# Research Papers — 2026-09 · bundle page 16 of 266 > Full-text markdown of papers #751–800 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/16 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/15.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/17.md) ### Beyond Lip Sync: Reference-Grounded Oral Refinement for Audio-Driven Portrait Animation Authors: Bangxun Tang Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.38019v1 We present RGOR (Reference-Grounded Oral Refinement), an audio-driven lip-sync framework that renders the mouth of the specific person being dubbed rather than a generic one. Existing lip-sync systems follow the audio closely and keep the face recognizable, yet the mouth they render is an average mouth: the shape and texture of the lips, the arrangement of the teeth, and how much of them shows as the mouth opens are not that person's. The problem persists because nothing in current training or evaluation asks for the person's own mouth: perceptual losses accept any plausible mouth, face identity is carried mostly by the skin around it, and the released inference code of inpainting systems uses the unmasked target frame as the reference, which hides the gap. To address this, RGOR conditions every generated frame on frames from separate enrollment recordings of the same person and on HD patches of the mouth that bypass the VAE, and trains the generator against a paired judge that compares each rendered mouth with the person's reference and learns to reject a realistic mouth of someone else. We further build an evaluation protocol and use it to compare open-source and commercial lip-sync systems on held-out identities. Experiments show that RGOR achieves the best or second-best result on most metrics, and preserves the person's own lip and dental detail while keeping synchronization and the rest of the face intact. --- ### Prompts Live on an Arc: Gaussian Curricula in Fisher--Rao Coordinates for Rollout-Efficient GRPO Authors: Mei Okonkwo, Pixel Nomand, Julian Berg, Elena Voss, Lena Park, Marcus Hale, Adrian Cho, Sofia Reyes Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.38018v1 Group relative policy optimization (GRPO) learns only from prompts whose sampled responses disagree: a group that is entirely correct or entirely incorrect has zero reward variance, contributes no gradient, and still consumes its rollouts. Prompt-selection methods reduce this waste by steering sampling toward intermediate pass rates, but they choose the target, its width, and the uncertainty model heuristically, in raw pass-rate or logit coordinates. We show that GRPO comes with a natural coordinate for pass rates: the arc length $ψ=\arcsin\sqrt{p}$ on the Bernoulli Fisher--Rao manifold. In arc length, the expected GRPO update is uniform up to two boundary ramps; the probability of a zero-variance group is bounded by two Gaussian boundary layers of width $1/\sqrt{2G}$; pass-rate evidence has constant noise; and the gradients of the pass@$k$ and pass$^k$ objectives are Gaussians whose center and width follow from $k$ in closed form. A prompt curriculum for GRPO is therefore a Gaussian in arc length, and choosing its center amounts to choosing the objective. We turn this observation into ARCUS, a drop-in sampler that tracks every prompt with a Kalman filter in arc length, scores prompts by an objective-matched Gaussian kernel times the predicted probability of an informative group, keeps only informative groups for the unchanged GRPO update, and paces the target toward the hardest objective whose predicted yield stays within a small slack of the best. Across six mathematical reasoning benchmarks and three backbones, ARCUS improves the average accuracy of GRPO by 2.8--2.9 points and that of dynamic sampling by 1.1--1.2 points, while generating 48--57\% fewer rollouts than dynamic sampling. --- ### Brain-SAD: A Brain-Inspired Safe Autonomous Driving Control Framework with Dynamic Fear-Oriented Constraint on Dual-Policy Authors: Huan Rong, Chao Yin, Anouar Imel, Yijie Xia, Tinghuai Ma Published: 2026-09-29 Categories: cs.AI, cs.CV, cs.NE, cs.RO arXiv: https://arxiv.org/abs/2609.38016v1 Constrained Reinforcement Learning has recently gained increasing attention in the field of Safe Autonomous Driving, where the general mechanism is to maximize the expected reward while keeping the overall action risk bounded. In this way, the safety issues arising in AD can be mitigated through constrained actions. However, existing Constrained RL methods still lack dynamics on the imposed constraints. For instance, the action cost adopted by the existing Primal-Dual/soft-constrained methods is often defined as static state-to-cost mapping, and the safe-action projection in hard-constrained methods relies on the static projection with the fixed feasible region boundary estimated from offline demonstrations. The above drawback tightly couples the imposed constraints to the training scenarios, leaving the AD policy hard to handle different interaction scenarios, due to the improper state-level action-cost and the static projection boundary. Consequently, in this paper, we propose Brain-SAD, a brain-inspired safe autonomous driving control framework with dynamic fear-oriented constraints. By perceiving the current vehicle-interaction scene, Brain-SAD generates dynamic fear signal as fear reaction to online decide long-term policy for regular interaction or short-term policy for urgent-collision defense. In such two policy, the above fear-reaction will be constructed as the dynamic fear constraints, respectively reflecting the overall fear cost directly coupled with action-impact, and the dynamic fear boundary of the feasible region derived from different risky neighbors, both of which will in turn serve for the online policy optimization. Experimental results show that Brain-SAD outperforms existing methods, achieving higher success rate in shorter task-completion and collision-recovery time, and exhibits stronger reliability across continuous intersections of fluctuating complexity. --- ### When do data mixtures improve scaling laws? Insights from high-dimensional regression Authors: Diyuan Wu, Lehan Chen, Theodor Misiakiewicz, Marco Mondelli Published: 2026-09-29 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2609.38011v1 Modern machine learning systems are trained on mixtures of data from different domains, and choosing the right mixture can substantially improve downstream performance. Despite an extensive literature on data mixing and reweighting, existing work is largely empirical and it remains unclear when auxiliary data genuinely improves scaling laws rather than merely providing more samples. To gain insight into this question, we study a high-dimensional mixed-data regression model with a shared regression function, heterogeneous covariances and noise levels, and dataset sizes that may grow at different rates. We establish the minimax risk under an ellipsoidal parameter constraint for the general covariance structure and derive deterministic equivalents for the test error of ridge regression under commutative covariances. We then specialize to a target domain and an auxiliary domain with aligned power-law covariance spectra, where the theory yields explicit scaling laws in terms of spectral decay, target regularity, and the relative growth of the two datasets. These laws identify regimes in which combining data mixtures provably yields a faster scaling rate than using either dataset alone. In particular, improving the scaling law requires a specific interplay between spectra and relative sample sizes of the domains. Our numerical experiments on language models exhibit the same qualitative phenomenon: appropriate data mixtures yield a faster decrease in target-domain test loss than training on either domain alone. --- ### From Unity Simulation to Diffusion-Based Augmentation: Quantifying Dataset Balance for Robust Object Detection Authors: Mohamed Benkedadra, Aissa Saoudi, Maxime Gloesener, Sidi Ahmed Mahmoudi, Matei Mancas Published: 2026-09-29 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.38010v1 Modern computer vision models achieve high accuracy when trained on large-scale annotated datasets. In critical domains such as construction safety monitoring, data collection is costly, hazardous, and ethically constrained. This paper presents a systematic study comparing two complementary data generation paradigms, (1) Unity Simulation-based rendering and (2) Controllable Diffusion-based generation (CIA), for object detection under real data-scarce conditions. A unified experimental framework enables controlled dataset mixing across real, simulated, and generative sources, while maintaining identical model and training settings. Quantitative evaluation using Precision, Recall, mAP, and custom $Δ$-metrics, reveals that neither simulation nor generative augmentation alone achieves optimal transferability. Unity-only training yields an mAP@0.5 drop of $-50\%$ relative to real data, while CIA-only training shows a milder $-16.5\%$ degradation. Hybrid compositions significantly improve performance, with the 90\% real + 10\% Unity configuration achieving the best overall mAP@0.5 of $62.68\%$ ($+7.64\%$ over baseline), and the 90\% real + 10\% CIA configuration maximizing precision at $74.45\%$. Results demonstrate that limited synthetic inclusion enhances generalization, while excessive substitution induces domain drift. --- ### HybridCUA: Learning to Orchestrate GUI and CLI for Computer-Use Agents Authors: Tongbo Chen, Junbo Niu, Zhengxi Lu, Niu Lian, Fei Tang, Yuchen Yan, Yike Hong, Yong Du, Yizhou Liu, Bofan Chen, Yongliang Shen Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.38008v1 Computer use agents (CUAs) have demonstrated strong capabilities in completing digital tasks. However, existing CUAs either rely solely on graphical user interface (GUI) interactions, which are often inefficient and error prone, or augment GUI interactions with application specific APIs or tools, which require substantial engineering effort and are difficult to scale across applications. We argue that the next generation of CUAs should combine GUI interactions with the command line interface (CLI), leveraging the generality of the GUI and the efficiency of shell commands. A critical challenge, however, is that current models do not know when or how to use the CLI during task execution. To address this challenge, we develop a data construction pipeline that produces three types of trajectories: GUI only, CLI only, and interleaved GUI and CLI trajectories. This pipeline results in HybridCUA-8K, containing 5K hybrid trajectories and 3K verified RLVR tasks. Building on these data, we propose a training framework with two stages: supervised fine