# Research Papers — 2026-09 · bundle page 101 of 266 > Full-text markdown of papers #5001–5050 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/101 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/100.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/102.md) ### Multi-View Fusion for Encrypted C2 Detection: A Leakage-Controlled Measurement Study of Evaluation Pitfalls Authors: Hoang-Huy Nguyen-Huu, Van-Tri Phan, Khuong Nguyen-An Published: 2026-09-23 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2609.27311v1 Command-and-control (C2) traffic increasingly hides within TLS, so defenders now apply machine learning to traffic metadata. Many studies assume that combining two metadata views, namely flow statistics and TLS handshake fingerprints, improves both accuracy and robustness. We tested this assumption on 17,577 TLS flows from 62 real Cobalt Strike captures. Our evaluation removes the data leakage that leads to overly optimistic reported scores. We report three findings that matter more than the fusion result itself. First, an incorrect preprocessing step increases the F1 score by 0.28. This step computes the frequency encoding across the entire dataset rather than within each cross-validation fold. The increase is about ten times larger than any real effect we measured. Second, both the labels and the behavioral features depend on the destination address. Because of this, the 17,577 flows form only 2,132 independent groups, and the positive rate of 55.1\%, which looks balanced, drops to 4.2\%. Therefore, class balance is just a result of how we analyze the data, specifically whether we count flows or endpoints, and not a real feature of the task. Third, 20 of the 62 captures (32\%) have no TLS flows to any known C2 address, so they contain only benign samples. We checked these captures directly and confirmed that this is a gap in the ground truth, not a labeling error. In this context, fusion beats the best single view by only 0.022 in F1. When an attacker forges both feature surfaces simultaneously, every model performs worse than a simple baseline that always predicts positive (F1 = 0.711). For encrypted C2 detection, the evaluation design is not a preliminary step. It \emph{is} the main result. --- ### FairTest: Search-Based Fairness Testing for Multi-Agent Reinforcement Learning Systems Authors: Xiaotong Wang, Xuan Xie Published: 2026-09-23 Categories: cs.SE, cs.LG arXiv: https://arxiv.org/abs/2609.27309v1 Multi-agent Reinforcement Learning (MARL) trains a team of agents that share one environment and learn their policies together. Training maximizes the team return, and a high return does not imply that the rewards are shared fairly among the agents in every episode. Testing is an established way to discover the failures of deep reinforcement learning, yet few methods address the fairness of MARL. In this work, we propose FairTest, a search-based testing approach that seeks the unfair executions of a MARL policy. The design combines search guidance with test prioritization. The guidance scores each candidate with three fitness functions. One measures the fairness of the runs already performed, another predicts the fairness from abstract states and fairness features, and the third reads the decision uncertainty from the policy. Crossover and mutation derive further candidates from the observed executions. The prioritization ranks the candidates by the predicted fairness and the decision uncertainty, so that the runs reach the candidates where failures are expected. FairTest is evaluated on three environments and two MARL algorithms, and four baselines are given the same budget. It detects the most fairness failures compared to three baselines with statistical significance and large effect sizes. The failure count exceeds that of the strongest baseline by 221% on average and coverage improves by an average of 23%. --- ### Learn How to Act from Your Own Interactions: On-Policy Self-Distillation for GUI Agents Authors: Yan Zhang, Daiqing Wu, Huawen Shen, Liang Li, Gang Cao, Zhi Gong, Wei Dai, Xiaode Zhang, Can Ma, Yu Zhou Published: 2026-09-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.27307v1 Graphical User Interface (GUI) agents enable the fulfillment of complex user instructions through multi-turn interactions with software environments, requiring step-wise reasoning and long-horizon memory to guide actions and retain task-relevant information, respectively. Recent on-policy self-distillation (OPSD) methods have achieved strong performance on GUI grounding, a foundational subtask for GUI agents, owing to dense token-level supervision from privilege-conditioned self-teachers. However, extending existing OPSD methods to multi-turn GUI agents is hindered by self-teachers' limited privilege-following ability and insufficient privileged guidance. In this paper, we introduce GUI-SD-v2, the next version of GUI-SD, which extends OPSD from GUI grounding to multi-turn GUI interaction and addresses key limitations through a two-stage training framework. Specifically, GUI-SD-v2 first strengthens privilege following by jointly optimizing rollouts with and without privileged guidance from the same GUI states. Furthermore, it selectively distills step-specific reasoning and memory guidance through a privilege-conditioned self-teacher, supporting action decisions and the retention of task-relevant information for subsequent interactions. Extensive experiments on two representative GUI agent benchmarks, AndroidWorld and MobileWorld, show that GUI-SD-v2 compares favorably with existing OPSD baselines while consistently outperforming the evaluated state-of-the-art methods in both Pass@1 and Pass@3 success rates. Code and training data will be publicly released. --- ### Discrete Diffusion Models via Evolving Variational Autoregressive Networks Authors: Kewen Pan, Ying Tang Published: 2026-09-23 Categories: cs.LG, cond-mat.dis-nn, stat.ML arXiv: https://arxiv.org/abs/2609.27306v1 Conventional score-based diffusion models learn scores without representing normalized densities, whereas tractable normalized models support both sampling and direct likelihood evaluation. A recent tensor-network approach provides such a representation but is largely restricted to low-dimensional lattices. Here we introduce a discrete diffusion model that parameterizes normalized probability distributions using variational autoregressive networks. Explicit Markov jump operators govern the forward noising and reverse denoising dynamics, extending discrete diffusion models with normalized distributions to spin systems on higher-dimensional lattices. We apply this framework to the two- and three-dimensional Ising models across ordered, critical, and disordered regimes, accurately computing thermodynamic quantities including free energy, energy, and magnetization. We further integrate the framework with Monte Carlo sampling, using adaptive diffusion steps to maintain high acceptance rates even at low temperatures while enhancing sample diversity. These results establish a neural-network framework for the discrete diffusion model with normalized probability distributions. --- ### Live Assistant: Learning Whether, When, and Whom to Assist in Real-World Live Social Streams Authors: Shujian Gao, Jiamei Yan, Yuchen Yang, Penghao Zhou, Qinglei Wang, Tiehan Fan, Yuan Wang, Zuxuan Wu, Yu-gang Jiang Published: 2026-09-23 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.27303v1 Livestreams are long-lasting interactive environments where audiovisual content, viewer activity, host behavior, and platform signals evolve together, creating assistance needs that emerge from the stream itself. We introduce \liveassistant, a framework for mixed-initiative, role-conditioned assistance that formulates livestream interaction as four coupled decisions: \textbf{whether to act, when to act, whom to address, and what to communicate}. At each 10-second interval, one autoregressive policy consumes native audio and video with synchronized comments, gifts, viewer dynamics, and room metadata, then selects \textsc{OBS}, \textsc{MEM}, or \textsc{ANS}. \textsc{OBS} remains silent, \textsc{MEM} records a private semantic update, and \textsc{ANS} specifies a recipient, task, and grounded message. To support this task, we build a trajectory engine that reconstructs real livestream sessions into structured causal supervision, yielding over 320 hours of optimization trajectories and a human-reviewed benchmark of 275 clips and 13,812 decision intervals. We train the policy with Marker-Aware Multiturn Supervised Fine-Tuning (MA-MSFT), which strengthens sparse structured decisions, followed by Streaming Multiturn GSPO (SM-GSPO), which optimizes self-generated trajectories with turn- and trajectory-level credit. On the held-out benchmark, \liveassistant reaches 71.14 state accuracy, 72.67 recipient accuracy, and 58.41 task accuracy, with consistent gains over representative streaming and general multimodal baselines. Together, the formulation, benchmark, and training framework establish livestream assistance as selective participation in a shared social stream. --- ### Beyond the Illusion of Power: Calibrating Quasi-Experiments in Observational IS Authors: Spandan Ghose Chowdhury Published: 2026-09-23 Categories: stat.ME, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.27299v1 Information systems (IS) researchers increasingly use quasi-experimental methods such as difference-in-differences (DiD) and instrumental variables (IV) to recover causal effects from observational panel data. Power calculations that justify these designs assume i.i.d. errors, but the deeper problem is what even a cluster-robust calculator cannot see. We report a Monte Carlo study over 9837 parameter conditions (approx 9.8 million datasets) and decompose the planned-versus-achieved power gap. The serial-correlation component is recoverable by an AR(1)-aware calculator when rho is known, and partially when rho must be estimated from short pre-periods, but panel attrition, staggered-adoption bias, and parallel-trends pretesting are captured by no closed-form formula; exogenous