We address this challenge through symbol spotting: given a candidate alphabet specified as a set of rendered font glyphs, the task is to determine whether and where its characters appear in an unseen handwritten document, without any labeled examples from the target script. The main difficulty lies in the domain gap between clean, digitally rendered font queries and degraded handwritten manuscript symbols. We propose a three-stage pipeline that bridges this gap without manual annotation, combining a joint SimCLR+DANN encoder for domain-invariant glyph representations with an embedding-space style-adaptation mechanism applied at retrieval time, requiring no re-training. Experiments on fourteen pages from seven encrypted manuscript collections show that our method outperforms zero-shot foundation models, including CLIP and DINOv2, by a large margin ($+0.194$ P@1 over CLIP ViT-L/14), and surpasses task-specific trained baselines by $+0.138$ P@1. We further demonstrate that the Raw-Cover metric, computed in a fully unsupervised setting, provides a meaningful script-family fingerprint that identifies the underlying alphabet of an unknown document. This capability is of direct practical relevance to palaeographers, historians, and other researchers working with undeciphered manuscripts. --- ### Mind the Approximation: Fisher-Weighted SVD Compression for ViTs Authors: Moritz Thoma, Maximilian Groezinger, Maximilian Forstenhäusler, Emad Aghajanzadeh, Ryan Pegoud, Manoj Rohit Vemparala, Pierpaolo Mori, Alexander Frickenstein, Daniel Mueller-Gritschneder, Ulf Schlichtmann Published: 2026-09-07 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.07155v1 Model compression is key to mitigate deployment challenges of ever growing machine learning models. In this area of research, singular value decomposition (SVD)-based compression offers a compelling trade-off between computational efficiency and model accuracy. Fisher-weighted SVD in particular provides principled, loss-aware compression. However, we find that improving the fidelity of Fisher approximation used in the compression is poorly predictive of post-compression accuracy for Vision Transformers (ViTs). Motivated by this observation, we propose FACTS, a structured Fisher Approximation tailored to Compressing ViTs with Fisher-weighted SVD, which enforces token-local aggregation while preserving within-token activation-gradient dependence. Additionally, we introduce a fast Constrained Rank Search (CoRS), that optimizes layer-wise rank allocation while adhering to a fixed floating point operation (FLOP) constraint. Extensive experiments across ViTs and hybrid architectures demonstrate that FACTS consistently improves accuracy-efficiency trade-offs without requiring finetuning. Notably, it outperforms the strongest SVD baseline by up to +5.8 percentage points (p.p.) Top-1 on Swin-B, with further gains driven by our search method. Code is available at https://github.com/MoritzTho/FACTS. --- ### Ambient @ EgoLongQA 2026: Distilling Long-Video perception into a Sub-2B Model Authors: Logesh Kumar Umapathi Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07154v2 We describe our entry to the EgoLongQA track of the Wearable-AI Challenge in ECCV 2026, which placed first in the <=2B parameter division with 0.8279 on the held-out test set. Our system is a single 2B vision-language model that answers multiple-choice questions about ten-minute egocentric videos in one greedy forward pass; It is obtained by distilling the junior perception module of a tool-using agentic pipeline, not the agent itself into a small student, using teacher traces filtered to those that answered correctly. it reaches 89% of the accuracy of the large agentic pipeline using 1.1% of its parameters. This raises a 27.1% base model to 81.4% on our held-out questions. The 2B backbone has 2.2132B parameters and therefore over the divisional limit, to make the entry admissable we prune the multilingual embedding table from 248,320 to 143,469 rows, reaching 1.9985B with provably identical logits on retained rows. --- ### FreqBLiMP: Frequency-Controlled Minimal Pairs Reveal Robustness and Fragility of LLMs Under Lexical Rarity Authors: Tyrone White, Yuki Arase Published: 2026-09-07 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.07153v1 Minimal-pair benchmarks such as BLiMP evaluate linguistic knowledge by testing whether language models (LMs) prefer acceptable sentences over minimally different unacceptable ones. However, these benchmarks largely ignore lexical frequency variation, despite lexical frequency being a pervasive and highly skewed property of natural language use. Consequently, existing evaluations do not test whether grammatical preferences remain stable when contrasts involve rare lexical items. We introduce FreqBLiMP, a frequency-controlled extension of BLiMP that regenerates all 67 paradigms under explicit Zipf-frequency regimes while preserving each minimal-pair's grammatical contrast. Evaluating multiple open-weight LLM families across scales, we find that decreasing lexical frequency produces a consistent, monotonic decrease in sentence likelihood, but only a modest reduction in overall contrastive acceptability accuracy. However, this aggregate stability masks substantial variation across linguistic phenomena, with LLMs remaining robust on overt morphosyntactic generalization while degrading on phenomena that require lemma-specific information. --- ### An Auditable Symbolic-RAG-Generative AI Architecture for Goal-Oriented Conversation Orchestration Authors: Ramon Gonzalez, Antonio Diaz Published: 2026-09-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.07152v1 Goal-oriented conversational systems must answer factual questions, understand visitor-provided information, and advance business objectives without becoming rigid questionnaires. This paper proposes a Symbolic-RAG-Generative architecture centered on the Goal-oriented Retrieval-Augmented Conversation Engine (GRACE). An instruction-constrained Business Goal Compiler transforms business intent into an immutable objective set, normalized priority vector, canonical questions, and initial state vector. At runtime, GRACE receives the complete conversation history, latest visitor message, current state, and grounded answer generated by a separate RAG component. It updates completion only from visitor-authored evidence and selects one contextually modulated follow-up. The core policy maximizes expected business progress subject to a minimum visitor-utility constraint. We formalize the state, monotonic transitions, source separation, question modulation, and constrained policy; present the reference architecture; and define an evaluation comprising 24 English real-estate and 10 Spanish professional-cleaning conversations, totaling 119 protocol-defined visitor turns. Across both domains, GRACE achieves 84.9% exact state-transition accuracy, 91.6% evidence precision, 89.6% evidence recall, 100% monotonicity, and 94.1% terminal-state accuracy. The evaluation establishes compelling symbolic-state performance across standard, multi-goal, RAG-detour, validation, refusal, and