tuning on the constructed trajectories, followed by reinforcement learning with our CLI aware rewards that encourages agents to use the CLI selectively and reliably. Experiments show that HybridCUA-9B achieves 53.6% accuracy on OSWorld, improving over the base model by 14.8 percentage points, and improves performance on WindowsAgentArena by 4.0 percentage points. These results demonstrate the effectiveness and cross platform generalizability of the hybrid GUI and CLI paradigm for computer use agents. --- ### HARISSA: Inference-Time Self-Checks for Efficient and Safe Local Language Model Deployment Authors: Kenan Alkiek, Moontae Lee, David Jurgens, V. G. Vinod Vydiswaran Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.38006v1 Running a language model locally offers advantages in privacy, latency, and cost, but local hardware fits only small models, which are less capable than frontier models. The usual remedy for a hard query, escalating it to a cloud model, gives up the privacy and cost advantages of running locally. A deployment that stays local faces two decisions for hard queries instead. First, it can spend more computation on a query, e.g., reasoning before answering, which raises accuracy at a cost in latency, so it must decide which queries are worth the extra computation (efficiency). Second, some queries are beyond the local model, and delivering a wrong answer is worse than deferring the query to a human in the loop, so it must decide which answers are safe to deliver (safety). We show that both decisions can be made from the model's own hidden states. The prefill state, computed before any token is generated, predicts whether the model will answer correctly, and the answer state, at the end of the generated answer, predicts whether that answer is correct. HARISSA fine-tunes the model so that both states predict correctness, then makes both decisions with one policy that cascades through the ways of answering from cheapest to most expensive, skipping a way the prefill state predicts will fail and deferring the query when the answer it stops with is predicted wrong. On a device running a single model, HARISSA is within one accuracy point of chain-of-thought at 2.7 times lower latency. On a server holding four sizes of one model, HARISSA is more accurate than the FrugalGPT and Self-REF cascades at the same latency, and at the same deferral rate the answer state leaves fewer wrong answers than the standard confidence signals in five of six task and setting pairs. --- ### Diagnosing and Improving Probabilistic Reasoning in Large Language Models Authors: Huaman Sun, Dingcheng Wang, Jason Hartline, Jessica Hullman Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.38005v1 Large language models (LLMs) are increasingly proposed as decision assistants who must reason probabilistically from available evidence under explicit decision costs. We propose a decision-theoretic framework that decomposes LLMs' decision loss into two components: forming accurate beliefs from provided evidence and translating those beliefs into actions that optimize a provided utility function. Using a synthetic benchmark with known ground truth, we apply the decomposition to characterize probabilistic reasoning in frontier and open-sourced models. We further evaluate whether RL interventions targeting beliefs, decisions, or both improve these components across three domains, whether improvements transfer across components and elicitation formats, and whether decision performance can improve without improvement in belief formation. We find that targeting one component of probabilistic reasoning redistributes decision loss, improving the target without necessarily transferring to others, and that jointly targeting belief formation and decision-making improves both but hinges on matched formats between training and evaluation. --- ### No Scale Left Behind: Multi-Scale Autoencoder with Bi-directional Attention for Time Series Anomaly Detection Authors: Jiaheng Guo, Haochen Zhang, Yu-Chao Huang, Jinhao Duan, Nicholas Konz, Tianlong Chen Published: 2026-09-29 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.38004v1 Time series anomaly detection (TSAD) plays a crucial role in healthcare, finance, industrial monitoring, and other sectors. Within and between these settings, anomalies span vastly different temporal scales, from sub-second point spikes to multi-hour drift patterns. However, most existing TSAD methods commit to a single temporal granularity, and multi-scale designs either analyze different scales in isolation or are constrained to a predefined coarse-to-fine hierarchy, both failing to sufficiently capture multi-scale interactions. To resolve this limitation, we propose Multi-Scale Autoencoder with Cross-Scale Attention for TSAD (MSCAD), a simple yet powerful semi-supervised TSAD framework founded on parallel autoencoder branches corresponding to different patch sizes. A stack of symmetric bidirectional cross-scale attention blocks enables every pair of scales to exchange information before reconstruction without allowing any single scale to be privileged. On the comprehensive TSB-AD benchmark (40 datasets, 530 series), MSCAD achieves large performance gains against 50 baselines across multiple metrics, with VUS-PR of 0.57(+9.6%) on the univariate split and 0.47(+9.3%) on the multivariate split compared to the state-of-the-art. --- ### Mutual Information Constrained Chernoff Bottleneck Authors: Dier Tang, Guangyue Han Published: 2026-09-29 Categories: cs.IT, cs.LG, math.OC, math.ST, stat.ML arXiv: https://arxiv.org/abs/2609.37994v1 The classical information bottleneck (IB) measures the relevance of a representation $U$ of $X$ to a target $Y$ by $I(U;Y)$, which does not directly characterize the error of downstream decisions. For a binary hypothesis $Y$ inferred from many separately encoded observations, the optimal error exponent is the Chernoff information between the two conditional distributions of $U$ given $Y$. We study the mutual information constrained Chernoff bottleneck, which seeks an encoder that maximizes this Chernoff information subject to a rate constraint $I(U;X) \leq R$. We show that its optimal value $C(R)$ increases strictly up to $R = H(V)$, where $V$ merges the symbols of $X$ with equal likelihood ratio, remains at the uncompressed exponent beyond, and, unlike the IB curve, need not be concave. We further show that $k+1$ outputs suffice to attain $C(R)$, where $k$ is the cardinality of $V$. We propose an alternating algorithm that updates the encoder via a generalized Blahut--Arimoto algorithm and the Chernoff parameter $s$ via a nonlinear equation, and prove that its iterates remain feasible, with nondecreasing and convergent Chernoff information. Numerical experiments confirm the theory, and on real topic-detection data from the 20 Newsgroups corpus, compressing each word to only $17\%$ of its entropy retains $90\%$ of the error exponent and nearly the accuracy of the uncompressed classifier. --- ### BITEM at the NTCIR-19 R2C2 Task: Predicting Confidence from Agentic RAG Pipeline Signals Authors: Julien Knafou, Luc Mottin, Alexandre Flament, Paul van Rijen, Esteban Gaillac, Patrick Ruch Published: 2026-09-29 Categories: cs.CL, cs.AI, cs.IR arXiv: https://arxiv.org/abs/2609.37993v1 The BITEM team entered both subtasks of the NTCIR-19 R2C2 task with a single agentic pipeline, in which a model searches, reads and records evidence over a movie corpus while an orchestrator holds the record and rules on what may be submitted. A claim is admitted only once an entailment cascade has checked it against the passage it cites, and an answer is released only once enough checked evidence stands behind it. Each question is run three or four times, every pass retrieving from a corpus stripped of what the earlier passes have already seen. The confidence filed with each answer is computed by the orchestrator from what the run leaves behind and is never asked of the model, which is offered no way to rate itself. The two retrieval runs placed 4th and 5th of 22, pooling the passes was worth 0.0709 nDCG@20, and the gain was largest on the multi-hop and post-processing-heavy questions, where the organisers rank the pooled run top of the field. Sixteen of the 25 answer runs were built on passages these two runs supplied, 12 of them filed by other teams. HMR rewards a system whose confidence is high where it answers right and low where it answers wrong. The pipeline reached an accuracy of 0.9219, 6th of 25, while the confidence filed with those answers gave an HMR of 0.4915, 13th. A few rules crafted over those same recorded signals, with no further model call and no further retrieval, raise that to an accuracy of 0.9375, 5th, and an HMR of 0.6985, 9th. Ranking on HMR alone can reward a system for answering wrongly with low confidence, so we propose accHMR, the accuracy multiplied by HMR, which reports the reward in proportion to the accuracy, and on which the revised rules would have scored 0.6549, 5th. For future work, fitting a model on the numbers the pipeline already produces, rather than writing such rules by hand, would be a real step forward. --- ### Which Attention Heads are like the Human Head? Not the Ones that Compute Authors: Christopher Pinier, Gustaw Opiełka, Hannes Rosenbusch, Taylor Webb, Michael D. Nunez, Claire E. Stevenson Published: 2026-09-29 Categories: cs.AI, q-bio.NC arXiv: https://arxiv.org/abs/2609.37991v1 Brain-AI alignment is often interpreted as a sign that model and brain perform similar computations. Whether the aligned units are causally involved in model computation is rarely checked. On an abstract pattern-completion task (AAABAAA $\rightarrow$ B), we compare LLM attention-head representations with human EEG and test how ablating those heads affects task performance. Alignment and causation dissociate: brain-aligned heads contribute to performance, but their removal is substantially less disruptive than removal of heads selected via attribution patching. We compare two head sets that prior interpretability work defines without reference to the brain: concept vectors (CVs), which represent abstract patterns across formats, and function vectors (FVs), selected for their contribution to correct-answer prediction. Brain alignment shows little association with FV scores, while its association with CV scores varies across models. Among brain-aligned heads, we find recurring attention profiles: one emphasizes distinctive elements (novelty heads), the other repeating elements (repetition heads). The novelty family tracks salience and attends to the same elements that humans look at, yet its removal is less damaging than random ablation on average. Repetition heads contribute modestly to performance and are associated