attrition alone costs approx 8 to 11 percentage points at the few-hundred-to-thousand sample sizes IS studies use. Treatment-correlated, outcome-dependent attrition instead induces bias, not just power loss. For IV, holding first-stage F fixed, larger N neither raises power nor curbs exclusion bias, though with a fixed instrument more data does sharpen the first stage, so identification rests on instrument strength, not sample size. --- ### StateComp: Learning When to Compress History in Long Horizon Agents Authors: Mingxuan Wang, Hongyue Chen, Yinglong Guo, Fei Luo, Chao Ning, Bo Wang, Guorun Yao, Yanbiao Ma, Jungong Han Published: 2026-09-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.27298v1 Long-horizon agents continuously accumulate interaction history during task execution, yet the importance of past interactions changes as the agent state evolves. Existing context management methods largely compress history based on fixed windows, periodic schedules, or current relevance, overlooking a more fundamental question: when has a past interaction become safe to replace? Premature compression may remove information still needed for future actions, while overly conservative retention leads to substantial context overhead. To address this, we propose State Conditioned Compression (StateComp), a framework that determines when historical interactions can be safely compressed according to the current agent state. StateComp constructs KEEP and READY supervision through a two-stage annotation procedure and trains an imbalance-aware router on hidden representations from a frozen language model. A bounded state representation further reduces the cost of evaluating long histories, while adjacent READY interactions are grouped into continuous spans and replaced with compact summaries during execution. Experiments on WorkBuddyBench show that StateComp reduces total agent and summarization tokens by 52.27% while maintaining task performance, and achieves a 12.67-fold speedup in representation extraction. --- ### Large Knowledge Model: From Papers to a Scientific Reasoning Landscape Authors: Yuan Huang, Sihan Hu, Hongyu Gu, Chao Ma, Jiaxing Zhang, Zhiyong Zou, Caiyu Fan, Yan Xiao, Mingjun Xu, Chenyu Xie, Mingzhen Ju, Zhehao Ma et al. Published: 2026-09-23 Categories: cs.AI, cs.CL, cs.IR arXiv: https://arxiv.org/abs/2609.27297v1 Accumulated scientific knowledge advances inquiry when prior findings help researchers choose new questions, design investigations, and interpret results. Realizing this value at scale requires access to the reasoning that connects research problems, scientific procedures, conclusions, and evidence. We introduce the Large Knowledge Model (LKM), a scientific knowledge infrastructure that transforms the literature into a shared, computationally accessible reasoning resource. LKM represents papers as source-grounded reasoning graphs, couples structural traversal with semantic retrieval over the same objects, and aligns related questions, claims, and reasoning chains across papers. This representation forms a Scientific Reasoning Landscape with three connected views: a Question Landscape that organizes research problems and open directions, a Workflow Landscape that exposes reusable scientific procedures, and an Evidence Landscape that connects conclusions to their support, disagreement, and conditions. The unified substrate supports reasoning-aware scientific search, evidence-grounded question answering, comparative evidence analysis, and research planning. Researchers and agents can retrieve relevant work through its scientific intent, synthesize answers with inspectable supporting arguments, and develop research plans informed by established workflows and unresolved evidence. We describe a corpus-scale system and evaluate scientific retrieval and knowledge-intensive question answering. With the answering model fixed, LKM retrieval improves accuracy by 9.30%, 4.20%, and 14.69% on ChemBench, PubMedQA, and SciBench, respectively. By connecting knowledge access to scientific reasoning and action, LKM provides a common foundation for discovering relevant research, reusing scientific knowledge, and coordinating cumulative inquiry across researchers, agents, and research cycles. --- ### Teach-to-Crash: A Closed-Loop Student-Teacher LLM Framework for Collision-Inducing Test Scenario Generation Authors: Zaid Ghazal, Khouloud Gaaloul, Bruce Maxim Published: 2026-09-23 Categories: cs.SE, cs.AI, cs.MA, eess.SY arXiv: https://arxiv.org/abs/2609.27296v1 Validating Autonomous Driving Systems (ADS) in simulation requires testing architectures that can discover rare, safety-critical failures while generating scenarios that are executable, diverse, and useful for downstream failure analysis. We introduce Teach-to-Crash, a closed-loop testing framework that combines a constrained ego-centric scenario representation, stagnation-aware search control, and a dual-LLM architecture for adaptive failure discovery. A high-reasoning Teacher LLM acts as an adaptive search controller, while a low-reasoning Student LLM emits simulator-executable scenarios in a strict JSON schema. The Teacher intervenes only when rolling collision rate and time-to-collision metrics stagnate, providing strategic guidance to redirect the search. In a CARLA case study with two experimental setups that vary the ego vehicle's speed policy, Teach-to-Crash achieves the highest Collision Hit Rate (90.79%), the shortest mean Time-to-Collision (18.31 s), and a competitive Collision Discovery Rate (136.21). PAFOT attains a higher mean CDR (179.44), but with substantially larger variance. Teach-to-Crash also yields the highest diversity (0.547) and, averaged across both setups on the CARLA Traffic Manager controller, the highest avoidability-based usefulness proxy (60.04%) among the compared methods. These results, within the evaluated CARLA scope, provide evidence that closed-loop dual-LLM reasoning can steer adversarial simulation-based testing over a constrained executable program space, generating failures that are frequent, structurally diverse, and assessed as more frequently avoidable. --- ### KITE: KV-Invariant Transformer Expansion for Efficient Agentic LLM Scaling Authors: Zhiheng Hu, Yixun Wei, Jian Zhou, Yizhuang Zhou, Ji Li, Xing Chen, Yang Li, Bojun Wang, Yibo Zhu, Xiangyu Zhang, Daxin Jiang Published: 2026-09-23 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.27294v1 Scaling a language model is not only a question of final quality: the architectural choice determines how much computation is spent during training, prompt processing, and autoregressive decoding to achieve certain model quality. An ideal model architecture should lower all above computation costs to facilitate scaling to a larger model, while ensure the larger model indeed outperforms smaller baselines. We introduce KV-Invariant Transformer Expansion (KITE), a scaling paradigm that achieves this goal. It trains the model from a smaller size to a larger size (i.e., saving training costs via upcycling), while places newly added parameters in regions that do not affect attention KV. Consequently, during inference, prefilling KV only relies on the smaller part of the model, so the inference costs are saved. As a concrete instantiation, we present Step Scale Transformer (SST), a two-tower decoder in which one tower produces KV and the other reads them. At comparable cumulative training compute, SST, a 67B MoE model with 2.15B active body parameters per decode token, achieves lower training loss than 47B and 63B MoE Transformers with 1.48B and 2.02B active body parameters, respectively, while reducing estimated inference cost by 6.7% and 31.6%. --- ### NGN: Learning Neural Network Size as a Differentiable Count Authors: Lixing Li Published: 2026-09-23 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.27291v1 Neural network size is usually chosen before training, separating architecture selection from weight optimization. We introduce the Neurogenesis Network (NGN), a differentiable parameterization for learning how many ordered structural components a model should use. For each ordered component group, one learnable boundary selects an active prefix while the model parameters are trained. The boundary can grow from a compact initialization and can be deployed by discarding components beyond the learned boundary. Controlled experiments examine convergence of the learned boundary, the performance of deployed prefixes, and comparisons with fixed-size models and alternative approaches to learning capacity. We then apply the same mechanism to MLPs, convolutional and graph networks, Transformers, state-space models, LoRA, and adapters. Across these settings, deploying only the learned prefix usually changes performance little, and the selected architectures perform similarly to fixed models trained at the same size. These results show that structural capacity can be optimized directly as a count. --- ### Sparse-Observation Atmospheric Thermal Forecasting with Physics-Informed Neural Networks for Climate-Aware Digital Twins Authors: Tannaz Goodarzvand Chegini, Elyas Shivanian, Behzad Karimi, Faraz Dadgostari Published: 2026-09-23 Categories: cs.AI, math.AP, math.NA arXiv: https://arxiv.org/abs/2609.27290v1 Short-horizon forecasts of atmospheric temperature are needed to support climate-aware digital-twin systems, but such forecasts must be produced where thermal observations are incomplete. This study evaluates a physics-informed neural network for potential-temperature forecasting, constrained by a pressure-coordinate thermodynamic advection-source equation and a diabatic-source closure fit from the preceding 12-hour period and frozen before future-time training. Using hourly ERA5 reanalysis at three pressure levels, the model is evaluated as a conditional hindcast at lead times of one, two and three hours against persistence, local-trend, and two matched neural-network baselines, one of which receives the same future meteorological forcing as the PINN, helping distinguish the physical constraint from access to future forcing. In an Oklahoma development case, mean RMSE