robustness scenarios. --- ### Vishing-Tactics-Bench: Forecasting Exploitation Trajectories in Voice Phishing Calls Authors: Jeongmin Lee, Dongmyung Sul, Seung Yun, Jinxia Huang Published: 2026-09-07 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.07151v1 Voice phishing (vishing) unfolds in real time; by the time a call has ended and post-hoc classification is possible, the harm has already been done. The more actionable question is which concrete harm (Information Gathering or Financial Exploitation) an ongoing call is tactically progressing toward. We present Vishing-Tactics-Bench, a benchmark grounded in Endsley's situation-awareness (SA) framework that recasts vishing defense from after-the-fact fraud classification to harm projection: predicting at each turn whether the call will reach either terminal harm. We adapt MITRE ATT&CK to vishing as a 6-tactic taxonomy (Vishing-Tactics) and label 35,340 scammer utterances across 5,645 synthetic Chinese calls. We define Exploitation Trajectory Forecasting, a survival-style protocol over the two terminal harms with three metrics: AP@k, C-index, and divergence error. Baselines ranging from a Markov heuristic to fine-tuned LLMs show that the tactical trajectory serves as an interpretable representation of the call's tactical state, supporting harm-specific forecasting, which can then be used for the downstream application of intervention selection; a stratified lead-time analysis at a tight false-alarm budget further identifies at what point in a call the trajectory signal yields early warning. --- ### Stable-MM-R1: Anchoring Multimodal Reasoning Dynamics via Entropy-Guided Stratification Authors: Yimeng Ye, Shuang Chen, Wenxuan Huang, Manyuan Zhang, Kaituo Feng, Zhangquan Chen, Jiayu Chen, Yucheng Zhou, Yicheng Xiao, Zhiyuan Feng, Tianyu Shi Published: 2026-09-07 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2609.07148v1 While Reinforcement Learning (RL) effectively incentivizes reasoning in Large Language Models, current pipelines are hindered by training instability and rapid entropy collapse. These limitations often stem from "Rollout Silencing" and low-quality gradient signals in standard sampling procedures. In this work, we propose a robust, data-centric framework to stabilize RL training. We first introduce Potential-Aware Query Mining (PAQM), which filters data dynamically to focus on the "Distillation Zone"---samples with high potential for capability elicitation. Furthermore, we present Hybrid Stratified Replay (HSR), a novel mechanism that restructures batches by stratifying rollouts based on Path Entropy, a rollout-level confidence proxy, and outcome reward. Within each optimization step, HSR reuses current-policy "Stability Anchors" and "Hard Negatives" to construct high-contrast optimization groups, then clears its buffers before the next step. This approach mitigates entropy collapse while improving the utilization of learning signals under limited compute. Our method outperforms strong baselines on complex reasoning tasks, offering a principled solution for stable and efficient RL fine-tuning. --- ### Fine-grained Distributed Backdoor Attacks in Federated Learning Authors: Jian Wang, Hong Shen, Wei Ke, Xue Hua Liu Published: 2026-09-07 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.07147v1 Federated learning, as a privacy-preserving distributed machine learning paradigm, faces significant threats from backdoor attacks. Compared to centralized attacks, distributed backdoor attacks are more harmful but require more poisoned samples to compensate for the loss of trigger strength due to decomposition. Fixed trigger patterns are also easily detected by robust aggregation algorithms, increasing the risk of attack exposure. To address these challenges, we propose a fine-grained distributed backdoor attack framework (FDBA). This framework uses dynamic trigger generation and embedding vector optimization to perform attacks with fewer poisoned samples. First, we design a dynamic trigger generation method based on image edge structures using the Canny algorithm to extract edge features, which are then injected with Laplacian noise. RGB channel decomposition is applied for covert adaptation of the distributed trigger, reducing detection chances. Second, we introduce an embedding vector contrastive learning strategy that forces poisoned samples to approach the target class center in the feature space, enhancing attack effectiveness. On CIFAR-10, piecewise-linear estimates for target ASRs between 70\% and 90\% show that FDBA reduces the required poisoning ratio by 37.4\%--48.4\% compared with DBA. In non-independent and identically distributed (Non-IID) scenarios, FDBA retains 84.7\% of its IID attack performance under extreme heterogeneity, whereas DBA drops to 73.5\%, and the framework successfully bypasses mainstream defense mechanisms. This study offers new insights into federated learning security and emphasizes the potential threats and defense challenges posed by fine-grained distributed attacks. --- ### Encoded Early, Used Late: Where Transformers Begin to Act on an Inferred Partner's Expertise Authors: Mika Okamoto, Gabriele Sarti Published: 2026-09-07 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.07139v1 A transformer can make an attribute linearly decodable in its residual stream at a depth where that attribute does not yet influence the output. This gap between where information is readable and where it is used has been shown for attributes stated directly in the input. We ask whether it also holds for an attribute the model must infer gradually over a conversation, namely how expert its dialogue partner is. Using ExpertCollab, a corpus of multi-turn research-planning dialogues between model-played personas at four expertise levels, we find that partner expertise is most decodable in the early layers and falls to near chance before the midpoint of the network. Counterfactual patching shows that injecting the expertise difference at the layer of peak decodability barely changes a fixed late-layer readout, whereas the same difference injected past the midpoint propagates almost completely, a separation of more than an order of magnitude. A content-matched random control and a probe-free diagnostic place the transition at the same early layer, and a statically specified control attribute stays decodable throughout. An inferred relational attribute is therefore represented well before it becomes causally active, which bounds where any attempt to read out or steer partner-conditioned behavior must intervene. We use one model on a synthetic corpus as an initial demonstration. --- ### Flow3D-OPD: Multi-Teacher On-Policy Distillation for 3D Geometry Generation with Flow-Matching Diffusion Transformer Authors: Zhiwei Ning, Zhen Zhou, Puhua Jiang, Xintong Han, Gengming Zhang, Jie Yang, Zhonglong Zheng, Yuanjie Zheng, Wei Liu, Chunchao Guo Published: 2026-09-07 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.07137v1 Recent image-to-3D generation models built on flow-matching diffusion Transformers (DiT) can produce