with abstract-pattern representation (CVs). Across 17 models spanning 3B-72B parameters, FV-ranked removal is substantially more disruptive than brain-ranked removal. Brain alignment thus captures how the model reads the stimulus, and only faintly captures how it represents the pattern and solves the task. --- ### TabFM-Auto: Self-Evolving Pipelines for Tabular Foundation Models Authors: Deqing Fu, Huangyuan Su, Rajat Sen, Taman Narayan, Sujay Sanghavi, Abhimanyu Das, Weihao Kong Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37989v1 Tabular foundation models achieve strong zero-shot accuracy on structured data by pretraining on synthetic tables, but they ignore the column names, task descriptions, and auxiliary files that carry dataset semantics. Meanwhile, self-evolving machine learning engineering (MLE) agents train models from scratch on each dataset, yet jointly searching over features, architectures, and hyperparameters is noisy and prone to overfitting. We introduce TabFM-Auto, which pairs a tabular foundation model, TabFM, with a language model agent that evolves the data pipeline around it. Guided by dataset metadata and validation feedback, TabFM-Auto iteratively refines data cleaning, feature engineering, context selection, and post-processing to reduce TabFM's error. Across all 51 datasets of the TabArena benchmark, five TabFM-Auto configurations with different agents and language models take the top five overall positions, and the best raises TabFM from 1785 to 2013 Elo. The discovered pipelines also transfer to other frozen tabular foundation models (+69 to +143 Elo) with no further search. On the 8 tabular competitions of MLE-Bench, TabFM-Auto ranks first overall among MLE agents. --- ### KV-Kaizen: Learning Context-Adaptive Cache Compression Choices Authors: Joao Monteiro, Louis Béthune, Anastasiia Filippova, Sonia Laguna, David Grangier, Marco Cuturi Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37988v1 As the context size of text processed with an LLM grows, the size of KV caches can outstrip the memory allocated for the original model weights. This impacts LLM throughput negatively, since decoding is memory-bound and decode cost grows with cache size. Recent work alleviates this bottleneck by discarding the least relevant tokens. Eviction introduces a tension, since a one-off decision to discard content may prove detrimental later. Instead, we focus on alternative choices that can lead to cache compression without evicting tokens. We achieve this by learning a selector that is able to produce, based on context, a per-layer cache configuration towards an overall compression budget. The selector operates along three axes: sharing one cache across layers (depth), caching at fewer bits (precision), or truncating the low-rank latent cache representations (rank). We call the resulting method KV-Kaizen, for the many small per-layer choices it compounds. We observe that these interventions taken independently and uniformly over all layers limit achievable compression because they degrade accuracy. Crucially, composing them locally and adaptively to the context can instead preserve accuracy while achieving large memory savings. At inference, the selector runs once, before pre-fill. In evaluations on instruction following and reasoning tasks, our selectors reach the Pareto frontier of accuracy against cache size, against learning-free and post-hoc baselines. On long-context tasks, KV-Kaizen improves on eviction and can be composed with it, reaching a 32x smaller decode-time cache on a 14B model while preserving accuracy. A 4x cache size reduction incurs no accuracy degradation from 7B parameters up, and a compressed model is more accurate than a smaller uncompressed one with the same cache size. Together, these findings support pre-training large models and compressing them only afterwards. --- ### ORMA: Optimization-based Monocular 4D Reconstruction of Articulated Animals Authors: Xuyi Hu, Francesco Palandra, Shangzhe Wu, Daniel Cremers, Riccardo Marin, Silvia Zuffi Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37986v1 Recovering articulated 4D representations of animals from monocular videos remains challenging due to the large diversity of quadruped morphologies and lack of animal 4D supervision data. Existing learning-based reconstruction methods operate on individual images and rely on synthetic or model-fitted 3D supervision, which inherits the constraints of strong parametric priors and limits generalization to out-of-distribution species. When applied to out-of-distribution animals, they often recover a plausible pose while producing inaccurate geometry because the underlying shape model cannot faithfully represent the observed instance. We present ORMA, a training-free reconstruction framework that decouples articulation from shape, using the predicted pose as reference for optimization while leveraging generative 3D priors for accurate shape reconstruction. Given a reference image, we reconstruct the animal geometry and register it to the parametric model SMAL+, yielding an articulated shape adapted to the observed instance. We then combine per-frame articulated pose estimates with globally consistent camera poses to recover animal motion in a shared world coordinate frame, and further refine the reconstruction using self-supervised DINO correspondences and temporal consistency. To enable quantitative evaluation, we introduce PAW4D, a synthetic multi-species benchmark with ground-truth 3D geometry and camera motion. Experiments on PAW4D, PFERD, and challenging in-the-wild videos demonstrate that ORMA improves reconstruction accuracy while recovering globally consistend animal motion across diverse quadruped species. --- ### $S^3$: Spectral Null-Space Swap Makes Reasoning Models Efficient Authors: Hongbo Ma, Sansheng Cao, Jiajun Fan, Bangji Yang, Ge Liu Published: 2026-09-29 Categories: cs.LG, cs.AI, cs.CL, cs.CV arXiv: https://arxiv.org/abs/2609.37976v1 LLMs trained with Chain-of-thought excel in reasoning capability, but often come with excessive token cost. We find that the core of reasoning capacity lies in the Thinking model's weight component within the null space of a projection defined by the corresponding Non-thinking model's dominant singular directions, and removing the subspace component can largely improve reasoning efficiency without hurting the accuracy gained during thinking-mode post-training. Unlike existing efforts that mostly operate within the dominant subspace, we are the first to unveil the critical role of the null space and harness it for model optimization. Motivated by this finding, we propose Spectral Null-Space Swap ($S^3$), a training-free composition of paired Non-thinking and Thinking checkpoints. Our method keeps the Non-thinking model inside its own dominant subspace and takes the Thinking checkpoint outside it, improving reasoning efficiency while maintaining accuracy. We extensively evaluate $S^3$ on 2B-30B dense and mixture-of-experts (MoE) architectures spanning 28 evaluation environments across mathematical, multimodal, and audio reasoning domains. $S^3$ establishes new empirical Pareto Frontiers among training-free model composition strategies: across all settings, it reduces inference token overhead by an average of 27.4% compared to full Thinking models while simultaneously improving overall task accuracy by 1.0 percentage point (e.g., yielding +8.3% accuracy on HMMT25 alongside a 33.0% token speedup). We further use attention entropy for explanation and find that the retained component produces more concentrated attention, and we use a simplified analytical model about optimization to demonstrate why null-space can effectively reduce attention entropy, thereby improving the efficiency of reasoning. --- ### On Trajectory-Aware Training for Masked Diffusion Language Models Authors: Manuel Madeira, Amitis Shidani, Alice Bizeul, Victor Turrisi, Louis Béthune, Bhavika Devnani, Dan Busbridge, Pierre Ablin, João Monteiro Published: 2026-09-29 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.37974v1 Masked diffusion models (MDMs) generate text by unmasking several tokens per step, but they are trained and sampled under different conditions. The model is trained on randomly masked sequences, whereas inference follows a trajectory shaped by the model's own predictions. Additionally, each step has no access to what the previous one computed. Recent methods narrow these limitations from separate angles, leaving open how these choices interact. We introduce PUMBA, a unified framework for trajectory-aware training that trains the denoiser on consecutive steps of policy-induced trajectories, passes information between steps, and optimizes them jointly by backpropagation through time. A controlled study of this design space shows that i) exact train--inference alignment fails due to local overfitting, whereas a looser alignment still brings training masks closer to those seen at inference; ii) passing continuous information outperforms discrete gradient estimators through the commitment at each step; and iii) performance improves as backpropagation through time spans more steps, which we support theoretically. Combined, these components match the best checkpoint of a same-size autoregressive model. Building on these findings, we scale PUMBA to supervised fine-tuning of LLaDA-8B, where it improves the trade-off between performance and number of function evaluations (NFEs) in both full-canvas and block diffusion generation. At matched performance, it needs up to 22% fewer NFEs than standard fine-tuning with twice the budget in full-canvas generation, and up to 26% fewer than standard fine-tuning for the same number of steps in block diffusion. --- ### Dagger: Decoupling-based Model Stealing Attack against Graph Neural Networks Authors: Ying Song, Xiaowei Jia, Balaji Palanisamy Published: 2026-09-29 Categories: cs.CR, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.37972v1 As Graph Neural Networks (GNNs) are widely deployed as Machine Learning-as-a-Service (MLaaS) APIs, model stealing attacks have emerged as a critical security threat. By querying a victim model's black-box API, an adversary can construct a functionally equivalent surrogate model, compromising proprietary intellectual property and downstream security. Existing GNN stealing attacks, however, rely on overly permissive assumptions, such as soft-label outputs, large query budgets, full-graph query access, and prior knowledge of victim backbones that rarely hold in real-world deployments. In this work, we formalize a strictly constrained black-box, hard-label and backbone-agnostic threat model for GNN stealing attacks under a tight query budget. Given these realistic restrictions, we identify four