improvement over the strongest baseline grew from 8.1\% at one hour to 23.8\% at three hours; under an observation-density sweep down to 5\% of candidate locations, this 3-hour advantage remained 14.6--16.9\%, with no evidence that lower density improves performance. Under a fixed protocol transferred to an Alabama heat event with three virtual-observation layouts, three-hour improvement ranged 19.7-24.4\% with consistent origin-level wins. A parallel Montana stress test, in which fixed pressure levels intersected complex terrain, produced a three-hour degradation of roughly 17.5\%, identifying a terrain-related applicability limit of the formulation. Together, these results indicate that the physics constraint's benefit grows with forecast horizon, persists under severe observation sparsity, and transfers across regions, but is bounded by the validity of a fixed vertical-coordinate representation over complex terrain, evidence relevant to physics-constrained components of climate-aware forecasting and digital-twin systems. --- ### Ruby-ASR: Evidence-Preserving Supervision for Joint Orthographic and Lexical-Reading Recognition Authors: Hao Shi, Yun Liu, Xuehao Yang, Jun Liu, Chuanbo Hua, Xuanjun Chen, Lianbo Liu, Shiao Zhu, Zixiong Su Published: 2026-09-23 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.27289v1 Conventional Japanese automatic speech recognition (ASR) is supervised by an orthographic transcript, although the same written form can correspond to different lexical readings realized in speech. Such utterances receive an identical target, so their reading distinction is absent from the supervision interface and cannot be recovered reliably by post-hoc text-only grapheme-to-phoneme conversion. We present Ruby-ASR, which refines the conventional target into a span-bound orthographic--lexical-reading sequence. Unlike separate full-sentence orthographic and phonological outputs, the ruby representation locally binds each written span to its realized reading and permits deterministic recovery of both views. We instantiate the target under subtitle-style and verbatim-style transcription conventions using a Qwen3-ASR backbone; a mora-level CTC objective provides auxiliary monotonic reading supervision. The experimental results across five Japanese benchmarks show that refining the recognition target can improve lexical-reading recovery without sacrificing readable orthographic transcription. We release the checkpoints and inference code. --- ### PotARCin: Multi-Dimensional Evaluation of Skill Acquisition in Abstract Reasoning Tasks Authors: Claas Beger, Ryan Yi, Melanie Mitchell Published: 2026-09-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.27288v1 The Abstraction and Reasoning Corpus (ARC) has become a prominent benchmark for evaluating general abstract reasoning and fluid intelligence in AI models. Yet standard ARC evaluation considers only a single capability: producing the correct output grid for a test input. We argue that this narrow format fails to evaluate the diversity of abilities that genuine abstract skill acquisition should enable. We introduce PotARCin, a benchmark that extends ARC by assessing understanding of a task's underlying abstract rule across five dimensions: Definition, Classification, Constrained Generation, Editing, and Inversion. PotARCin employs programmatic methods to generate new task instances and transform given inputs for a given ARC task, enabling dynamic generative sampling beyond fixed input-output pairs. Across five state-of-the-art models evaluated on the ARC-AGI-1 training set, we observe a 25-52 percentage-point performance gap between standard ARC evaluation and evaluation on PotARCin, and find that multi-dimensional evaluation reorders models that standard accuracy ranks alike. We further investigate effects of generative sampling, difficulty of corruption types, and questions of self-consistency, showing that models frequently contradict their own formalized rule even where they have stated it correctly. We also introduce P-ARC, a held-out hand-crafted test set, on which models achieve 1-8% accuracy across all five dimensions, underscoring the importance of more holistic evaluations of abstract reasoning capabilities. --- ### SR-Fraud: An Outcome-Supervised Reflective LLM Agent Framework for Non-Stationary Payment Fraud Detection Authors: Xuwei Tan, Yao Ma, Xueru Zhang Published: 2026-09-23 Categories: cs.LG, cs.CR arXiv: https://arxiv.org/abs/2609.27287v1 Real-time payment fraud detection is a non-stationary streaming prediction problem: adversaries adapt before supervised labels mature, and localized burst attacks can cause losses before retraining. Production systems typically rely on tabular classifiers and rules, which can struggle to capture these emerging sequential patterns before periodic retraining occurs. We present SR-Fraud, an outcome-supervised reflective LLM framework that decouples request-time decisions from offline adaptation. A frozen, stateless agent scores each transaction from a Hybrid Episodic Window to track behavioral shifts, while an offline reflection agent proposes boundary hypotheses from matured errors. A deterministic verifier then admits only supported hypotheses into an executable knowledge state. On a production payment-fraud benchmark, SR-Fraud improves all detection metrics over its frozen decision agent, obtains higher point estimates than static and periodically retrained CatBoost, and detects an emerging fraud burst. --- ### Memory Control Signals Emerge Before Action in Long Horizon Agents Authors: Mingxuan Wang, Guorun Yao, Fei Luo, Yinglong Guo, Chao Ning, Bo Wang, Hongyue Chen, Yanbiao Ma, Jungong Han Published: 2026-09-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.27286v1 Long horizon language model agents continuously accumulate interaction history, increasing computational cost while making relevant information harder to preserve and reuse. Existing context management methods mainly focus on how to compress or retrieve history, but largely leave open whether the model itself already represents the need for these memory operations before they occur. We study the hidden state immediately before each agent action and find that compression and recall needs are already encoded in the model's internal representations. These signals cannot be explained by simple context length or interaction progress, and they exhibit distinct formation patterns across model depth. We further show that most memory decision information is preserved in a compact recent context, while selectively restored historical evidence complements the long range dependencies that recent context misses. Based on these findings, we propose Preaction Memory with Evidence Retrieval (PaMER), which combines state guided compression with external evidence retrieval. PaMER+ further introduces step level evidence selection to recover only the historical information required by the current task. Experiments on WorkBuddyBench, across multiple context management baselines and model backbones, show that our framework substantially reduces context consumption while maintaining competitive task performance. --- ### Hunyuan-A13B Technical Report Authors: Tencent Hunyuan Team, Ao Liu, Botong Zhou, Can Xu, Chayse Zhou, ChenChen Zhang, Chengcheng Xu, Chenhao Wang, Decheng Wu, Dengpeng Wu, Dian Jiao, Dong Du et al. Published: 2026-09-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.27284v1 We present Hunyuan-A13B, an open-source large language model based on a Mixture-of-Experts architecture. It contains 80 billion total parameters but activates only 13 billion during inference, balancing model capability, computational efficiency, and deployment cost. The model is pretrained on a rigorously filtered 20T-token corpus with enhanced STEM data curation, improving factual reliability and reasoning ability. High-quality supervised fine-tuning and large-scale reinforcement learning further enhance its overall performance. Hunyuan-A13B also introduces a dual-mode Chain-of-Thought framework that adapts reasoning depth to task complexity: fast thinking for routine queries and slow thinking for complex, multi-step problems. Evaluations show competitive performance across mathematics, science, programming, general language understanding, and agent tasks, often approaching that of much larger models. Its high inference throughput makes it suitable for latency-sensitive applications. We release Hunyuan-A13B to support open research and practical LLM deployment. --- ### Multitask Regression with Pairwise Fusion Authors: Xiaodong Li, Zhentao Li Published: 2026-09-23 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2609.27280v1 We study multitask regression when coefficient sharing can differ by predictor. For a given predictor, many tasks may have the same coefficient while a few differ, and the exceptional tasks need not be the same for another predictor. We describe this structure by two quantities: the number of active predictors and the total number of task coefficients that differ from the most common value for their predictor. We estimate the coefficient matrix by penalizing all pairwise coefficient differences across tasks, with an additional group penalty when predictor selection is needed. The resulting upper and lower bounds have the same dependence on these two quantities. We also consider the stronger setting in which a large set of tasks shares one entire coefficient vector. Under explicit sample-size conditions, the same pairwise estimator pools those tasks exactly, while allowing the remaining tasks to differ. Simulations and household energy data illustrate the transition between broad sharing and task-specific coefficients. --- ### EnSIMem: Entity-Structured Indexing for Long-Term Agent Memory Authors: Xuanyu Meng, Xing Fan, Xinyi Fan, Chenlei Guo, Yixuan Xie, Jiawei Han Published: 2026-09-23 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.27279v2 An agent that interacts with users over long periods must recall facts, preferences, events, and changes from a continuously growing interaction history. Existing memory systems often