high-fidelity meshes, yet their post-training strategy remains largely unexplored. There exist several critical bottlenecks in reinforcement learning: the inherent difficulty of defining comprehensive rewards for 3D geometric quality, and the gradient interference that arises when jointly optimizing heterogeneous objectives. Inspired by the practicability of on-policy distillation (OPD) in large language models and image generation, we propose \textbf{Flow3D-OPD}, a two-stage post-training framework that introduces multi-teacher distillation into 3D geometry generation. In the first stage, we utilize the semi-policy to enhance the foundational capability of the pretrained model and then design an agentic verifier for 3D geometric quality evaluation. Based on the verifier, we could cultivate domain-specialized teacher models via direct preference optimization (DPO). In the second stage, we consolidate heterogeneous expertise into a unified student model through on-policy distillation with hard task-routing sampling and gradient accumulation, which could mitigate the gradient interference in joint optimization. Without relying on elaborate modifications, our straightforward yet effective design achieves consistent improvements across all geometric quality dimensions and surpasses all teacher models in the average metric. Extensive experiments demonstrate that our approach provides an effective paradigm for reinforcement learning in 3D generation. --- ### NutriBench-Kitchen: Benchmarking Embodied AI for Nutrition Management Authors: Yulin Wei, Xiangchen Wang, Jianhui Pan, Jinyu Xiao, Zheng Tan, Ruozai Tian, Guanhua Chen, Feng Zheng Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07135v1 An embodied kitchen assistant must do more than recognize food in isolated frames. It must track ingredient states over time and integrate visual observations with recipe and nutritional knowledge to support constraint-aware decision-making. We formalize this capability as \emph{Embodied Nutrition Management}: perceiving nutrition-relevant events, maintaining a persistent food state, and using it for knowledge-grounded planning. Existing benchmarks evaluate static food understanding or embodied cooking actions, but do not measure whether an agent can continuously update and use nutrition-relevant states in dynamic kitchens. To fill this gap, we introduce \textbf{NutriBench-Kitchen}, a benchmark containing 1,500 manually verified question--answer pairs from 160 cooking videos. It covers five task families: Ingredient Entry, Memory Management, Recipe Query, Long-Term Planning, and Short-Term Planning, spanning food-state construction, maintenance, knowledge retrieval, and decision-making across different planning horizons. Evaluations of proprietary and open-source large vision-language models reveal a substantial gap from human performance, particularly in quantitative ingredient estimation, long-term state tracking, and reasoning under interacting constraints. We further introduce \textbf{Nutri-Vgent}, a diagnostic long-video agent with separate episodic, food-state, and recipe memories. Its consistent improvements demonstrate the value of explicit state representations and structured memory for nutrition management. Together, NutriBench-Kitchen and Nutri-Vgent provide a testbed for studying persistent state tracking and knowledge-grounded reasoning in dynamic kitchens. Code is available at https://github.com/V1ol1n/NutriBench-Kitchen. --- ### Retrieval-Augmented Multi-Prompt Ensemble for Minor-Grain Breeding Information Extraction Authors: Hang Zhao, Jiahao Wang Published: 2026-09-07 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.07134v1 This paper presents our system for CCL2026-Eval Task 5: Minor-Grain Breeding Information Extraction (MGBIE), which jointly extracts 12 entity types and 6 relation types from minor-grain breeding literature. We propose RAME (Retrieval-Augmented Multi-Prompt Ensemble), a training-free framework that elicits multiple LLM outputs under controlled diversity and aggregates them by majority voting to obtain high-confidence predictions. RAME combines (i) retrieval-augmented few-shot selection via a hybrid BM25-embedding retriever, (ii) a three-prompt ensemble (Strict, Relaxed, Balanced) spanning the precision to recall spectrum, and (iii) large-scale repeated sampling with majority voting to filter noisy predictions. Built on DeepSeek-V4-Flash, RAME achieves a Total Score of 0.499 (NER 0.730, RE 0.346) on the leaderboard, ranking 1st and surpassing the official Track-A baseline powered by GPT-5.5 (0.448), representing an 11.4% relative improvement. Code is available at https://github.com/king-wang123/CCL26-RAME. --- ### AgentLeak: Cloning Stronger LLM Agent Capabilities onto Weaker Agents Beyond Skill Stealing Authors: Xiaoting Lyu, Yuhong Wu, Yufei Han, Shichang Liu, Liang Zhang, Bin Wang, Bin Wang, Xiaobo Ma, Wei Wang Published: 2026-09-07 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.07131v1 Large language model (LLM) agents increasingly achieve long-horizon tasks by combining foundation models with explicit skills and implicit procedural knowledge acquired through execution. The resulting task-solving capabilities have become valuable proprietary assets, raising a new security question: can a substantially weaker attacker-controlled agent acquire the capabilities of a stronger proprietary agent through limited black-box interaction? Existing skill-stealing attacks recover explicit skill artifacts, yet we show that artifact leakage does not necessarily transfer capability: a weaker agent may possess the same skills but still fail because it lacks procedural behaviors implicitly realized by the stronger agent. Our key insight is that the skill execution gap itself forms a leakage surface, where missing behaviors are exposed through observable differences between successful victim executions and failed attacker executions. Based on this, we present AgentLeak, a black-box capability-cloning attack that identifies capability-critical behaviors from these execution differences and incorporates them into attacker-side skills, while keeping the attacker's model, harness, and tools unchanged. Across 20 task scenarios comprising 600 instances, diverse agent systems, and multiple backbone models, AgentLeak improves task pass rates by over 40% compared with direct skill reuse and recovers more than 80% of the victim--attacker capability gap. Our findings reveal a confidentiality risk in LLM agents: protecting explicit artifacts alone is insufficient, as observable execution behavior can leak the procedural knowledge required to reconstruct proprietary task-solving capabilities in low-capability and attacker-controlled agents. --- ### Line-Coupled Language Model Authors: Shiyuan Li, Shaorong Zhang, Zhaorui Yang, Qian Zhang, Greg Ver Steeg, Bingyao Li Published: 2026-09-07 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.07129v1 Autoregressive language models generate one token per decoding step, limiting the useful output of each forward pass. Although diffusion models, insertion-based decoding, and