fundamental challenges: sparse local structures and isolated nodes that degrade victim label quality, insufficient supervision signals, systematic imbalance with incomplete class coverage, and backbone mismatch. To address these interlocking barriers, we propose Dagger, a novel two-phase decoupling-based attack framework. Specifically, in Phase 1, Dagger pre-trains a surrogate using decoupled information propagation to preserve structural context over sparse local subgraphs while handling isolated nodes, combined with manifold-level node mixup to synthesize continuous supervision signals and smooth decision boundaries. In Phase 2, Dagger freezes the encoder and fine-tunes the classifier head via class-balanced sampling paired with logit adjustment to rectify severe query imbalance without requiring extra victim queries. Extensive experiments across four benchmark graphs and four GNN backbones demonstrate that Dagger consistently outperforms state-of-the-art GNN stealing attacks, achieving up to 18.16\% higher fidelity while only utilizing 12.23$\times$ fewer queries than the strongest baseline. --- ### PhysWAM: Physically Consistent World Action Model for Autonomous Driving Authors: Dhruv Parikh, Fengcheng Yu, Quankai Gao, Jiawei Yang, Junjie Ye, Maulik Bhatt, Thang Vu, Charles Ochoa, Rowan McAllister, Igor Vasiljevic, Rajgopal Kannan, Viktor Prasanna et al. Published: 2026-09-29 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2609.37970v1 World-action models (WAMs) jointly predict how a scene will evolve and how an agent should act, however joint generation alone does not necessarily impose a shared geometric constraint on these predictions. We present PhysWAM, a unified world-action model for autonomous driving that co-denoises multiview video, metric depth, and ego motion within a single flow-matching transformer. To ground world and action generation in measured scene geometry, we introduce Coupled Point Projection (CPP) that unprojects the generated depth into 3D points, transforms them using the generated $\mathrm{SE}(3)$ ego motion, and minimizes their distance to LiDAR points transformed using the recorded ego motion. This geometric constraint promotes physical consistency with the measured scene by jointly supervising generated depth and motion alongside their standard flow-matching objectives. At inference, trajectory selection relies only on a simple label-free consensus rule, with no learned scorer or simulator feedback. We evaluate PhysWAM across NAVSIM v1 and v2 planning, zero-shot closed-loop transfer, and future video and metric-depth prediction. Despite PhysWAM's simple selection procedure, it achieves strong planning performance and transfers zero-shot to unseen driving environments. It also generates accurate metric depth and temporally coherent video, with CPP improving both planning and depth prediction. Together, these results demonstrate that the geometric relationship between scene depth and ego motion provides a direct way to couple world and action generation within a simple unified model. --- ### SoL-Refiner: Speed-of-Light One-Step Refinement for High-Resolution Video Authors: Haozhe Liu, Tian Ye, Shuchen Xue, Yitong Li, Junsong Chen, Haopeng Li, Jincheng Yu, Duomin Wang, Ruihua Zhang, Lei Zhu, Song Han, Enze Xie Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37969v1 High-resolution video generation is expensive, as its cost grows rapidly with the number of spatiotemporal tokens. A practical alternative first generates a lower-resolution video and then applies a refiner, but conventional multi-step refinement introduces a second sampling bottleneck. We present SoL-Refiner, a one-step video refiner that transforms low-resolution model outputs into 4K videos with a single denoising step. Our three-stage recipe combines high-resolution continual training, reinforcement learning (RL) post-training, and a final one-step distillation. We introduce Refiner-Bench, a video refinement benchmark constructed from the outputs of different video generators, and use a shared-input protocol to compare refiners at approximately 2K output resolution. At 2K, the one-step SoL-Refiner outperforms all external refiners on the VBench and UniPercept averages, while at $3840\!\times\!2176$ it improves both metrics over the three-step LTX-2.3 Refiner. With the complete acceleration stack, SoL-Refiner achieves an $8.91\times$ speedup in refinement latency over the same baseline in our 2K latency setting. --- ### SelfSearch: Reward-Free Search for Self-Improving Agents Authors: Jungwoo Yang, In Jin Kong, Yohan Jo Published: 2026-09-29 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.37968v1 Advances in the coding capabilities of LLM agents allow them to inspect and modify their own instructions, tools, and execution procedures. Existing approaches use this ability to search for improved agents through repeated downstream evaluation, which incurs substantial costs and ties the search to the evaluated tasks. We introduce \textbf{SelfSearch}, a reward-free search procedure in which agents modify themselves using records of previous self-improvement episodes. These records capture the reasoning, tool actions, and outcomes of earlier modification attempts, providing concrete experience for improving both task solving and self-modification. Without downstream reward signals during search, SelfSearch improves population-mean success over the initial agent in all six model--benchmark settings, with individual agents gaining up to 11.2 percentage points on Terminal-Bench 2.1. On SWE-bench Multilingual, an agent improves success by \textbf{5.0} percentage points while reducing execution cost by \textbf{38.5}\% on tasks solved by both the initial and evolved agents. SelfSearch achieves competitive task success with evaluation-guided search baselines at lower search cost. With only \textbf{\$4.03} in search cost, it produces a harness that solves \textbf{82.0}\% of Terminal-Bench 2.1 tasks with DeepSeek V4 Flash under the settings of a public nine-harness comparison, matching the top-scoring harness, Codex. These results suggest that experience gained through self-modification can improve agents' downstream capabilities and efficiency. --- ### TabFM: A Zero-Shot Foundation Model for Tabular Data Authors: Weihao Kong, Erez Louidor Ilan, Shuxin Nie, Taman Narayan, Rajat Sen, Yichen Zhou, Deqing Fu, Samet Oymak, Abhimanyu Das Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37959v1 Tabular machine learning typically relies on per-dataset workflows, fitting tree ensembles or running AutoML searches from scratch for every task. We present TabFM, a 400M-parameter tabular foundation model that formulates supervised tabular prediction as in-context learning. TabFM produces calibrated zero-shot predictions in a single forward pass without task-specific tuning. Trained entirely on synthetic tables generated from structural causal models, TabFM learns general tabular representations that transfer zero-shot to real-world tasks. Across all 51 benchmark datasets in TabArena (38 classification and 13 regression), zero-shot TabFM ranks first among default tabular foundation models and outperforms tuned AutoML pipelines. Two extensions over the same frozen weights improve performance further on both tracks: multi-view feature expansion with ensembling and post-hoc calibration (TabFM+), and LLM-guided, dataset-specific data processing and feature engineering (TabFM-Auto). --- ### Kolmogorov-Arnold Classifier Systems as Universal Approximators Authors: Hiroki Shiraishi, Hisao Ishibuchi, Masaya Nakata Published: 2026-09-29 Categories: cs.LG, cs.NE arXiv: https://arxiv.org/abs/2609.37958v1 As the input dimension $n$ grows, rule-based machine learning, such as Learning Classifier Systems (LCSs), faces a fundamental scalability bottleneck for function approximation: both rule count and parameter count grow exponentially with $n$. Traditional LCSs partition the $n$-dimensional input space directly, requiring $\mathcal{O}(m^n)$ rules for adequate coverage, where $m$ is the per-variable resolution. This article breaks from this paradigm by reorganizing rules dimension-wise, guided by the Kolmogorov-Arnold representation theorem: any continuous $n$-dimensional function can be expressed as a finite superposition of one-dimensional functions. The proposed Kolmogorov-Arnold Classifier System (KACS) decomposes the target function into one-dimensional subproblems and assigns a dedicated ruleset to each, reducing the worst-case rule count from $\mathcal{O}(m^n)$ to $\mathcal{O}(mn^2)$ and replacing $n$-dimensional local models with one-dimensional models requiring only two parameters per rule, independent of $n$. We also provide the first constructive proof that an LCS, namely KACS, is a universal approximator for continuous functions on compact domains. Evaluated against a direct $n$-dimensional input space partitioning approach under otherwise identical conditions, KACS achieves competitive accuracy in many settings while using only 2\% to 40\% of the parameters. Our implementation is available at https://github.com/YNU-NakataLab/KACS. --- ### BrainNet Studio: A Unified Toolkit for Brain Network Construction, Intelligent Analysis, and Visualization Authors: Xiwei Zeng, Shengrong Li, Yiheng Liu, Chunwei Tian, Daoqiang Zhang, Qi Zhu Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37956v1 Brain networks characterize structural and functional relationships among brain regions and support research on cognition, brain disorders, and brain-computer interfaces. Their time-varying topology and higher-order spatiotemporal dependencies are not adequately represented by conventional static networks. Existing tools primarily focus on static connectomes and provide limited integration of dynamic network modeling with modern graph and sequence learning methods. We present BrainNet Studio, an integrated toolkit for static and dynamic brain network analysis. It provides a unified workflow encompassing network construction, feature extraction, predictive modeling, candidate biomarker identification, visualization, and assisted interpretation. The toolkit integrates 27 algorithms, including deep learning, graph neural networks, and spatiotemporal sequence models, to support classification and the identification of discriminative brain regions and connections. A large language model generates researcher-verifiable summaries of functional connectivity, structural connectivity, and structure-function coupling at individual and group levels. Within a consistent computational framework, users can configure analytical tasks, compare methods, inspect outputs, and extend functionality without repeatedly assembling application-specific pipelines. BrainNet Studio provides a practical