compress interactions into generic summaries or retrieve anonymous text chunks, making it difficult for an agent to identify the correct entity, property, and supporting evidence. We present EnSIMem, an entity-structured long-term memory architecture for an agent. During offline construction, the system organizes interactions into theme-coherent episodes and builds dialogue-grounded index entries of the form (entity)(entity_type)(property: value). Each entry preserves its source turns, temporal information, and available multimodal fields. During online interaction, the agent's request is decomposed into evidence requirements whose properties are aligned with the memory index. Entity-property lookup and adaptive retrieval then collect the evidence needed for point, temporal, compositional, and aggregation reasoning. The agent generates its response from the preserved source evidence rather than from lossy memory summaries. On long-term agent-memory benchmarks, EnSIMem achieves high answer accuracy while maintaining compact, evidence-focused contexts. These results show that entity-structured indexing and episode-level provenance provide a reliable foundation for long-term memory in agents. The code of our model is available at https://github.com/RamonMeng/EnSIMem. --- ### Graph Learning with Spectral Connectivity Priors for Scarce Data Authors: Mingxiao Liu, Bahar Oveisgharan, Bingyan Zou, Gene Cheung, H. Vicky Zhao, Feifei Gao Published: 2026-09-23 Categories: cs.LG, eess.SP arXiv: https://arxiv.org/abs/2609.27278v1 Learning a sparse graph from scarce data is practically important but challenging. Motivated by the desirable combination of local sparsity and strong global connectivity exhibited by expander-like graphs, we propose spectral connectivity-regularized graph learning (SCoGL), a framework that incorporates a family of Laplacian spectral priors to explicitly promote global connectivity. Specifically, SCoGL augments a combinatorial-Laplacian-constrained graphical lasso (GLASSO) objective over a target adjacency matrix $\mathbf{W}$ with a general connectivity prior computed from Laplacian eigenvalues. We derive gradients for several representative connectivity priors and develop a projected gradient descent (PGD) algorithm with Armijo backtracking to efficiently optimize $\mathbf{W}$. Experiments show that the proposed SCoGL variants improve graph recovery and enhance downstream tasks such as graph signal denoising when signal observations are scarce. --- ### TimeEvo: Failure-Driven Self-Evolution of a Time Series Agent Authors: Jie Yang, Yan Zheng, Jiarui Sun, Xiran Fan, Junpeng Wang, Liang Wang, Zelin Xu, Qinghua Liu, Zhengyu Fang, Yiwei Cai, Philip S. Yu Published: 2026-09-23 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.27277v1 Time series agents answer analytical questions by calling external tools, and which tools they carry is decided by people before the agent runs. However, we identify two failures in this setup. Human-Agent Tool Misalignment: a library of 21 expert-curated tools helps on some tasks and hurts on others, dropping anomaly accuracy under every backbone we test. Silent Harm: one round of generic self-revision changes 147 answers and breaks 56 of them, while the final score moves by less than a point. Both follow from the same gap: whether a tool helps is decided question by question at runtime, while tools are supplied in advance and judged by a single average. To address this, we propose TimeEvo, which clusters an agent's diagnosed failures into capability gaps, plans a measurement for each, synthesizes evidence-only tools that fill them, and admits the candidate library only through a paired admission gate. Experiments on ten time series QA tasks and three backbones show that TimeEvo, starting from an empty library, improves accuracy on every task and every backbone, and that a library grown on a cheap model still gains when it is installed into stronger ones. Code is available at https://github.com/Muyiiiii/TimeEvo. --- ### DRSR: Learning Set-Level Deletion Risk for Efficient Long-Horizon Agents Authors: Mingxuan Wang, Bo Wang, Fei Luo, Guorun Yao, Chao Ning, Yinglong Guo, Hongyue Chen, Yanbiao Ma, Jungong Han Published: 2026-09-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.27276v1 Long-horizon language-model agents accumulate reasoning traces, tool exchanges, and observations whose relevance changes with the current decision. Existing compression strategies often score historical units independently, but the safety of deleting several units is generally not determined by their singleton scores: redundant evidence, accumulated small effects, and the information that remains after deletion all matter. We introduce Direct Relational Set-Risk Pruning (DRSR), which formulates agent-history compression as risk-constrained selection over deletion sets. Offline, DRSR constructs exact counterfactual supervision by jointly deleting protocol-valid history Blocks and measuring the change in teacher-forced likelihood of the same recorded next output. A lightweight scorer then predicts set-level harm from online-visible relations between candidate history and the current pre-action state, together with deleted-retained and pairwise set structure. At deployment, DRSR evaluates a small set of structurally valid deletion candidates with the lightweight scorer and removes the largest feasible set under recency, protocol, budget, and learned-risk constraints, abstaining when no set is sufficiently safe. On WorkBuddyBench Full260, DRSR increases mean reward from 0.699 to 0.802 while reducing total model tokens by 20.820%. On the fixed Eval40 comparison, it obtains 0.794 reward at 1.211M tokens per task, using 35.850% fewer tokens than the uncompressed agent. Mechanistic analyses and ablations further show that decision-conditioned relations, retained-context information, pair interactions, and abstention each contribute to reliable pruning. --- ### High Dynamic Range Video Reconstruction from Single-Exposure Raw Sequences Authors: Tao Zhang, Peixian Su, Xingyu Gao, Yunhao Zou, Yu Lu, Zunjie Zhu, Bolun Zheng, Ying Fu, Chenggang Yan Published: 2026-09-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.27274v1 Due to the limited dynamic range of conventional image sensors, captured low dynamic range (LDR) video often suffers from highlight clipping and shadow detail loss, making high-quality high dynamic range (HDR) reconstruction from single-exposure sequences highly challenging without alternating exposures or extra hardware. Alternating-exposure HDR methods sacrifice frame rate and struggle with motion alignment, making them impractical for real-world capture. To address this, we propose RawHDRV, an end-to-end framework for single-exposure Raw video HDR reconstruction, that fundamentally exploits the linear response and channel-specific characteristics of Bayer data. Specifically, it features a channel-decomposition temporal alignment and fusion strategy that processes Bayer channels separately to exploit their distinct exposure characteristics, together with exposure-aware weighted fusion. It further incorporates an exposure complementarity mask-guided restoration module that leverages inter-frame exposure redundancy to adaptively fuse reliable information and suppress saturation artifacts, and introduces a mask-guided color loss that combines normalized error constraints with gradient smoothing to enhance highlight recovery. Furthermore, we construct a large-scale mobile Raw-HDR video dataset with per-frame HDR annotations. Experiments show that our method achieves the state-of-the-art results in all metrics, demonstrating superior spatial quality and temporal stability under extreme exposure conditions. The code is available at https://github.com/supeixian/RawHDRV. --- ### CAVEAT: Towards Robust Computer-Use Agents in Incentive-Misaligned Environments Authors: Yuxuan Li, Will Epperson, Wesley Deng, Zezhou Huang Published: 2026-09-23 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.27273v2 Computer-use agents (CUAs) increasingly act on behalf of users online. What happens when the environments they operate in have incentives of their own? Online marketplaces, for example, may favor some products over others, steering agents away from the user's objective. Existing CUA benchmarks cover cooperative settings or explicit attacks, but do not test whether agents preserve user objectives when the environment itself has a stake in the outcome. We introduce CAVEAT, a controlled benchmark spanning nine marketplace environments and a taxonomy of eight common steering mechanisms. Across five model families, agents purchase the user-optimal product in 78.6% of matched-control episodes but only 17.3% when steering mechanisms are enabled. Larger models and more reasoning improve robustness, but substantial failures persist. Trajectory analysis and targeted ablations identify three weaknesses in how agents decide: they (1) prematurely narrow the set of alternatives they consider, (2) impose priorities the user never stated, and (3) commit before resolving decision-relevant evidence. Guided by this diagnosis, we develop CAVEAT-Harness, which targets these failures and raises the optimal purchase rate by up to 80.0 percentage points, and show that targeted post-training further improves a smaller open model. These results establish incentive robustness as a distinct challenge for delegated agents, diagnose failure modes, and show how targeted interventions can substantially improve robustness. --- ### GaussPDE: Graph-Based Partial Differential Equation-Driven Rendering for 3D Gaussian Splatting Authors: Haoyuan Yue, Fengyuan Ye, Ziyin Li Published: 2026-09-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.27264v1 We present GaussPDE, a framework that injects physically structured partial differential equation (PDE) dynamics into pretrained 3D Gaussian scenes without mesh extraction, voxelization, or retraining. Our key observation is that PDE rendering requires not only accurate appearance, but also a reliable discrete computational domain. We therefore first introduce camera-aware