multi-token prediction enable parallel generation, they either incur additional training-time token traffic or struggle to predict strongly dependent future tokens. We introduce the Line-Coupled Language Model (LCLM), an autoregressive model that advances multiple text lines together by predicting the next token for every active line while coupling the lines through shared causal context. LCLM interleaves line tokens into a single causal sequence and uses line-staggered rotary positions, retaining the standard next-token objective and causal attention. Controlled experiments show that cross-line targets are substantially less dependent than consecutive same-line targets, supporting lines as parallel generation units. With 881M parameters, LCLM produces an average of 2.94 content tokens per forward pass with a validation cross-entropy loss of 2.44, compared with 1.00 token per forward pass and a loss of 2.39 for the vanilla autoregressive baseline. Most notably, even when LCLM generates 16 tokens per forward pass, its loss is only 0.09 higher than that of the vanilla autoregressive baseline (2.34 vs. 2.25). --- ### EEG-Driven Decoding Framework for Passenger Hazard Perception in Highly Automated Vehicles Authors: Yingkai Yang, Ashton Yu Xuan Tan, Bowen Li, Xiaorong Gao, Sifa Zheng, Jianqiang Wang, Xinyu Gu, Yang Zhao, Yuxin Zhang, Sharon X. Huang, Tania Stathaki, Jun Li et al. Published: 2026-09-07 Categories: cs.AI, cs.LG, eess.SP arXiv: https://arxiv.org/abs/2609.07128v1 Reliable risk assessment remains a central challenge for Autonomous Vehicles (AVs). Despite advances in automation, passenger cognition provides a non-intrusive auxiliary signal that improves both objective and perceived safety without requiring active human intervention. We introduce an Electroencephalogram (EEG)-based Brain-Computer Interface (BCI) that decodes passenger neural responses for both Risk Prediction (RP) and Danger Identification (DI), explicitly modeling humans as passengers to match real-world AV use. To achieve this, we propose the Passenger Cognitive Model (PCM), Risk-aware Sequential Labeling (RSL), and the Passenger EEG Decoding Strategy (PEDS), which integrates a 3D Convolutional Recurrent Neural Network (3D-CRNN) model for joint EEG decoding. Experimental results show that 3D-CRNN achieves a Balanced Accuracy (BA) of $95.3\% \pm 2.7\%$ in RP and improves single-subject DI from $80.9\% \pm 3.9\%$ to $85.0\% \pm 3.2\%$ with RSL. Event-wise analyses further show that 3D-CRNN consistently outperforms other models across different event types in RP and DI. In generalization experiments, 3D-CRNN achieves $77.0\% \pm 5.3\%$ BA in cross-session DI and $77.4\% \pm 1.1\%$ BA on seen subjects in cross-subject evaluation, while maintaining a $64.9\% \pm 8.5\%$ BA on unseen subjects, demonstrating promising generalizability and transferability across both intra-subject and inter-subject variability. These findings establish an Electroencephalogram (EEG) decoding framework for AV passenger hazard perception and suggest that passenger cognitive signals can provide auxiliary supervision for future AV decision-making and Safety of the Intended Functionality (SOTIF) support. --- ### Beyond Task Success: Stage-Wise Reliability of World Model Planning under Sensing Degradation Authors: Geonmyeong Lee, Byoung-Tak Zhang Published: 2026-09-07 Categories: cs.RO, cs.LG arXiv: https://arxiv.org/abs/2609.07126v1 In world model planning, sensing inputs pass through an encoder and predictor before affecting planner decisions, so final task success alone cannot reveal where sensing disturbances attenuate or persist in the pipeline. We apply 10 visual and temporal sensing degradations to a world model planner and track their effects across representation, future prediction, planner preference, and physical outcome using paired evaluation on the same 50 tasks. The relative impact of degradations was not preserved across stages: large representation shifts could attenuate downstream, while smaller initial shifts could persist to the outcome, and internal-response ordering did not directly match physical-outcome ordering. Temporal degradations also showed distinct patterns: even with similar overall changes in observation history, responses differed substantially with the location of corrupted information and the planner's actual exposure. This non-uniform stage-wise response was also observed in secondary evaluations with another manipulation task and a different world model. Stage-wise diagnosis can therefore identify where sensing disturbances attenuate or persist and help prioritize subsequent model verification and sensing mitigation. --- ### A Two-Stage Framework for Ego-Centric Key Object Identification via Object State Prediction Authors: Shihong Ling, Yue Wan, Xiaowei Jia, Na Du Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07125v1 This paper presents a novel framework designed to enhance key object identification in autonomous driving. Existing methods primarily focus on either detecting objects independently or leveraging visual relationships, but they do not explicitly consider the ego vehicle's perspective in determining object importance. To address this gap, we propose a structured approach that integrates a virtual ego-vehicle representation and a modular object state predictor, enabling a more accurate estimation of object behaviors relative to the ego-vehicle. Subsequently, our framework employs spatial-temporal reasoning to refine key object identification, prioritizing objects based on their states and relative spatial information rather than relying solely on visual relationships. Experimental results on real-world driving datasets demonstrate the effectiveness of our approach in accurately detecting critical objects in complex traffic environments. --- ### PTCG: Persona-guided Tree-based Counterargument Generation Authors: Eunbeen Son, Yohan Jo, Joonsuk Park, JinYeong Bak Published: 2026-09-07 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.07120v1 The ability to generate counterarguments is important for critical thinking and balanced discourse, yet existing approaches typically produce only a single counterargument, failing to capture the diversity and persuasiveness required in real-world debates. To address this limitation, we propose Persona-guided Tree-based Counterargument Generation (PTCG), a framework that combines Tree-of-Thoughts-inspired step-wise generation and pruning with speaker persona selection. By estimating the author's persona from the original argument and incorporating speaker personas representing distinct perspectives, PTCG operationalizes perspective-taking and enables the generation of diverse counterarguments. Results from LLM-as-a-Judge, classifier-based assessment, and human evaluations indicate that PTCG shows consistent improvements in both the diversity and persuasiveness of counterarguments compared to baseline methods. --- ### The Illusion of Debiasing: Persona Steering Redistributes Rather Than Reduces Bias in LLMs Authors: Ziyue Feng, Hongbo Fang, James A. Evans Published: 2026-09-07 