and extensible platform for connectome analysis in cognitive neuroscience, exploratory studies of brain disorders, and brain-computer interfaces. The toolkit is publicly available at https://github.com/xbrainnet/Brainnet-Studio. --- ### Topological Coherence for Self-evolving Multi-agent Systems Authors: Sen Zhao, Ruiqi Kong, Zuyu Zhang, Lifeng Shen, Xinyu He, Xu Zhang, Qinghua Zhang Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37953v1 Complex tasks inherently couple workflow structure, agent responsibility, collaboration, and memory access: task regions delimit responsibility and tool scope, cross-region dependencies give rise to handoffs, and ownership boundaries delimit private and selectively shared memory. Existing methods can jointly optimize agent and communication structures, yet such optimization does not by itself require responsibility, handoff, and memory boundaries to remain consistent with task dependencies. We term this requirement topological coherence. We introduce TOCOMAS, a Topology-Coherent Multi-Agent System. TOCOMAS grounds a task graph in tool interfaces, organizes compatible task nodes into reusable responsibility domains, and derives dependency-induced and profile-conditioned collaboration together with boundary-regulated memory visibility. During online self-evolution, TOCOMAS proposes coupled changes to agent, collaboration, and memory policies, retaining for subsequent tasks only candidates that satisfy structural constraints and improve evaluated reward. Across BBEH, WorkBench, SWE-Bench-Verified, and CoMemBench, TOCOMAS improves task success over baselines across backbones. CoMemBench also shows gains over the self-evolving baseline in verified progress, handoffs, and memory isolation. --- ### Scene-Consistent Illumination Transfer for Inserted Advertising Graphics Authors: Rameshwar Mishra, Bishshoy Das, A. V. Subramanyam, Guan-Ming Su Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37951v1 Replacing a visible advertisement in a broadcast frame is geometrically straightforward but photometrically delicate. A pasted graphic can have the correct perspective and still appear detached when its brightness, shading, or shadow disagrees with the surface beneath it. This paper presents Ad-Relight, an inference-only procedure for transferring scene illumination to a supplied advertising graphic without collecting a banner-specific training set. The procedure first separates slowly varying shade from graphic structure, then probes a pretrained diffusion relighter with two nearly identical backgrounds to isolate the contribution of the target region. A final pass combines this residual with a smoothed luminance field and a soft attenuation mask. Across 560 generated placements, the approach improves structural similarity, perceptual distance, and illumination agreement over geometric compositing and direct relighting baselines. Human judgments and an automated preference study show the clearest gains on floor-mounted graphics with nonuniform lighting. The current study is image based; temporal stabilization remains an open extension. --- ### Video-RSI: Recursive Self-Improvement of Video Understanding Agents via Harness Evolution Authors: Bingjun Luo, Jialin Guo, Siqi Li Published: 2026-09-29 Categories: cs.AI, cs.CV arXiv: https://arxiv.org/abs/2609.37950v1 Video understanding agents acquire evidence through an executable harness that controls what they observe and how they use those observations. However, execution traces contain only the evidence acquired by the current harness, leaving competing explanations for failure unresolved and limiting the basis for self-improvement. We introduce Video-RSI, a framework for recursive self-improvement in which a video understanding agent uses its own language model to revise its harness. Through active video investigation, the model revisits the original training videos to test competing failure explanations with additional observations, grounding proposed changes in evidence beyond the existing trace. Cost-aware harness evolution turns these diagnoses into reusable revisions and determines which revisions to retain by considering both answer accuracy and visual cost. Across our evaluation settings on video understanding benchmarks, the evolved agent improves accuracy while processing fewer frames and achieves competitive accuracy-efficiency trade-offs against existing video understanding agents. These results demonstrate the potential for video understanding agents to improve their own evidence acquisition and use through harness evolution. Code is available at https://github.com/bingjunluo/Video-RSI . --- ### A shape-similarity latent space for fluid interfaces: invertible reduced-order modelling of droplet morphology Authors: Ali R. Hashemi, Mohammad R. Hashemi, Pavel B. Ryzhakov Published: 2026-09-29 Categories: physics.flu-dyn, physics.comp-ph, physics.data-an, stat.ML arXiv: https://arxiv.org/abs/2609.37947v1 A droplet breaks up in tens of microseconds, and a recording captures perhaps a dozen frames. The states in between cannot be recovered without repeating the experiment, and simulating them is too costly to sweep an operating envelope. Yet they are present in the corpus as a whole: a campaign spanning a device's actuation range produces morphologies resembling those any single recording missed. Exploiting that requires a representation that is low-dimensional, invertible, and faithful to shape rather than to sampling. Proper orthogonal decomposition supplies the first two but measures distance in sampled coordinates; manifold learning supplies the third but no map back to a shape; elastic shape analysis supplies a shape metric but no reduced coordinates. SHROM composes all three. Interfaces are represented by their square-root velocity functions, a neighbour graph is built over that shape space, and an autoencoder is trained to reconstruct while penalising latents in which graph neighbours are not latent neighbours. The demonstration uses 301,539 inkjet droplet contours and four filmed break-up sequences. The graph term does not improve reconstruction. It determines whether position in the latent carries meaning: clustering the latent of an otherwise identical model recovers the shape-space partition at chance level (ARI = 0.063 +/- 0.059), and at 0.781 +/- 0.049 with the term active. Waveform parameters predict the full contour at R^2 = 0.878 +/- 0.026. Negative results are reported in the same terms. The graph metric proved immaterial across five choices, and interpolation error saturates at the reconstruction limit, so a plain autoencoder leads that task. The main limitation is interpolation across a topology change: no component of the regularised latent holds both a single-component and a post-break-up shape, whereas an unregularised autoencoder mixes them freely. --- ### Identifiability Guarantees for Drivers and Dynamics of Delayed Physical Systems Authors: Julien Boussard, Antoine Débouchage, Théo Saulus Published: 2026-09-29 Categories: stat.ML, cs.LG, math.DS arXiv: https://arxiv.org/abs/2609.37944v1 A wide range of methods have been proposed, including physics-informed neural networks, which are powerful but do not guarantee identifiability of the dynamics, symbolic regression, which requires a set of precomputed operations, and causal discovery, which is more principled but usually relies on strong assumptions that physical systems may violate. In this work, we develop a theory-grounded method and prove that under a set of permissive assumptions, the structural drivers and drift of stochastic delayed differential equations are identifiable. Our method outperforms others on a benchmark for driver identifiability, and on a second benchmark to evaluate physical consistency of the learned dynamics. --- ### An Efficient Machine Learning Approach for Degradation Forecasting in AEM Water Electrolysis Authors: Marco Veneriano, Ani Gjergji, Sebastiano Bellani, Andrea Riva, Vito Paolo Pastore, Matteo Santacesaria Published: 2026-09-29 Categories: cs.LG, eess.SY, physics.app-ph arXiv: https://arxiv.org/abs/2609.37941v1 This study provides a data-driven analysis of a novel dataset of single-cell Anion Exchange Membrane water electrolyzers (AEMWE), operated under constant current load across multiple heterogeneous experimental campaigns. We train and evaluate a range of machine learning models with different complexity, including linear baselines, LSTMs and CNNs, to perform medium-term forecasting of the cell voltage degradation curve. The models are assessed within a rigorous training and evaluation framework specifically designed for heterogeneous industrial data. --- ### Does Local Video Understanding Transfer Across Encounters? The EgoGears Benchmark Authors: Yuedong Tan, Lei Qi, Yu Liu, Di Wen, Ruiping Liu, Xiaoye Wang, Yufan Chen, Junwei Zheng, Chengzhi Wu, Chen Zhang, Zhihang Chen, Haiwen Sun et al. Published: 2026-09-29 Categories: cs.CV, cs.AI, cs.RO arXiv: https://arxiv.org/abs/2609.37938v1 Embodied systems must make knowledge acquired during one encounter usable in another despite changes in viewpoint, motion, and illumination. Yet aggregate cross-video accuracy conflates failures of local perception with failures to preserve observation identity, establish correspondence, and compose evidence, obscuring whether local video understanding actually transfers. We introduce EgoGears, a complementary single- and multi-video benchmark designed to diagnose this transition. It contains 567 single-video and 1,487 multi-video questions derived from 126 human-collected egocentric recordings covering 39 outdoor routes. Repeated traversals across movement speeds and lighting conditions ground comparisons in shared physical environments; 531 questions require alignment across independent recordings. Single-video questions measure the local visual, spatial, and motion evidence available to a model, while multi-video questions test whether evidence remains bound to the correct observation and can be composed into consistent route relationships. We report 29 single-video and 31 multi-video MLLM configurations across six model families in the main leaderboard. Among the 20 configurations evaluated comparably on both splits, every model performs worse on multi-video questions, with a mean decrease of 22.5 percentage points, and the gap persists when answer format and scoring are held fixed. The gap is not explained simply by additional videos or recording boundaries. The central bottlenecks are observation--evidence binding and ordered route-state tracking. The code and benchmark are publicly available at https://github.com/lei-qi-233/EgoGears. --- ### Look Closer: Patch-wise Supervision for AI-Generated Image Detection Authors: Zhida Zhang, Tao