regularization during 3DGS reconstruction to suppress camera-near floaters and oversized primitives that would create unstable graph topology. We then construct an active Gaussian graph using covariance-aware distances and opacity, appearance, and boundary-aware conductance, enabling mass-weighted graph Laplacian PDE evolution directly over Gaussian primitives. The evolving scalar PDE state is coupled back to rendering by modifying the direct-current spherical harmonic color coefficients while preserving geometry, opacity, and view-dependent rendering behavior. Experiments on real and synthetic scenes show that GaussPDE produces stable, controllable, and spatially coherent dynamic visualizations, with reduced cross-boundary leakage compared with baselines. --- ### Can One Adapted Model Do It All? Fine-Tuning Strategy Selection for Customer Support LLMs Authors: Md Tahmid Rahman Laskar, Xue-Yong Fu, Shashi Bhushan TN Published: 2026-09-23 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.27262v1 Production customer-support systems often require LLMs to support multiple skills, such as intent classification, question answering, summarization, or tool-use decisions. A central deployment question is whether these skills should be handled by separate task-specialist models or by a single model trained through multi-task training, sequential updates, or model merging. We study this question using thirteen models spanning five families (Qwen3, Qwen3.5, Gemma-3, Llama-3.1, and Mistral) from 0.6B to 32B parameters across eight customer-support datasets, spanning four public and four proprietary datasets with approximately 74.5k training and 8.7k evaluation samples. Under a fixed training protocol, we train more than 200 checkpoints. Our experiments reveal that multi-task full fine-tuning is the strongest operational default at every model size we test. Specialist models are strong on their target tasks but often degrade sharply off-task, making reliable routing important. Sequential Low-Rank Adaptation (LoRA) preserves earlier skills better than sequential full fine-tuning, while merging a specialist with its base model improves off-task robustness with limited same-task loss for larger models. We conclude with practical guidelines for selecting fine-tuning strategies in real-world settings. --- ### UniDataAgent: An Ontology-Grounded Agent for Enterprise Question-to-Report Automation Authors: Yutai Duan, Yahui Zhao, Zhangti Li, Yu Ma, Zhenfeng Qi, Shaoyang Yuan, Jing Fan, Jie Liu Published: 2026-09-23 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.27257v1 Enterprise data agents must preserve organization specific semantics, not just translate questions into queries. We present ChinaUnicom DataAgent (UniDataAgent), an ontology grounded system for reusable question-to-report analysis that separates semantic acquisition from online execution. Ontology Acquisition and Validation stage (OAV) builds versioned enterprise ontologies from metadata, business knowledge, and supporting materials through expert authored business skills, constrained generation, question verification, and selected expert review. Question-to-Report Execution (QRE) stage retrieves semantic contracts for each question, coordinates skills and data tools, validates results, and produces evidence linked reports. Across 27 enterprise tables and roughly thousands of metric types, ontology construction took a few hours instead of about one week manually. It took just a few minutes to generate the reports, instead of several working days. Ontology grounding achieved 95.0\% strict accuracy on real business questions, versus 72.5\% for document RAG, especially on structured and compositional tasks. The system has already been deployed to generate cost savings and has the potential to be replicated in other enterprises. --- ### Functional Causal Discovery via Conditional Covariance Ordering Authors: Keyu Li, Ruoxu Tan Published: 2026-09-23 Categories: stat.ME, stat.ML arXiv: https://arxiv.org/abs/2609.27256v1 We study causal discovery where each node is a random function. Previous studies on this topic rely on structural assumptions, e.g., linearity or non-linearity, and distributional assumptions, e.g., Gaussianity or non-Gaussianity. In contrast, we make use of covariance operators to avoid these assumptions. Under functional additive noise models, we propose a new sufficient condition to identify a valid topological ordering based on comparing norms of conditional covariance operators. Taking advantage of this identifiability condition, we develop a new mixed regression model that subsumes linear and non-linear models. Together with variable selection, our procedure yields an estimation of the causal directed acyclic graph (DAG) for functional variables. In theory, we develop the least-squares-type theory of this regression model, and derive asymptotic consistency of order determination, sparse regression, as well as identifying the DAG. Computational algorithms based on discrete observations are provided. Applied to simulated data, our approach performs satisfactorily among existing approaches. A real data example of brain effective connectivity is also presented. --- ### What Converges in the Platonic Representation Hypothesis? Structure over Geometry Authors: Junwon You, Mihyun Jang, Sangwoo Mo, Jae-Hun Jung Published: 2026-09-23 Categories: cs.LG, cs.CV, math.AT arXiv: https://arxiv.org/abs/2609.27252v1 The Platonic Representation Hypothesis suggests that increasingly capable models converge toward shared representations. Recent work narrows this claim to shared local neighborhood relationships, finding that capacity-dependent trends in several global similarity measures largely disappear after calibration. We challenge this interpretation by showing that prior local-global comparisons confound structural scale (local versus global) with what is compared: relational structure, defined by which samples are related, versus metric geometry, characterized by quantitative relations such as distances, similarities, or correlations. To disentangle these factors, we construct a controlled $2\times2$ framework that evaluates both relational structure and metric geometry at local and global scales. We introduce $H_0$ skeleton overlap as a global counterpart to mutual $k$-nearest neighbors, together with matched distance-aware variants. Across vision-language models, relational structure exhibits robust representational convergence at both scales after calibration, whereas increasingly stringent distance agreement substantially weakens alignment and progressively flattens the capacity-dependent trend. We further extend the analysis beyond ambient Euclidean geometry by evaluating distance agreement under a Riemannian metric approximation and recover the same structure-geometry pattern. The pattern is also reproduced in video-text representations. Together, these results show that relational convergence extends beyond local neighborhoods to global spanning structure, whereas metric geometry exhibits substantially weaker convergence. --- ### The Risk-Sensitive Schrödinger Bridge: Is Not a KL Projection Authors: Hamidreza Behjoo Published: 2026-09-23 Categories: cs.IT, cs.AI arXiv: https://arxiv.org/abs/2609.27250v1 The Schrödinger bridge owes its computational power to a single structural fact: by Girsanov's theorem the controlled problem is a Kullback--Leibler (KL) projection onto a fixed reference measure, solvable by alternating projections. This letter shows that the fact does not survive risk sensitivity. When the expected path cost is replaced by the entropic risk measure and both endpoint marginals are kept as hard constraints, the resulting fixed-point bridge value $J_θ$ (the soft-problem value at the multiplier that enforces the terminal constraint) admits no representation as a constrained KL minimum against any fixed path-space reference with a regular endpoint law (a class strictly larger than the uniformly elliptic diffusion references: no Markov property is required), even allowing an additive normalisation depending on the initial marginal. Moreover, no single reference generates the one-parameter family in the risk parameter. The obstruction is computed in closed form: the Gaussian bridge value violates, by exactly $θ/2$, a heat equation that any Gaussian smoothing of a fixed endpoint density must obey. In place of the projection, the theory rests on a terminal-multiplier fixed point and an asymmetric factorisation penalising the score energy of the backward factor. --- ### Repurposing Pre-trained LLMs as High Fidelity Continuous Text Autoencoders Authors: Arkanath Pathak, Unnat Jain, Alexander C. Berg Published: 2026-09-23 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.27248v1 Next-token prediction has enabled highly fluent autoregressive language models, but it represents global structure only indirectly through sequential factorization. In contrast, high-fidelity autoencoders have become a standard primitive in image generation, enabling generative models to operate over continuous latent spaces; text lacks a comparably faithful continuous representation. We propose LLMAE, a method for repurposing a pretrained decoder-only language model as a continuous text autoencoder by exposing an intermediate fixed-length latent bottleneck within its internal activations. Instantiated with a parameter-efficient 270M Gemma 3 model, LLMAE uses structured attention masks, LoRA adaptation, and KL regularization to learn an autoencoding interface that leverages the generative prior of the original LLM. We train LLMAE to reconstruct text sequences up to 1024 tokens, significantly improving on this task to achieve near-perfect reconstruction. Furthermore, we demonstrate the downstream utility of this representation by training a latent text diffusion model for detailed image captioning using the learned LLMAE autoencoder. By mapping text into a fixed-length continuous latent space, our approach provides an effective substrate for downstream adaptation while benefiting from the fluency of the original LLM. --- ### Listening and Mirroring: The