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.07117v1 Prompt-based interventions: system prompts, personas, role instructions, reliably reshape what a language model says, but it is unclear which layer they reach. Do they reconfigure internal structure, or only modulate the output channel? We use persona conditioning as a controlled probe, measuring its effects along a depth axis from self-report, through open-ended generation, to word-level parametric association, across three instruction-tuned models. We find a graded dissociation. Personas are legible but not structural: models follow single-trait instructions yet fail to reproduce human inter-trait covariance. The dissociation deepens with depth: personas hold or amplify closed-form QA bias, shift absolute tone while leaving between-group disparity unchanged, and barely perturb an already saturated associative baseline. Prompt-based steering thus operates in the output channel and has a structural reach limit that surface manipulability can mask. --- ### From LLM-Generated Specifications to Learned Quadruped Locomotion Authors: Merve Atasever, Keyan Azbijari, Cagan Bakirci, Alfredo Reina Corona, Tolga Izdas, Richard Yang, Erdem Biyik, Jyotirmoy V. Deshmukh Published: 2026-09-07 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.07111v1 Quadruped robot locomotion policies are often trained using reinforcement learning, which in turn relies heavily on hand-crafted reward functions. Designing reward functions requires substantial manual engineering, and it is often unclear which local rewards will induce the desired global behavior. Shaped rewards from formal specifications in languages like Signal Temporal Logic (STL) can make rewards more interpretable, but writing STL specifications itself still requires domain expertise. We study whether large language models (LLMs) can fill this gap by generating Parametric Signal Temporal Logic (PSTL) specifications that are subsequently used for policy learning. Given a natural language locomotion objective and a constrained specification grammar, GPT-5.5 and Qwen 3.6 independently propose STL templates for command tracking, safety, and gait structure. We instantiate the parameters of the generated PSTL templates using expert trajectories and retain only specifications that are consistent with demonstrated expert behavior. The resulting specifications are then transformed into smooth, finite-history reward functions and used to train a quadruped locomotion policy with Proximal Policy Optimization (PPO) in MuJoCo XLA (MJX). We evaluate both \emph{gait-aware} and \emph{gait-agnostic} settings. The former specifies walking-trot, trot, and bound regimes, while the latter allows contact patterns to emerge from the task objective. We compare against hand-engineered rewards, Text2Reward-style LLM-generated reward code, and an expert-switching oracle. Gait-aware Qwen 3.6 specifications achieved 100\% survival and command success across all tested speeds (0.3--2.1 m/s) and matched the target gait at high speeds, whereas Text2Reward achieved 0\% for both metrics at $\geq 1.9$ m/s. Videos: https://stl-locomotion.github.io/ --- ### Discovering Natural Transformation Vulnerabilities in Black-Box Vision Models Authors: Dongsu Song, DaeYun GO, Jay Hoon Jung Published: 2026-09-07 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.07110v1 Natural adversarial examples (NAEs) reveal that vision models can fail under realistic semantic changes beyond norm-bounded perturbations. However, generating NAEs in a black-box setting remains challenging because existing generative attacks often rely on surrogate models, learned attack priors, or costly query-based optimization, whereas the natural transformations that expose model vulnerabilities are unknown a priori. We propose \textbf{Adversarial Scenario Attack (ASA)}, a query-based black-box framework that searches over natural-language editing scenarios using a multimodal language model and a modern text-guided generative editor. ASA jointly explores background, weather, and material/color transformations through winner--loser feedback, and uses a greedy explorer to compose only attack-improving scenarios. Across diverse ImageNet classifiers, ASA achieves substantially higher attack success rates than prior query-based generative attacks while requiring fewer victim-model queries and preserving competitive perceptual quality. Moreover, ASA exhibits both image-level and prompt-level transferability: its adversarial images remain effective across victim-model architectures, while its discovered editing scenarios can be reused across same-class images and, in some cases, across architectures. These findings suggest that vision models possess reusable vulnerabilities to natural transformation patterns, which ASA can efficiently identify in a black-box setting. --- ### Online Draft Co-Training for Speculative Decoding in Large-Scale, Long-Context RL Post-Training Authors: Zili Wang, Zhaopeng Qiu, Yuekai Zhang, Shuang Yu, Junjie Lai Published: 2026-09-07 Categories: cs.LG, cs.DC arXiv: https://arxiv.org/abs/2609.07108v1 Speculative decoding accelerates rollout generation, which dominates the cost of reinforcement learning (RL) post-training. Online co-training can further increase the draft's accuracy, yielding greater speedups. However, scaling this approach to co-training on large models with long contexts poses two obstacles: (1) branch attention is unsupported by standard causal context-parallel (CP) implementations, and (2) target features span across pipeline-parallel (PP) stages. We address both with an end-to-end system for large-scale online draft co-training. For CP, we extend packed, load-balanced zigzag ring attention by merging rank-local branch attention with causal main-sequence attention. For PP, TapChannel transports intermediate target features across stages via a separate path, leaving the pipeline schedule unaffected. Experiments demonstrate that co-trained drafts closely track the policy baseline while delivering substantial rollout and end-to-end speedups across model scales up to 122B. Our CP design achieves strong scaling at 256K tokens with significant memory savings over prior work, and our PP transport incurs modest overhead. Code can be found at https://github.com/NVIDIA-NeMo/RL/issues/3698. --- ### Beyond Sparse Rewards: A New Benchmark and Structure-Aware Graph Alignment for Micro-Drama Understanding Authors: Yixin Qin, Shi-Zhe Chen, Zhiqi Yu, Siyuan Cheng, Tao Cheng, Jinwen Luo, Zheng Wei Published: 2026-09-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.07107v1 Micro-dramas, characterized by ultra-short durations and hyper-dense storylines, pose unique challenges for video understanding that conventional benchmarks fail to address. To bridge this gap, we introduce M-Drama, the first large-scale bilingual benchmark for micro-drama comprehension, featuring over 35K instances across 9,138 clips. Furthermore, while reinforcement learning can enhance VLMs on complex narratives, existing reward metrics often suffer from sparse and superficial signals, failing to capture