Wu, Siyu Liu, Jie Cao Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37937v1 How much of an image does a detector need to see? Small RGB regions can retain useful evidence of image synthesis even when they reveal little of the full scene. Motivated by single-patch detection, we study patch-wise supervision: a shared backbone classifies explicit crops, each crop receives its own loss, and patch probabilities are averaged only at inference. The procedure requires neither handcrafted residual filtering nor a learned image-level fusion module. Experiments span single-patch selection, multiple generator collections, and four CNN and Transformer backbones. On GenImage, the reported patch-wise variants improve average accuracy over their whole-image counterparts across all four backbones. Comparisons of supervision granularity, source resolution, crop size, and inference coverage further characterize the approach, while post-processing tests and difficult-image evaluation reveal its limitations. The historical experiments include evaluation-based model selection, so their scores are not presented as a uniformly selected leaderboard comparison. Overall, the study identifies explicit local input and patch-level supervision as a simple, useful combination for investigating generalizable AI-generated image detection. --- ### Post-Anomaly Detection Inference for Deep SVDD Authors: Cao Le Cong Thanh, Dang Quang Vinh, Vo Nguyen Le Duy Published: 2026-09-29 Categories: stat.ML, cs.LG arXiv: https://arxiv.org/abs/2609.37935v1 Deep Support Vector Data Description (Deep SVDD) has become a prominent framework for unsupervised anomaly detection by learning latent representations that compactly characterize normal data around a center. Despite its empirical success, anomaly decisions produced by Deep SVDD are typically made solely based on anomaly scores without rigorous statistical guarantees, thereby limiting their reliability in safety-critical and high-stakes applications where false positives must be strictly controlled. In this paper, we propose PADI (Post-Anomaly Detection Inference), a novel framework that equips a trained and frozen Deep SVDD detector with statistically valid inference by leveraging the Selective Inference framework. Specifically, PADI performs inference conditional on the event that a test instance is identified as anomalous by Deep SVDD, thereby enabling rigorous statistical assessment of anomaly decisions. Based on this formulation, we derive valid selective p-values that quantify the statistical significance of the detected anomaly. Using these p-values, we theoretically establish control of the false positive rate (FPR) at a user-specified significance level $α$ (e.g., $α=0.05$). Furthermore, we extend the proposed framework to Deep Semi-Supervised Anomaly Detection (Deep SAD), providing a principled approach for statistically reliable inference in semi-supervised anomaly detection settings. Extensive experiments on both synthetic and real-world benchmark datasets robustly support the theoretical findings. The results demonstrate that PADI consistently achieves proper FPR control while attaining superior true positive rates compared with existing approaches. --- ### GRFBrain: Graph-Structured Rectified Flows for EEG Dynamic Modeling Authors: Haohui Jia, Zheng Chen, Jathurshan Pradeepkumar, Xu Cao, Yasuko Matsubara, Yasushi Sakurai, Takashi Matsubara Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37934v1 Forecasting time-varying functional connectivity from electroencephalography (EEG) requires modeling both history-dependent trends and structured variability across channels. Conditional flow matching provides a framework for distributional forecasting, yet it remains unclear whether graph-informed source distributions offer practical advantages over isotropic noise and strong deterministic predictors. We introduce a graph-structured residual flow framework that separates conditional mean prediction from stochastic residual transport. A history-only predictor estimates the future connectivity graph, while a graph Gaussian source encodes dependencies derived from past connectivity through a Laplacian-based covariance. A conditional velocity field transports source samples to future graph residuals, with transport time explicitly distinguished from physical EEG time. Our study identifies the conditions and controls needed to distinguish useful residual transport from improvements attributable to deterministic prediction, learned representations, and sampling effects. --- ### Learning When to Update: A Near-Optimal Timing Bandit Approach Authors: Qiulin Lin, Junyan Su, Liyuan Wang, Minghua Chen Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37932v1 Systems operating in dynamic environments require timely updates to sustain performance. For resource-intensive systems such as machine learning models and digital twins, strategically timing updates is essential. Updating too frequently wastes resources, while updating too infrequently leads to costly performance degradation. The problem is particularly challenging when the system's degradation pattern is unknown a priori, as is common in new operating environments. We formalize this challenge as a novel \emph{timing bandit} problem, where each arm represents a candidate update interval with a fixed update cost and an unknown, stochastic degradation cost. Three structural properties distinguish this setting from standard multi-armed bandits: selecting an interval commits the learner to multiple time slots before the next update; arm costs are composed of per-step degradation costs and a fixed update cost; and selecting a longer interval naturally reveals degradation at every intermediate step, providing consecutive feedback relevant to shorter intervals. By exploiting these structures, we develop Balanced Consecutive Arm Elimination (BCAE). BCAE achieves $\tilde{O}(\sqrt{T})$ regret, improving upon the $\tildeΩ(K\sqrt{T})$ regret of standard bandit algorithms in this setting, where $K$ is the number of candidate update intervals. We further propose an Optimism-Enhanced variant (OE-BCAE) that integrates lower-confidence-bound principles to improve empirical adaptivity while preserving the same regret order. Moreover, the regret bound achieved by our algorithms matches the theoretical lower bound up to logarithmic factors. Simulation results demonstrate that our algorithms achieve low regret and remain stable as both the number of arms and the update cost vary. --- ### Learning What to Remember: Long-horizon Counterfactual Memory Optimization Authors: Jiaming Tang, Mingyan Liu, Armin Sarabi Published: 2026-09-29 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.37930v1 Persistent textual memory allows language models to carry information across long interactions, but learning what to remember is fundamentally a credit-assignment problem. A memory rewrite may only become useful many steps later, while much of the observed utility may be inherited from information already stored before the rewrite. We introduce Memory Gain Policy Optimization (MGPO), which isolates the incremental value of each memory rewrite by crediting it for its marginal contribution to current and future downstream utility. This turns delayed memory utility into a direct learning signal for optimizing what information should persist. We study MGPO on document-level information extraction, where structured supervision makes the effects of individual memory updates directly measurable. MGPO improves extraction while reducing average memory length by nearly 80% relative to the initial memory policy before optimization. The learned memory policy also supports reuse and transfer across domains, downstream models without further training. These results show that effective memory learning depends not only on preserving useful information, but on identifying which memory updates create lasting incremental value. --- ### Rollout-Marginal Distillation for Long-Horizon Autoregressive Video Generation Authors: Chenjian Gao, Zhihao Hu, Jianqi Ma, Jun Zhang, Weidong Zhang, Tianfan Xue Published: 2026-09-29 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.37925v1 Autoregressive (AR) video diffusion enables low-latency, streamable video generation, but prediction errors often accumulate over long rollouts. Training the generator on its own rollouts exposes it to these imperfect histories. However, existing video-level distribution matching distillation (DMD) scores the whole rollout jointly. Because a chunk is evaluated together with its past and future, its correction can favor matching artifacts in the surrounding context merely to preserve temporal consistency. To provide a clearer visual-quality signal, we introduce Rollout-Marginal Distillation (RMD). RMD retains the generated history for AR prediction but scores each chunk independently against a chunk teacher, ensuring its quality correction is not compromised by an imperfect temporal context. To compensate for the lack of temporal context in independent chunk scoring, RMD subsequently applies video-level DMD to restore temporal coherence. Extensive experiments demonstrate that RMD maintains high visual quality far beyond its training horizon and outperforms video-level DMD baselines. Code and video results are available at https://cjeen.github.io/RMD --- ### Time-Anchored Diffusion Language Models: Latent-Space Caching for Fast Generation Authors: Joel Anto Paul, Litu Rout, Aditya Akella, Sanjay Shakkottai Published: 2026-09-29 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2609.37924v1 Recent work on anchored diffusion language models improves denoising by shaping an intermediate latent space with supervised important-token targets. In this work, we introduce time-based (self-supervised) anchoring, which learns and reuses latent anchors without requiring such targets. Our key observation is that anchors encode persistent properties of the clean sequence, such as its semantic intent, global structure, or intermediate plan. Although their hidden representations become stale as the token canvas evolves, their semantic content remains useful across nearby diffusion times. This is implemented through a two-stage architecture consisting of a relatively expensive anchor network that generates the latent cache state and a lightweight denoising network that intelligently combines the cached latent state with the current state at each reverse step using a fusion module. This gives anchoring a latent-space caching interpretation: the anchor network is evaluated periodically, while its cached representation is reused across multiple reverse steps. We instantiate this framework as TADM:Post-train, which time-anchorizes pretrained DLMs, and TADM:Pretraining, which