Effects of Verbal Attunement and Behavioral Mimicry on Social and Empathic Perceptions of Embodied AI Agents in VR Authors: Nathalia Gomez, Haig Shamlian, Omar Khan, Tiffany D. Do Published: 2026-09-23 Categories: cs.HC, cs.AI arXiv: https://arxiv.org/abs/2609.27246v1 As embodied agents take on increasingly social and relational roles in VR, visual realism and embodiment alone may be insufficient; users must also perceive these agents as emotionally attuned, supportive, and humanlike. Prior work suggests that verbal attunement and nonverbal mimicry can each improve users' social evaluations of embodied agents. However, behavioral mimicry has largely been studied outside of real-time, conversational AI interactions, leaving limited understanding of how users respond when an agent simultaneously generates contextually responsive dialogue and adapts its nonverbal behavior during an immersive conversation. To address this gap, we developed an embodied AI counselor that combines conversational AI with real-time facial-expression and posture mimicry, while producing either verbally attuned or neutral responses. We evaluated the system in a 2 X 2 within-subjects study with 20 participants, manipulating verbal attunement and behavioral mimicry. Results showed that verbal attunement was the most reliable driver of perceived empathy. Behavioral mimicry showed a marginal relationship with perceived humanness, while greater mimicry exposure showed preliminary, exploratory positive associations with empathy, positivity, and humanness, particularly among female participants. Together, these findings show that multimodal synchrony is not a simple additive strategy for designing empathic conversational agents in VR and underscore the need to consider how verbal and nonverbal behaviors are combined during real-time interaction. --- ### Full-Covariance Smoothing of Bayesian Neural Networks for Online Adaptation Authors: Oren Wright, Haoming Jing, Qiaoan Shen, Koichiro Niinuma, Yorie Nakahira, José M. F. Moura Published: 2026-09-23 Categories: cs.LG, eess.SY arXiv: https://arxiv.org/abs/2609.27244v1 A neural network's layers can be treated as time steps of a state-space model, turning Bayesian training into a smoothing problem: a forward pass propagates Gaussian moments through the network, and a backward Rauch--Tung--Striebel pass updates the weight posteriors in closed form. Such methods learn from each observation in a single pass, in an uncertainty-aware manner, and without gradient-based iterations or replay, which makes them well suited for online adaptation and data-efficient learning. Existing smoothing-based methods, however, are restricted to diagonal covariances across activations, discarding correlations between neurons. We overcome this limitation via a cross-covariance identity that enables full-covariance propagation through a network's nonlinear activations. We derive a one-step-per-layer smoother that approximates as Gaussian only each layer's affine output, and that applies both to deterministic systems with noisy observations and to stochastic systems described by output statistics. We demonstrate this method in non-stationary classification, online dynamics learning, and policy adaptation of a vision-language-action model, and find that it is generally more accurate than other smoothing-based methods. --- ### Combining LLMs and Genetic Search for ARC-AGI-2 Authors: Val Dyachenko Published: 2026-09-23 Categories: cs.NE, cs.AI arXiv: https://arxiv.org/abs/2609.27242v1 LLMs can generate programs for ARC-AGI-2 tasks, but the provided compute only allows a small number of attempts to generate, debug and validate solutions. Genetic algorithms can search and test many more programs, but random search rarely starts in a useful neighborhood of the solution space. We combine the two methods through a compact domain specific language (DSL). First, a quantized Qwen3.5-4B LLM generates an initial set of programs for each ARCAGI-2 task. Then, we use those programs to seed an initial population of starting programs, and use genetic algorithms to evolve these programs towards a solution to the given task. The DSL is designed such that every mutated program remains valid and can be executed. The initial programs proposed by the LLM solve 2 (3.3%) of the first 60 tasks of the ARC-2 public evaluation set. The genetic algorithm solves an additional 4, giving 6 correct test outputs in total (10.0%). If we try using evolving solutions without this LLM seeding, we do not arrive at any solutions at all. The results show that genetic search can improve programs generated by LLMs and produce additional correct solutions. --- ### On the Sample Complexity of Active Learning with Membership Queries Authors: Ganghua Wang, Shaddin Dughmi Published: 2026-09-23 Categories: stat.ML, cs.LG, math.ST arXiv: https://arxiv.org/abs/2609.27241v1 This work revisits a fundamental question in active learning: how powerful is the ability to synthesize arbitrary queries? Compared to pool-based active learning, where the learner only selects queries from a given unlabeled pool, we find that this seemingly mild change in query ability may dramatically alter the difficulty of statistical learning. In particular, some hypothesis classes that are inherently slow to learn in the pool-based setting, achieving only polynomial error decay in the number of samples, become exponentially learnable once synthesized queries are allowed. This striking gap suggests that membership query synthesis induces a fundamentally different mode of learning, one that is not adequately captured by existing active learning theory and calls for new analytical tools to characterize its complexity. Motivated by this phenomenon, we develop several sufficient conditions, present intriguing examples, and propose a conjectural perspective toward understanding which hypothesis classes admit efficient learning through synthesized queries. --- ### Strip Convolution and Direction-Aware Exclusion Loss for Oriented Ship Detection Authors: Bin Chen, Yuanyuan Liu, Peng Yang, Chao Lu Published: 2026-09-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.27238v1 Oriented ship detection in very high resolution (VHR) remote sensing imagery remains challenging due to elongated hull geometry and dense target distributions in complex port scenes. Existing methods typically address geometric representation and duplicate suppression separately. To jointly tackle these issues, we propose an oriented ship detector with two complementary components. The C3k2_Strip module employs orthogonal strip convolutions to better capture elongated hull structures, while the Class-Aware Direction-Aware Exclusion Loss (CA-DAEL) suppresses redundant predictions using class, direction, and confidence cues. Experiments on HRSC2016 and DIOR-R achieve 78.45% and 53.71% mAP50:95, respectively, with only 2.91M parameters. On HRSC2016, the proposed method improves mAP50:95 by 6.32 percentage points over the YOLOv11-OBB baseline, demonstrating its effectiveness for accurate oriented ship detection. --- ### Discover, Falsify, Revise: Auditing Input-Use Claims from Source Code to Predictive Contribution in Agent-Discovered Cell Models Authors: Mengran Li, Bo Li, Chengyang Zhang, Yang Yan, Jinfeng Xu, Zhenchao Tang Published: 2026-09-23 Categories: cs.LG, q-bio.QM arXiv: https://arxiv.org/abs/2609.27234v1 AI virtual cells aim to predict cellular responses to specified interventions, yet held-out predictive performance alone does not establish use of the supplied perturbation information. This prediction-claim gap matters in agentic model discovery, where language-model agents generate and revise predictors using score-based feedback. We introduce CELLAUDIT, which audits input-use claims by asking whether an input can enter the cited computation, whether fitted predictions depend on it, and whether that dependence improves prediction of observed response. On a paired morphology-transcriptomics perturbation benchmark (BBBC047), an agent-selected predictor attains a mean held-out Global Pearson correlation coefficient (PCC) of 0.3153 but remains invariant to compound replacement; a control-profile-only predictor reaches 0.3142. Source inspection identifies a compound-query pathway blocked by singleton key-value attention, and the invariance persists after refitting with disjoint control wells. In a stratified audit of 48 candidates across two linked tasks, 47 change predictions under compound replacement on both held-out folds, but only 20 show target-loss gains with intervals above zero on both folds. On BBBC047, falsification-guided revisions recover positive mean compound contributions while retaining gains over the control-profile-only baseline. In matched sci-Plex searches, audit-enriched feedback yields higher held-out performance and larger mean compound and dose contributions across five trajectories, although paired intervals span zero. Refitting fixed designs on an independently acquired cohort shows predictive generalization need not imply generalization of input-use claims: dose contribution persists, whereas support for compound identity does not. CELLAUDIT adds a falsification layer to agentic model discovery, moving from generate-score-revise toward discover-falsify-revise. --- ### Distilling Sequential Computation in Transformer Language Models Authors: Zixuan Lan, Jessica Yang, Yanhong Li, Karen Livescu, Jiawei Zhou Published: 2026-09-23 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.27233v1 Transformer language models process sequences token by token in an autoregressive manner, making growing contexts increasingly expensive. Yet many adjacent token spans are highly predictable or frequently occur as stable units, suggesting that their representations may be compressible. We introduce a method for distilling sequential computation by replacing spans of input tokens with collapsed representations, computed on the fly by a lightweight merge module. This module generates a single surrogate embedding from a sequence of static token embeddings that captures the functional role of the multiple tokens, allowing