intricate character identities and temporal structures. We propose SAGA (Structure-Aware Graph Alignment), a novel graph-matching reward function that models narratives as heterogeneous graphs. SAGA computes dense, rigorous rewards via decoupled semantic triplet and structural temporal matching. Extensive experiments on Qwen3-VL-8B-Instruct demonstrate that SAGA outperforms existing baselines, delivering substantial improvements in open-ended accuracy and summary quality, while maintaining competitive out-of-domain generalization. Code is available at https://github.com/qyx1121/MDrama_SAGA. --- ### Revisiting Complete Reasoning Traces for Post-Training Authors: Jaehui Hwang, Sangdoo Yun, Byeongho Heo, Dongyoon Han Published: 2026-09-07 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.07103v1 Large language models (LLMs) are often post-trained on pre-collected reasoning trajectories to improve their reasoning capability. Such trajectories tend to be long due to complex, interwoven paths, which often include detours on the path toward the answer. However, it has been underexplored whether LLMs indeed benefit from learning complete trajectories in post-training, such as supervised fine-tuning (SFT). Starting from our pilot study, we find that full trajectories provide only limited benefit, while partial trajectories are effective even under heavy truncation. We analyze redundancy in reasoning trajectories through attention-based analyses and controlled token-removal studies, both of which show that intermediate tokens contribute minimally to final reasoning quality. This suggests that avoiding redundant information may allow LLMs to internally infer coherent alternatives by inferring missing steps from their internal knowledge, given known trajectory endpoints. Furthermore, we show that training LLMs using endpoints leads to consistent changes in reasoning behavior, and that it also benefits post-training methods based on reinforcement learning or on-policy distillation, highlighting the need to revisit complete reasoning traces. Code is available at https://github.com/naver-ai/revisiting-trace. --- ### AstroSpecLM: A Spectrum-Language Model for Evidence-Grounded Astronomical Spectral Analysis Authors: Jinghang Shi, Yanxia Zhang, Ali Luo, Changhua Li, Xiao Kong Published: 2026-09-07 Categories: astro-ph.IM, cs.AI arXiv: https://arxiv.org/abs/2609.07102v1 Astronomical spectra encode rich physical information, but drawing scientific conclusions from spectral features typically requires expert interpretation. This paper presents AstroSpecLM, a spectrum-language model that connects one-dimensional DESI spectra with Qwen3-4B to answer questions and provide explanations grounded in spectral evidence. Instead of generating question-answer pairs directly from templates or raw catalog fields, we first distill each spectrum into a compact set of catalog- and spectrum-derived facts, then use these facts as references to generate instruction-following conversations. The resulting model is competitive with specialist supervised baselines on classification and redshift estimation, while additionally producing natural-language explanations that reference specific spectral features. Our results indicate that grounding a language model in one-dimensional scientific spectra is feasible, and that fact-mediated instruction data yields a model capable of both prediction and explanation. --- ### Temporal Heterogeneous Graph Transformer for Credit Card Fraud Detection Authors: Qinwen Yan Published: 2026-09-07 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.07100v1 Credit card fraud detection typically relies on tabular features, while repeated attributes can also provide useful relational signals. This paper proposes THGT-FD, a Temporal Heterogeneous Graph Transformer for Fraud Detection. Each transaction is represented using one transaction token and six types of relation tokens and incorporates Time2Vec encoding into the transaction representation. A Transformer learns the interactions among these tokens within each individual transaction and then outputs a fraud probability. Experiments were conducted on 150,000 transactions sampled from the IEEE-CIS Fraud Detection dataset and chronologically partitioned according to TransactionDT. On the test set, THGT-FD achieved an AUC-ROC of 0.8536, an average precision of 0.4164, and a Recall@5% of 0.4708. The class-weighted histogram-based gradient-boosting baseline achieved an AUC-ROC of 0.8722. The results indicate that relation tokens provide useful information for fraud-risk ranking, although the current model does not yet incorporate entity-level historical aggregation. --- ### Ambient @ EgoProactive 2026 : Proactive Egocentric Assistance with Visually Grounded Supervision Authors: Logesh Kumar Umapathi Published: 2026-09-07 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.07099v2 We present our submission to the EgoProactive track of the ECCV 2026 Wearable AI Challenge, which ranked first in the large-model division and second in the <=2B division. The task requires a wearable assistant to decide after each eight-second segment of egocentric video whether to intervene or remain silent. Our approach has two main components. First, we reformulate intervention timing as single-token classification. Rather than generating either $interrupt$<utterance> or $silent$, the model predicts yes or no, and we derive the decision from the renormalised probabilities of these two tokens. This formulation improved macro-F1 by 0.249 and G-mean by 0.30 over free-form generation. Second, because labelled data were limited to the released validation set, we generated additional supervision using a tool-calling video agent that inspects each clip and assigns intervention timestamps. A narration-only alternative was four times larger and ten times cheaper, but transferred worse than supervision from an unrelated real corpus, suggesting that visual grounding is more important than annotation volume for this task. --- ### Risk Is Not Review Value: Wrong-Answer Exposure Under Bounded Review Budgets Authors: SangJin Park, Myungsub Choi, Jineok Kim, Minseung Kang Published: 2026-09-07 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.07095v1 LLM assistants often produce more answers than humans can review before users see them. Most evaluations ask whether an answer is wrong, unsupported, or low-confidence. Bounded review budgets instead ask which answers should be checked first under a fixed review budget. Risk alone is not enough: a high-risk answer may be hard to repair, while a moderately risky answer may be directly correctable from available evidence. For generated-answer evaluation, we model review prioritization as exposure reduction, where review value combines estimated wrongness, intervention affordance, impact, and cost. We evaluate review queues with Wrong-Answer Exposure Ratio (WAER), the fraction of wrong answers left unreviewed, and post-repair residual exposure (PRRE), the fraction still exposed after deterministic benchmark-supported repairs. PRRE uses repairability rules that do not numerically