learns time-based anchors during pretraining. Applied to DiffusionGemma-26B, TADM:Post-train improves throughput by approximately 49% to 79% on several math, code, and STEM benchmarks (GSM8K, AIME26, GPQA-Diamond, LiveCodeBench-v6, HumanEval, MMLU-Pro). TADM:Pretraining reduces Transformer-layer computation by up to 38% relative to a standard single-stage DLM, achieves up to 73% higher measured throughput than ADLM. --- ### EpiCon: Collective Agent Learning through Co-Evolving Multimodal Memory Authors: Ziyun Zeng, Hang Hua, Shaden Alshammari, Rogerio Feris, William T. Freeman, Jiebo Luo Published: 2026-09-29 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.37923v1 Agents can learn from past executions, but enabling different agents to reuse and build on one another's experience remains challenging. We introduce EpiCon, a shared multimodal memory framework for agent collective learning without updating host model parameters. EpiCon links question-level memory evolution to a persistent experience bank through two independently trained 2B models: a memory controller and a tree self-organizer. The controller jointly refines textual guidance and visual evidence across attempts and selectively includes visual memory. The self-organizer consolidates lessons hierarchically and retrieves experience and rules for new problems. We evaluate EpiCon on eleven benchmarks spanning four multimodal task domains, using two harnesses and multiple backbones. A frozen bank improves other systems even with a single solving attempt. A second harness raises the original system's macro-average score by 2.6 points across eleven benchmarks. Across four host configurations, EpiCon improves macro-average scores by 1.7 to 4.9 points over No Memory and reduces memory-operation time by 67\% to 74\% relative to backbone-sized memory models. --- ### Pattern Formation in Transformers Authors: Erkan Turan, Gaspard Abel, Maks Ovsjanikov Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37921v1 What are the inductive biases of a Transformer architecture? Existing theory on how the forward pass shapes representations either considers whether Transformers escape from rank collapse or demonstrates that self-attention drives tokens toward cluster patterns. The latter view arises from an elegant dynamical systems perspective, but relies on simplified architectural assumptions, and does not explain the rich structures observed in practice. This leaves a major open question: when a full Transformer escapes rank collapse, how does it structure token representations? Using pattern-formation theory, we show that the dynamical view of Transformers can account for Positional Encoding, Multi-Head Attention, and Output-Value geometry. We demonstrate that a full Transformer architecture imposes an inductive prior by selectively amplifying a rich set of previously unreported patterns, including traveling or rotating waves among others. We characterize the role of each architectural component in controlling which pattern is amplified, which ones stabilize, compete, or coexist. Finally, we show that these structures can act as a controllable dynamical prior that facilitates learning. By choosing both task-aligned positional encoding and weight initialization, we demonstrate improved data efficiency and accelerated optimization on controlled sequence tasks and with ConViT on CIFAR-10. --- ### SYNCR: Diagnosing and Learning Cross-Video Reasoning from Simulation Authors: Sara Ghazanfari, Siddharth Garg, Prashanth Krishnamurthy, Farshad Khorrami Published: 2026-09-29 Categories: cs.CV, cs.LG arXiv: https://arxiv.org/abs/2609.37918v1 Reasoning across videos requires aligning events, matching identities, comparing motion, and integrating partial observations. Evaluating these capabilities and testing how to improve them requires both reliable labels and targeted supervision. We introduce SYNCR, a simulator-grounded framework that connects these two needs through shared task generators. Built on Habitat, Kubric, and CLEVRER, SYNCR derives answers from environment state and provides 4,000 evaluation questions and 15,960 training questions over disjoint videos, spanning eight cross-video reasoning tasks. Visual ablations and human evaluation assess dependence on the supplied evidence and answer recoverability. Evaluation of 22 multimodal large language models reveals persistent difficulties in physical comparison and scene integration that increasing model size does not consistently resolve. Supervised fine-tuning raises Qwen3-VL-8B's average SYNCR accuracy from 32.6% to 61.6%, with gains extending to task configurations and video sources absent from training for those tasks. Transfer to real footage is most consistent for temporal ordering: accuracy improves by 9.0-20.5 percentage points on constructed Assembly101 and Panoptic ordering sets across three checkpoints spanning two model families and two model sizes, with additional gains on existing temporal reasoning benchmarks. These results establish SYNCR as a controlled setting for diagnosing cross-video reasoning failures, testing their learnability, and identifying where synthetic supervision transfers. --- ### Search Dimension in Unlabeled Projection Pursuit: A Scaling Law for Subspace Restriction Authors: Rares Dimitrie Grozavescu, Mark Girolami Published: 2026-09-29 Categories: cs.LG, math.ST, stat.ML arXiv: https://arxiv.org/abs/2609.37917v1 Projection pursuit searches for a direction along which the data look least Gaussian. When the observation space contains a large Gaussian complement, the empirical objective can be minimized by a direction that carries no signal, with empirical kurtosis as low as at the truth. Sample splitting exposes rather than repairs this failure. Appending coordinates independent of the latent regime degrades the search while leaving Bayes recoverability unchanged. Restricting the search to the column space of a known forward operator removes the failure exactly on the negative-kurtosis branch. Estimating a principal subspace from the data is the alternative. In a controlled two-component model, the leading sufficient scalings differ in the gain with which the operator transmits the discriminant: $ς^{-4}$ for covariance-spike estimation and $ς^{-8}$ for fourth-moment search. At fixed search dimension, the measured threshold ratio collapses onto $n/p^2$ with exponent $0.156$, close to the predicted $1/8$. This is an empirically supported scaling motivated by sufficient bounds, not a proved asymptotically tight law. When the search dimension is varied, the measured exponent is $0.325$, substantially larger than $1/8$, and the tested range does not identify its functional form. The crossing location also depends on calibration and model configuration. Under a downstream excess-error criterion, the scaling largely disappears. --- ### RLX: A Unified Multi-Backend Tensor Compiler and Distributed Runtime in Rust Authors: Eugene Hauptmann, Nataliya Kosmyna Published: 2026-09-29 Categories: cs.DC, cs.AI arXiv: https://arxiv.org/abs/2609.37916v1 Production machine learning (ML) stacks often split graph compilation and kernel execution across different layers and languages, making backend behavior, deployment guarantees, and performance fallbacks hard to reason about end-to-end. RLX addresses this gap with a single Rust codebase that combines compiler and runtime roles around one primitive-level, three-level intermediate representation (IR), plus a transparent dispatch contract that resolves each operator to native, common-IR, or rewritten lowering and fails compilation when legalization is not possible. The same IR targets fourteen runtime devices (cpu, metal, mlx, ane, cuda, rocm, oneapi, tpu, hexagon, gpu, vulkan, opengl, directx, webgpu) and two specialty codegen paths (Cortex-M INT8 and FPGA), ingests safetensors, GGUF, ONNX, and rten formats, supports F16/BF16/F64/C64 and quantized INT4/INT8 flows with AMP/PTQ/QAT, and scales via tensor-/pipeline-parallel collectives over TCP and RDMA transports. Beyond neural workloads, RLX also extends to scientific/physics-style domains through sparse and dense linear algebra extensions (e.g., CSR LU/CG/matvec and LAPACK- backed factorizations) and 3D Gaussian splatting operators. We evaluate RLX against PyTorch, TensorFlow, JAX, candle, burn, tch, rten, MLX, CoreML, IREE, Glow, TensorRT, and tinygrad under identical input generation and p50 measurement methodology on one host. On all-MiniLM-L6-v2, RLX-Metal is fastest at every batch (e.g., 16.6 ms at batch 32 vs. PyTorch-MPS 26.7 ms). In the MNIST training table, RLX also has the top-throughput entry (graph-fused MLP: 946,487 img/s), above NumPy+BLAS (787,349 img/s), while retaining 100% top-1 parity on reference checks (e.g., Qwen3). --- ### Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning Authors: Md. Ismail Hossain, Humaira Kousar, Isidora Chara Tourni Published: 2026-09-29 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2609.37915v1 On-policy self-distillation (OPSD) trains a student to match a privileged teacher distribution along its own sampled trajectory. Standard OPSD applies this supervision to unverified student rollouts while conditioning the teacher on privileged context, typically a reference solution. We separate these roles in a factorial analysis and find that scaffold correctness has a stronger effect on downstream accuracy than context correctness. Unverified scaffolds create an imitation gap because the teacher can use information unavailable to the student. This gap shrinks with model scale, yet OPSD continues to supervise mostly unverified trajectories. In contrast, verified scaffolds remain effective even when the teacher is conditioned on the student's own unsuccessful rollout. Based on this finding, we introduce OASIS, which retains the OPSD objective but supervises mostly verified by label on-policy trajectories and replaces written solutions with unverified model-generated attempts as the teacher context. OASIS therefore requires only final-answer labels. Across Qwen3-1.7B, 4B, and 8B on AIME 2024, AIME 2025, and HMMT 2025, OASIS improves over the base model by 3.2--3.8 points on average, while OPSD's gain falls from 3.05 points at 1.7B to 0.14 at 8B. At 8B, OASIS improves over OPSD by 3.05 points, showing that verified on-policy scaffolds preserve the effectiveness of self-distillation as models scale. --- ### The Unequal Influence of Bad Advice: Using Training Data Attribution to Modulate Emergent Misalignment Authors: Gonçalo Paulo, Louis Jaburi, Nora Belrose, Lucia Quirke, Stella Biderman Published: 2026-09-29 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.37914v1 Fine-tuning large language models on narrow, misaligned tasks can undo their post-training alignment