pretrained models to operate on compressed inputs without architectural changes or re-training. We apply this approach during inference to compress both prompts and intermediate decoding steps, using a rollback mechanism to substitute stored multi-token KV cache entries with their single-step surrogates. Experiments across diverse models show that the merge module can be used to reduce effective sequence length by up to 40% with minimal accuracy degradation across language modeling evaluations and downstream tasks, including question answering, summarization, commonsense reasoning, and long-form mathematical reasoning. Additional lightweight adaptation of the merge module further improves the accuracy-compression trade-off in selected settings. These results demonstrate that sequential token computation in Transformers can be effectively approximated through condensed surrogate representations that approximate the original behavior without model updating. --- ### A Scaling Study for fMRI Foundation Models Authors: Wenhao Ye, Xuanye Pan, Junfeng Xia, Junxiang Zhang, Mo Wang, Quanying Liu Published: 2026-09-23 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.27232v1 Scaling laws have guided large-model development in computer vision and natural language processing, but the relationships among data, model size, and compute remain unclear for functional magnetic resonance imaging (fMRI) foundation models. Here, we conduct a controlled empirical study using pretraining data from more than 200 source datasets and over 10,000 GPU-hours of experiments. Holding the pretraining framework and downstream protocol fixed, we vary pretraining data size, model size, and training duration. Downstream performance generally improves with compute, yet models using similar compute can perform substantially differently. Additional pretraining data bring larger gains at larger model sizes, suggesting that data and model size should be scaled together. At matched compute, increasing pretraining data benefits more tasks than increasing model size, although the pattern varies across tasks. We then use in-distribution (ID) downstream performance to select the combination of pretraining data size, model size, and training duration at two fixed compute budgets. The resulting models are locked before out-of-distribution (OOD) evaluation. They achieve the highest average performance across the evaluated OOD tasks among the compared fMRI foundation models while using less pretraining compute. Overall, our results show that compute alone does not characterize fMRI scaling: performance depends on how pretraining data, model size, and training duration are combined. --- ### Surgical Kinematics from Monocular Video with Learned Articulated Motion Constraints Authors: Mehmet Kerem Turkcan, Soham Samal, Zoran Kostic Published: 2026-09-23 Categories: cs.CV, cs.RO arXiv: https://arxiv.org/abs/2609.27227v1 Objective assessment of robotic surgery uses instrument kinematics, which must be reconstructed when only video is available. We introduce a kinematic reconstruction network for estimating instrument position, orientation and jaw angle from monocular video. Our visual representation combines global attention pooling of frozen DINOv3 features with local pooling at instrument landmarks from fine-tuned SAM 3.1 masks. Our shared Transformer encoder and temporal convolutional heads integrate this representation with mask geometry, monocular depth and visual state estimates from arm-specific multilayer regression networks. Our position branch predicts displacement magnitude and direction separately to preserve traveled distance. We fit trajectories to predicted state observations and motion increments by differentiable weighted least squares, expressing quaternion observations relative to cumulative predicted rotations to obtain a quadratic orientation objective. We evaluate reconstruction across 2,802 Open-H episodes. Compared with LiveMAE on the main Open-H benchmark, our method reduces path-length mean absolute error from 0.45 to 0.34\,cm and increases temporal mean average precision for motion segmentation from 44.54\% to 54.44\%. --- ### Meet, Compare, or Abstain: LatWeave for Deterministic Multi-Hop Question Answering on Knowledge Lattices Authors: Yuze Ren, Shaoheng Fan, Tao Wang, Yabo Yan, Han Han Published: 2026-09-23 Categories: cs.CL, cs.AI, cs.IR arXiv: https://arxiv.org/abs/2609.27225v1 Probabilistic question-answering systems -- whether large language models (LLMs) themselves, retrieval-augmented generation (RAG), or trained multi-hop retrievers -- conflate "what is known" and "how to reason" into a single probabilistic computation: hallucination cannot be eradicated, evidence chains cannot be audited, and the system answers even when it does not know. We present LatWeave, which organizes knowledge into a multidimensional knowledge lattice and compiles multi-hop QA into three deterministic operators -- meet (constraint intersection), compare (lattice-order comparison), and abstain (structural abstention); LLMs appear only on the construction side (one-shot extraction) and the query-planning side, while the answer-generation path is zero-LLM, zero-task-training, and auditable end to end -- so that question answering over Web-published knowledge becomes reproducible item by item. Rather than claiming across-the-board SOTA, we characterize the operating envelope of this paradigm on six public benchmarks: when knowledge is complete (MetaQA, 39,093 questions) meet chains are near-lossless over three hops (any-hit 0.9975, on par with fully supervised KBQA); on templated multi-hop home ground (2WikiMultihopQA held-out n=1,258) EM 0.865, well above published structure-augmented RAG reproductions; on open-text deep composition (MuSiQue) and extraction-coverage gaps (HotpotQA) we report degradation honestly and attribute it to causes outside the lattice-algebra layer; and when information is incomplete (IIRC) we achieve structural abstention with abstain accuracy 0.971 and leak rate 0.029. Within the operating envelope, deterministic execution pays no performance penalty, and every step on the answer path can be recomputed -- precisely the source of end-to-end auditability. --- ### Physiologically Informed Digital Auscultation for Pneumonia Detection in Long-term Care Residents Authors: Nicholas Rasmussen, Oleg Zaslavsky, Zih-Ling Wang, Hongyu Yu, Joelle Fathi, Kaibao Nie, Amil Khanzada, Tomoko Ito Published: 2026-09-23 Categories: cs.SD, cs.CV, cs.LG, eess.AS, q-bio.TO arXiv: https://arxiv.org/abs/2609.27222v1 Pneumonia is difficult to diagnose in older long-term care residents; multimorbidity and atypical presentations obscure signs, motivating operationally efficient objective testing. We analyzed multi-channel digital stethoscope recordings from 185 Japanese residents (73 pneumonia, 112 symptomatic without), using radiologist-confirmed chest X-rays and clinician diagnoses as supervisory signals that train convolutional neural networks, multimodal fusion, and channel-based variants with time-domain Grad-CAM interpretability. Models were evaluated with repeated patient-level cross-validation showing models with X-ray supervision outperformed clinician supervision (F1 0.729, accuracy 0.783 vs. F1 0.637, accuracy 0.711). Additionally, a three-channel selection protocol maintained performance (F1 0.736; accuracy 0.803), with two mid-thoracic sites ranking highest and Grad-CAM attention overlapping adventitious sounds. These findings indicate automated multi-channel lung-sound analysis can aid long-term care pneumonia diagnosis, with X-ray supervision being more reliable than clinical, and fewer channels preserving performance while lowering acquisition times. --- ### Tail-Aware Geometry Learning for Conformal Ellipsoids Authors: Xiang Zhang Published: 2026-09-23 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.27221v1 This paper studies multivariate conformal prediction (CP), a distribution-free uncertainty quantification framework with finite-sample coverage guarantees. The efficiency of multivariate prediction sets hinges critically on the residual geometry encoded by the nonconformity score, while existing minimum-volume methods rely on quantile thresholds that ignore tail residual severity and implicitly bind geometry learning to coverage level. We propose a tail-aware geometry learning framework for conformal ellipsoids that decouples tail sensitivity in geometry learning from the final coverage guarantee. Using a two-split design, we learn the metric matrix via volume minimization under a CVaR constraint on an estimation split, then apply standard conformal calibration on a held-out calibration split. The resulting problem is convex and admits a bounded-reweighting interpretation that prioritizes high-residual samples. Moreover, we theoretically characterize the trade-off between ellipsoidal volume and tail severity. Experimental results demonstrate the effectiveness of the proposed method. --- ### LOCKR: A Hidden-State Trajectory-Guided Planner for Detecting and Repairing Stable-but-Wrong Lock-In in Diffusion Language Models Authors: Guoshenghui Zhao, Tan Yu, Weijie Zhao Published: 2026-09-23 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.27220v2 Diffusion language models generate text through iterative denoising, exposing intermediate trajectories before final answers are produced. We identify a recurring reasoning failure, stable-but-wrong lock-in, where an answer stabilizes early around an incorrect value while substantial denoising remains. Surface-level decoding signals such as confidence, entropy, margin, and answer stability are insufficient to reliably distinguish correct from erroneous lock-in. We formulate selective reasoning repair as a lightweight test-time planning problem and propose LOCKR, a hidden-state trajectory-guided planner that decides when to allocate additional computation, expands a structured set of targeted repair branches, and selects the most promising continuation using trajectory-aware verification. Across two diffusion language models and three mathematical reasoning benchmarks, hidden-state