reuse the affordance scores used for ranking. On a 720-item TAT-QA/SciFact stress benchmark, review-value ranking keeps answer-level WAER nearly unchanged at 20% budget (0.605 vs. 0.600) but lowers PRRE from 0.881 to 0.716. These results show that trustworthy LLM evaluation should measure not only error detection, but also how limited review capacity reduces exposed wrong answers. --- ### CASCADE: A Spatio-Temporal-Causal Reasoning Representation and Dataset for Driving Authors: Jenny Schmalfuss, Despoina Paschalidou, Simon Gerstenecker, German Ros, Jose M. Alvarez Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07094v1 Reasoning is a promising route to the generalization that autonomous driving requires in the long tail, as it can infer how the elements of a scene depend on one another and traverse those dependencies to conclusions beyond what is observed. Yet it is hard to tell whether a model's conclusions follow the scene's dependencies, because no driving representation makes them explicit enough to test against. Text-based reasoning traces lack spatio-temporal grounding, spatio-temporal scene graphs lack causal links, and reasoning annotations at scale are increasingly model-generated and hard to verify. To this end, we introduce CASCADE (Causal Spatio-Temporal Analysis of Driving Environments), which encompasses two components: (1) a structured scene representation for reasoning in driving scenes and (2) a human-annotated dataset built on it. For every actor that interacts with the ego vehicle, the CASCADE representation records frame-by-frame, for as long as the actor is visible, what action is taken, where it occurs, and how it depends on the actions and states of others. The resulting structure makes reasoning predictions machine-verifiable: they can be scored against it element by element, without relying on (M)LLM judges. The CASCADE dataset provides comprehensive human annotations for 2,066 driving clips of the PhysicalAI dataset, with over 34K elements that establish the spatio-temporal and causal context of each scene, including 8.6K time-stamped ego and agent actions, 3.7K causal links and 2.9K potential influences, and 6.1K annotations for agents, objects, traffic lights, and environments. Being entirely human-annotated, CASCADE provides the reference for this comparison: benchmarking the reasoning abilities of Physical AI models, and verifying the quality of automatically generated reasoning labels. The CASCADE dataset is available at https://huggingface.co/datasets/nvidia/cascade. --- ### Where to Look and What to Use: Retrieve-Localize-Generate for Long-Term Conversational Memory Question Answering Authors: Yifan Wang, Xinkui Lin, Yongxiu Xu, Shen Gao, Ruochen Yang, Kun Huang, Yubin Wang, Jie Wu, Wei Liu, Jian Luan, Hongbo Xu, Shuo Shang Published: 2026-09-07 Categories: cs.CL, cs.IR arXiv: https://arxiv.org/abs/2609.07093v2 Retrieval-augmented generation (RAG) enables large language models (LLMs) to answer questions by accessing external knowledge and has been widely adopted for long-term conversational memory question answering. However, existing methods suffer from two key challenges: (1) fragmented evidence scattered across temporally distant sessions, and (2) noisy content within retrieved sessions that triggers the lost-in-the-middle effect. To address these challenges, we propose MemLoc, a unified Retrieve-Localize-Generate framework for long-term conversational memory QA. For retrieval, MemLoc decomposes each session into multi-granularity memory units and performs query routing via an inner-memory graph with entropy-based granularity selection. It further models cross-session semantic and temporal dependencies through a cross-memory graph, enabling coarse-to-fine retrieval of top-K relevant memory candidates. For localization, we introduce a reasoning-based evidence locator trained with Self-reflective Hint Policy Optimization (SHPO), which performs progressive refinement by extracting query-relevant fragments within memory units to suppress noise and reranking across candidates to remove redundancy, producing a compact evidence set with lightweight location IDs. For generation, these IDs act as precise grounding signals that guide the LLM to the correct memory positions, mitigating the lost-in-the-middle effect while preserving original contextual integrity. Extensive experiments on four benchmarks demonstrate that MemLoc achieves state-of-the-art retrieval accuracy and response quality while maintaining efficiency. Our code is available at: https://github.com/Nikol-coder/MemLoc. --- ### Conditioned Initialization for Attention Authors: Hemanth Saratchandran, Simon Lucey Published: 2026-09-07 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2609.07086v1 Transformers are a dominant architecture in modern machine learning, powering applications across vision, language, and beyond. At the core of their success lies the attention layer, where the query, key, and value matrices determine how token dependencies are captured. While considerable work has focused on scaling and optimizing Transformers, comparatively little attention has been paid to how the weights of the queries, keys and values are initialized. Common practice relies on random initialization or alternatives such as mimetic initialization, which imitates weight patterns from converged models, and weight selection, which transfers weights from a teacher model. In this paper, we argue that initialization can introduce an optimization bias that fundamentally shapes training dynamics. We propose conditioned initialization, a principled scheme that initializes attention weights to improve the spectral properties of the attention layer. Theoretically, we show that conditioned initialization can potentially reduce the condition number of the attention Jacobian, leading to more stable optimization. Empirically, it accelerates convergence and improves generalization across diverse applications, highlighting conditioning as a critical yet underexplored area for advancing Transformer performance. Importantly, conditioned initialization is simple to apply and integrates seamlessly into a wide range of Transformer architectures. --- ### Single Image to Textured 3D Object Generation in Frequency Domain: From Theory to Pipeline Authors: Qisen Wang, Yifan Zhao, Jia Li Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07085v1 Single-view 3D reconstruction, also known as image-to-3D, is a persistently challenging task due to the extreme lack of information. Recently, diffusion models pre-trained on large-scale datasets served as 2D priors are used to solve the ill-posed task but suffer from color deviation and view inconsistency, which can be curbed by using diffusion models fine-tuned with 3D annotated data served as 3D priors. However, 3D priors lack high-frequency details, which cannot be solved by direct complementation with 2D priors in spatial domain for introducing erroneous low-frequency 2D prior guidance. In this paper, we revisit the characteristics of different diffusion priors from the frequency perspective. Based on our