and induce novel misaligned behaviors -- a phenomenon known as \emph{emergent misalignment} (EM). EM has been linked to persona-like representations, where fine-tuning might reduce loss by amplifying a harmful or 'evil' persona. It remains unclear which properties of the training data drive this effect: whether all harmful examples contribute approximately equally to misalignment and whether different models are equally affected by the same fine-tuning examples. In this work, we use training data attribution to quantitatively estimate how much each harmful example contributes to EM. We benchmark the quality of the attribution via retraining -- a sound attribution score should enable us to enhance or attenuate EM by filtering data on that score. Score-based filtering can substantially enhance or attenuate EM; we find that both data-attribution scores and a black-box harmfulness score can identify consequential examples. All models we test become misaligned when trained on the same dataset, and influence scores perform best when filtering data from the same model that computed them. We find cross-model generalization of influence scores from scores derived from the three model families we tested, but this generalization does not recover same model filtering performance. --- ### Generated Query Expansion Still Helps Strong Sparse Retrieval: A Controlled Study with SPLADE-v3 Authors: Ryan C. Barron, Cade W. Trotter, Maksim E. Eren, Kim Ø. Rasmussen, Liz D. Miller, Benjamin J. Migliori Published: 2026-09-29 Categories: cs.IR, cs.AI arXiv: https://arxiv.org/abs/2609.37911v1 Scientific queries are often brief, while relevant papers use specialized vocabulary. Generated query expansion can bridge this mismatch, but earlier work suggests that its value shrinks as the underlying retriever becomes stronger. We test the four generated formats of term lists, a pseudo-document, multiple pseudo-references, and corpus-steered text all together with SPLADE-v3 on NFCorpus, TREC-COVID, and SciDocs. Every condition searches the same frozen document index and follows the same query-side integration rule and 256-dimension budget, isolating the effect of the added content. All twelve method-collection comparisons improve aggregate nDCG@10, with best relative gains of 4.81%, 8.92%, and 9.47%. Eleven remain significant after Holm correction. The gain persists in 103 of 114 interpolation settings, including every setting that assigns at least 30% of the mixture weight to the original query. Shuffled-text and non-contextual lexical-bag controls also remain above baseline in all 24 aggregate comparisons, showing that the added vocabulary carries most of the benefit. A corpus-induced typed concept graph, by contrast, produces no consistent gain, and its relation, depth, validation, random, and gating controls do not rescue it. Generated vocabulary can therefore complement a strong learned sparse retriever, provided that the original query remains strongly represented. --- ### Pixels to Keys: Exploring Spatial and Motion Cues in Gameplay Inverse Dynamics Authors: Abhishek Pillai, Ekta Prashnani, Joohwan Kim, Iuri Frosio Published: 2026-09-29 Categories: cs.AI, cs.CV arXiv: https://arxiv.org/abs/2609.37907v1 Video games offer scalable environments for studying perception and control in embodied agents.Abundant online gameplay videos could supply demonstrations, but they rarely include player inputs for training. Inverse Dynamics Models (IDMs) have thus been proposed to infer inputs from frames. Large (up to 1B parameters) IDMs trained on $\sim$1K-2K gameplay hours demonstrate feasibility and cross-environment generalization at this scale, but researchers do not clarify what the key components are to recover individual actions and often report only aggregate accuracy that can mask rare-action failures. We study the problem in a data-constrained scenario to evaluate how spatial motion features, model architectures, and training objectives affect an IDM's outcome and we analyse our models on per-key and balanced metrics such as $F_1^{macro}$. Our experiments on Trackmania highlight the importance of factors like the model architecture and motion flow extraction in preprocessing, while also showing the limits of evaluation through unbalanced metrics. The application of the same architecture and training recipe to Cyberpunk 2077 reveals uneven performance across game mechanics. Our per-action evaluation and failure analysis highlight ambiguities from camera motion, delayed effects and imbalanced key-press frequencies that call for explicit modeling of 3D scene structure, long-term state and the adoption of proper losses in future implementations. --- ### Beyond Interaction Capacity: Estimator Scaling with Recursive Models for CTR Prediction Authors: Shivang Chopra, Fotis Iliopoulos, Zsolt Kira, Gaurav Menghani Published: 2026-09-29 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.37905v1 Click-Through Rate prediction, a core task in recommendation and advertising systems, relies on modeling interactions among sparse categorical features. Explicit cross networks are a central paradigm for CTR prediction, and recent progress has largely come from increasing the interaction capacity of a single predictor through deeper cross networks and more expressive cross operators. We revisit whether continually increasing interaction capacity remains the most effective way to improve predictive performance, and find that its benefits quickly exhibit diminishing returns even as capacity continues to grow. This motivates a complementary scaling direction that we call estimator scaling, where additional resources are used to incorporate multiple related estimators rather than only enlarging a single predictor. Through theoretical analysis, we show that the gains from estimator scaling are governed by the amount of non-shared predictive variation available across estimators. However, exploiting this variation naively can be expensive: independently trained models provide substantial estimator diversity but require deployment cost to grow with ensemble size. This motivates a parameter-efficient realization of estimator scaling that can incorporate diversity from multiple estimator sources without maintaining multiple full models. Building on this view, we introduce RECursive Averaged Predictor (RECAP), a parameter-efficient recursive CTR model that operationalizes estimator scaling at three levels: distillation across independently trained models, exponential moving averaging over training trajectories, and aggregation over inference-time routes within a weight-shared recursive backbone. Experiments across multiple benchmarks establish new state-of-the-art predictive performance on standard benchmarks, while placing the RECAP on a favorable performance-parameter Pareto frontier. --- ### You Cannot Pick a Provider From the Price List: Market-Aware Routing for Open-Weight LLM Inference Authors: Liang He, Jingbo Wen, Yixiong Chen, Yue Yang, Qizhen Lan, Kangning Cui, Xilu Wang Published: 2026-09-29 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.37902v1 Existing LLM routers choose among models using static per-model costs. We show that open-weight inference markets introduce a second, largely ignored decision axis: after choosing a model, a client must still choose which provider serves it. Measuring live endpoints across (nummodels) open models, competing providers, multiple task types, and three measurement waves, we find that provider choice cannot be inferred from the price list. The same model can vary sharply in quality, latency, availability, and price across providers; higher-priced providers are consistently faster, but price does not reliably predict quality or availability; and provider feasibility is task-selective, with one deployment nearly normal on knowledge tasks but catastrophically degraded on multi-step reasoning. We formulate same-model provider selection as a price-taker market-aware routing problem. A simple measured-map policy routes to the cheapest provider that is both quality-equivalent and healthy, yielding matched-quality savings while avoiding degraded endpoints. Because the map drifts, we introduce FACET, an online provider router that certifies per-(provider x task) feasibility facets and fails safe to an anchor before serving uncertified arms. Across relaxed deployment assumptions, FACET tolerates imperfect task assignment and sparse feedback, while systematic evaluator bias exposes a quality-signal trust boundary that can be mitigated with ground-truth probes or audits. Live provider runs further confirm that certification can move real traffic from a premium anchor to a substantially cheaper certified endpoint. Our results suggest that market-aware LLM routing must measure not only which model to use, but also who serves it. --- ### Scaling Zero-Order Pretraining through Model Sharding Authors: Francois Chaubard, Mykel J. Kochenderfer, Chris Ré Published: 2026-09-29 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.37899v1 Zero-order optimization (ZO) trains without backpropagation, making it relevant to forward-only hardware and non-differentiable loss, but its gradient variance grows with perturbed dimension, inhibiting large-model training. Sharded Optimization Mixture of Assemblies (SOMA) trains LSTM experts independently on $N$ data clusters using simultaneous perturbation stochastic approximation (SPSA), without exchanging gradients, activations or optimizer state. Its separable loss removes cross-expert perturbation noise at the cost of jointly learned representations across domains. Using 80,000 estimated RTX 5090 GPU-hours, we show modest sharding improves training compute efficiency over all tested monolithic ZO controls. At 8.44M parameters and 150 aggregate GPU-hours, SOMA $N=2$ with 64 perturbations reaches 1.76 test nats/byte, versus 2.00--2.11 for monolithic SPSA at 64, 256 or 1,024 perturbations and 2.21 for EGGROLL. On WikiText-103, these frozen checkpoints reach 2.07, 2.25--2.36 and 2.49, respectively. On a fixed separable objective with equal-size blocks, we prove independent losses reduce relative gradient variance to approximately $1/N$ of a shared-loss estimator's. Holding starting weights, data, perturbations and compute fixed, independent rather than summed losses lower SOMA $N=4$ test loss by 0.035 nats/byte after 1,000 updates across three seeds. Larger ensembles offer a separate inference benefit: at similar model size with top-$k$ routing ($k=4$), SOMA $N=256$ achieves 2.36M tokens/s versus 257k for SOMA $N=8$ ($9.19\times$, including routing), at lower test loss (1.68 versus 1.71), albeit using $59.9\times$ as much aggregate training compute. We release all training and evaluation code and checkpoints. ---