trajectories consistently outperform surface signals and single hidden snapshots for both wrong-lock-in detection and repair selection. On natural evaluation distributions, LOCKR yields absolute accuracy gains of 2.21--5.37 percentage points across all five evaluated settings, with repair rates ranging from 22% to 41%. These results establish hidden diffusion trajectories as actionable signals for selective test-time reasoning repair. --- ### Learning Spectral Allocation: A Fractional Diffusion Framework for Adaptive Volumetric Segmentation Authors: Yi-Hui Shen, Tie-Qiang Li Published: 2026-09-23 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.27217v1 We address adaptive computation in 3D medical image segmentation: instead of designing another backbone, we ask how much spectral mixing each network stage needs and let optimization answer. We derive FHEAT, a two-parameter operator family, from the discrete cosine transform (DCT) solution of a fractional heat equation. A fractional order alpha and a diffusion strength D govern the operator, and at D=0 it is exactly the identity. Reparametrized by the semigroup time tau = D*alpha, same-resolution instances compose exactly, so any distribution of diffusion across same-resolution stages amounts to a single Sobolev-type regularizer of learned strength. This identity limit lets the optimizer of each layer, not the designer, decide whether global mixing is needed and how sharp it should be. We instantiate FHEAT in a lightweight U-shaped architecture (Light-UNETR) paired with a Kolmogorov-Arnold mixer (KAN3D) with adaptive rational activations, yielding FHEAT-Seg. At 5% to 20% label rates on three public benchmarks, training produces gradient-driven spectral sparsification: seven of the eight stage-level operators drive D to zero, and the survivor saturates at the sharpest low-pass (alpha ~ 0.9) in the decoder layer feeding the semi-supervised attention map. The retired layers become exact identity shortcuts at inference, cutting FLOPs from 4.29G to 0.90G (a 79% drop) at 0.975M parameters. Under a standard semi-supervised protocol, FHEAT-Seg reaches Dice scores of 90.47% (left atrium), 78.79% (Pancreas-CT), and 81.90% (BraTS 2019), ahead of five semi-supervised methods and the Light-UNETR baseline. The large variant also surpasses Light-UNETR-L under full supervision (Dice 93.09%, 85.11%, and 87.19%) with 2.851M parameters and 55.75G FLOPs. These results suggest that the allocation of spectral computation is a learnable property of optimization dynamics, not a manual design commitment. --- ### KATOsuper: Surrogate-accelerated neural topology optimization with sensitivity-consistent Fourier neural operators Authors: Shengyu Yan, Jasmin Jelovica Published: 2026-09-23 Categories: cs.CE, cs.AI, cs.LG, math.NA arXiv: https://arxiv.org/abs/2609.27216v1 Topology optimization (TO) remains computationally intensive due to repeated finite element analysis (FEA) evaluations required at each iteration. While neural network-based surrogates offer potential acceleration, existing approaches often suffer from gradient inconsistency between predicted objectives and sensitivities, leading to optimization instability. This work presents KATOsuper, an objective-agnostic framework that couples neural-reparameterized topology optimization with a Sensitivity-Consistent Fourier Neural Operator (SC-FNO). The framework employs the forward_split architecture, which derives deployed sensitivities via automatic differentiation through the predicted objective field and thereby preserves consistency between the predicted objective and the gradient used for optimization. The case studies include three 2D benchmark problems and three 3D structures considering compliance or stress minimization. A physics-informed multi-channel input encoding with Fourier position embedding enables resolution-invariant learning, supporting zero-shot extrapolation beyond the training resolution, with useful performance at moderate scaling factors and topology-preserving exploration at up to 64x without retraining. The framework extends to 3D through KATO3D, featuring novel KANConv3D blocks with learnable B-spline activations. KATOsuper demonstrates 15--110x deployment-time speedup over MATLAB baselines while maintaining competitive optimality, with the clearest gains observed in complex 3D and stress-optimization cases. The insight that sensitivity direction matters more than magnitude enables robust optimization even with approximate physics evaluation, extensible to other differentiable physics-driven design objectives. --- ### Scalable Subgraph Sampling via Resistance Curvature Authors: Chaoqun Fei, Tinglve Zhou, Tianyong Hao, Yangyang Li Published: 2026-09-23 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.27209v1 Subgraph sampling reduces the training cost of large-scale graph neural networks, but sampling criteria may overlook the geometric roles of edges. We propose a resistance-curvature-guided sampling framework built on ERC-LG, a curvature approximation method for large-scale graphs. ERC-LG combines Johnson-Lindenstrauss projections with regularized multi-GPU batched conjugate gradient solvers, avoiding explicit Laplacian pseudoinverse computation and full embedding storage. The resulting curvature informs node- and edge-sampling probabilities for constructing GNN training subgraphs. Experiments show numerical agreement with pseudoinverse-based curvature and reduced runtime compared with CG-only computation. ERC-LG-based sampling variants achieve the highest mean accuracy on six of seven real-world datasets in downstream node classification. --- ### Benchmarking Active Spot Selection for Cost-Efficient Spatial Transcriptomics Authors: Zheyu Zhu, Junchao Zhu, Fengbei Liu, Tianyuan Yao, Gelei Xu, John Cannon, Haichun Yang, Yuankai Huo, Mert R. Sabuncu, Ruining Deng Published: 2026-09-23 Categories: cs.CV, cs.LG, q-bio.OT arXiv: https://arxiv.org/abs/2609.27208v1 Spatial transcriptomics (ST) measures gene expression in tissue context, but dense capture grids can be costly and may repeatedly sample morphologically similar regions. Most active learning strategies were developed for categorical labels and independent samples. We conduct a retrospective pool-based benchmark of active learning versus uniform Random sampling for ST, where expression vectors are high-dimensional and continuous and candidates are spatially correlated. Using two fully profiled public ST cohorts, we mask candidate expression vectors and simulate multi-round selection with uncertainty-based Monte Carlo dropout (MC-dropout) and temporal output discrepancy (TOD), and diversity-based CoreSet and TypiClust-inspired selection. We compare 160 completed configurations at 5%, 10%, 30%, and 50% of the fold-wide training spot pool under patient-level cross-validation, with a separate full-label reference. Within each budget, strategies share the selection schedule, morphology-to-expression predictor, and optimization protocol. We assess mean per-gene within-slide Pearson correlation coefficient (PCC), expression-cluster agreement, and Moran's I fidelity. On HER2-positive breast cancer, pooled mean PCC differences from Random across the four active strategies were -0.0176, -0.0117, +0.0056, and +0.0057 at 5%, 10%, 30%, and 50%, respectively. On cutaneous squamous cell carcinoma (cSCC), three strategies were below Random at 5%, and all four were below Random at 10%. On HER2-positive breast cancer, CoreSet and MC-dropout had lower PCC but higher expression-cluster agreement than Random at the two smallest budgets; this pattern did not reproduce on cSCC. Under the reported fixed training horizons, the evaluated active strategies do not consistently improve on Random at small budgets, and rankings depend on the evaluation measure. --- ### Prediction with Expert Advice: Anytime Regret with Many Experts Matches the Fixed-Time Constant Authors: Yang Cai, Vineet Gupta, Yanchen Jiang, Christopher Liaw, Aranyak Mehta, Grigoris Velegkas, Di Wang Published: 2026-09-23 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2609.27206v1 Prediction with expert advice is a fundamental problem in online learning. When the time horizon $T$ is known in advance, the minimax cumulative regret over $n$ experts is asymptotically $\sqrt{\frac{T \ln n}{2}}$. This is achieved by the Multiplicative Weights Update algorithm with a learning rate tuned to $T$, and is known to be tight. If instead the regret bound is required to hold simultaneously at every time $t$, the best known guarantee has been $\sqrt{t \ln n}$---a factor of $\sqrt{2}$ worse---and it has remained unknown whether this factor of $\sqrt{2}$ is necessary. We show that it is not. We give an algorithm, requiring no knowledge of the horizon, whose cumulative regret satisfies $R_t \le \bigl(1 + O(\sqrt{\ln \ln n / \ln n})\bigr)\sqrt{t \ln n / 2}$ simultaneously for every $t \ge 1$. --- ### Phonemizing User-Generated Text: A Benchmark, Taxonomy, and Compositional Approach Authors: MinJu Jeon, Younghan Park, Han Sung Park, Jong-Hwan Kim, Dong-Jin Kim, Hoyeon Lee Published: 2026-09-23 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.27205v1 Text-to-speech systems increasingly process user-generated text (UGT) such as ppl and imo, whose pronunciation must be inferred from the canonical rather than surface form. We introduce UGTPhon, the first grapheme-to-phoneme (G2P) benchmark for UGT in English, Vietnamese, and Korean, together with an inference-grounded taxonomy for fine-grained diagnosis. Existing G2P models and frontier LLMs exhibit a systematic canonical-to-non-canonical performance gap, reaching up to 66.8 PER points. As a benchmark baseline, we propose a simple compositional G2P approach that incorporates canonical-form evidence through exact-match lookup and staged decoding. Across matched ByT5 and Qwen2.5-0.5B backbones, explicit canonical-form modeling consistently reduces non-canonical G2P errors. The 0.5B variant also performs competitively with much larger few-shot frontier LLMs, highlighting the benefit of explicitly modeling canonical-form inference for UGT phonemization. ---