observations, we theoretically present a unified framework of hybrid optimization using multiple diffusion priors in frequency domain. Under this framework, we further propose Morpheus3D, a pipeline of 3D object generation from any single unposed image in the wild. Morpheus3D enhances 3D prior with high-pass image-prompt 2D prior guidance to reconstruct high-quality 3D objects while effectively suppressing view inconsistency, low-frequency color deviation, and high-frequency lacking problems. Both quantitative and qualitative experiments on the public and our collected datasets with complex textures show that our method exhibits significant improvements in generation quality. --- ### SupGRPO: Enhancing GRPO with Matching-based Online SFT for Text Spotting Authors: Xudong Xie, Yuzhe Li, Jing Shi, Zhifei Zhang, Curtis Wigington, Zhaowen Wang Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07081v1 Text spotting requires both accurate text recognition and precise spatial localization. Current specialised spotters excel at predicting tight bounding boxes in natural scenes, but falter on complex or artistic text, whereas multimodal large language models (MLLMs) possess strong recognition capabilities yet remain weak at localisation. To equip the text spotter with general and powerful recognition capabilities and to maximize its localization ability, we explore two MLLM-based fine-tuning methods: Supervised Fine-Tuning (SFT) and reinforcement learning fine-tuning based on Group Relative Policy Optimisation (GRPO). An interesting finding is that SFT is less effective than GRPO at enhancing recognition, while GRPO is less effective than SFT at enhancing detection. To compensate for each other's shortcomings, we introduce a joint training strategy, SupGRPO, which simultaneously optimizes the model using both SFT and GRPO. SupGRPO employs the specially designed reward functions and develops a matching-based online SFT applied solely to coordinate tokens. It both mitigates the reward sparsity problem of GRPO and avoids the instance order dependency problem of SFT. To evaluate particularly challenging cases, we curate ATS, a dataset for artistic text spotting. Experiments demonstrate that SupGRPO improves both text recognition and detection, and attains superior performance. Our code and dataset will be released at https://github.com/Psycho-9/SupGRPO. --- ### Comparing Self-Supervised and Domain-Invariant Features for Cross-Domain Voice Phishing Detection Authors: Jeongmin Lee, Seung Yun, Minkyu Lee, Ran Han, Yoonkyu Woo, Jinxia Huang Published: 2026-09-07 Categories: cs.SD, cs.CL arXiv: https://arxiv.org/abs/2609.07079v1 Voice phishing detection faces three critical challenges: real criminal recordings are unavailable due to privacy constraints; when available, only a handful of samples exist, insufficient for fine-tuning; and lightweight acoustic-only detection is needed as an alternative to large self-supervised models. We compare domain-invariant prosodic features and self-supervised representations (HuBERT, wav2vec2.0) through cross-domain evaluation-training on scenario-based actor recordings and testing on authentic criminal calls. Domain-invariant prosodic features achieve 69.5% F1 zero-shot and 71.0% with 5-shot learning. HuBERT achieves highest performance (94.2% F1, 5-shot), while wav2vec2.0 exhibits a precision-oriented detection profile (90.2% F1 with 99.4% precision, 5-shot). These findings reveal fundamental trade-offs: domain-invariant features enable zero-shot deployment when no real data exists, while SSL methods achieve higher performance but require real samples and compute. --- ### One for All: Generalist Foundation Model for Cross-Sensor Skeleton Representation Learning Authors: Jeonghyeok Do, Yun Chen, Munchurl Kim Published: 2026-09-07 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.07078v1 For learning generalizable motion representations from large-scale unlabeled data, Self-supervised learning (SSL) has become a widely adopted methodology. However, existing approaches are primarily limited by the inherent heterogeneity of skeleton data---characterized by varying joint counts, indexing protocols, and topological structures across different sensors---which typically necessitates training separate, sensor-specific, or even entirely dataset-specific models. To overcome this, we introduce SOfA (Skeleton One for All), the first generalist foundation model designed to achieve sensor-unified skeleton representation learning across diverse sensors. To accommodate the dimensional gap caused by varying joint counts, we introduce a fixed-size set of learnable Canonical Joint Slots, acting as a universal vessel that seamlessly accommodates arbitrary skeletal topologies. SOfA fills these slots via an attention mechanism that dynamically aggregates skeletal information from sensor-specific inputs. Furthermore, we resolve joint index misalignment between various sensors by introducing a Semantic Joint Embedding derived from a pre-trained text encoder, rather than relying on absolute positional embeddings. To validate our approach, we standardized ten 3D skeleton datasets for unified training. Extensive experiments demonstrate that SOfA can serve as a truly universal encoder, achieving state-of-the-art (SOTA) performance across a wide range of downstream tasks and sensor types, often outperforming dataset-specific specialist models with a single foundation model. --- ### A Hierarchical Consistency Framework for Auditing Retrieval-Augmented Generation Systems Authors: Ramon Gonzalez, Antonio Diaz Published: 2026-09-07 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.07075v1 Retrieval-augmented generation (RAG) is commonly evaluated by whether the final answer is correct. That test is insufficient: an answer can match its reference while the context that produced it contains a direct contradiction, leaving the contested evidence invisible to answer-only review and retrieval relevance scores. This paper presents the Hierarchical Consistency Framework (HCF), a post-hoc, model-agnostic audit of three distinct levels of a RAG process: the knowledge corpus, the final retrieved context, and the generated answer. HCF represents corpus conflicts as source-linked atomic facts, thereby identifying the documents responsible, and returns each Answer Consistency Score (ACS) with an explanation of supporting and contradictory contextual statements. We evaluate HCF on several controlled corpora spanning five domains and 100 query-corpus instances. A human evaluator compares every generated response with its supplied ground-truth response. The results show that the three diagnostic levels can dissociate: the corpus with the highest mean retrieval similarity has the lowest mean ACS, while a structurally degraded corpus performs worse at corpus level but better at answer level. Most importantly, HCF identifies contradictory retrieved evidence in several cases where the answer still matches the ground truth. HCF does not certify factual truth; it makes the evidence supporting and challenging an answer inspectable and attributable. ---