ICML 2026 论文列表

Large language models excel at short-horizon reasoning tasks, but performance drops as reasoning horizon lengths increase. In this work, we introduce a scalable method to bootstrap long-horizon reasoning capabilities using only existing, abundant short-horizon data. Our approach synthetically composes simple problems into complex, multi-step dependency chains of arbitrary length. We train models on this data using outcome-only rewards under a curriculum that automatically increases in complexity, allowing RL training to be scaled much further without saturating. Empirically, our method generalizes remarkably well: curriculum training on composed 6th-grade level math problems improves accuracy on longer, competition-level benchmarks. It also transfers significantly to diverse out-of-distribution ReasoningGym domains and long-context benchmarks, indicating broader generalization. Importantly, our long-horizon improvements are significantly higher than baselines even at high pass@k, showing that models can learn new reasoning paths under RL. Theoretically, we show that curriculum RL with outcome rewards could achieve an exponential improvement in sample complexity over full-horizon training, providing training signal comparable to dense supervision. h1 therefore introduces an efficient path towards scaling RL for long-horizon problems using only existing data.

Memory-based test-time adaptation (TTA) assigns streaming test samples into class-specific memory slots based on pseudo-labels predicted by models like CLIP, and retrieves them to facilitate subsequent predictions under distribution shift. However, this process introduces two challenges: (1) Each sample is hard-assigned to a single class based on CLIP's prediction, where inaccurate CLIP prediction leads to memory contamination that biases subsequent prediction. (2) Samples are evicted under biased selection due to fixed memory capacity, which risks discarding informative samples and undermining the efficacy of the memory. To address these challenges, we propose A²Memory (Attributive-Associative Memory for Test-time Adaptation). For challenge (1), we propose Attribute-centric Memory Construction that builds prior textual representations from class-shared representative and diverse visual attributes, and applies soft assignment to generate surrogate visual representations. For challenge (2), we design Class-wise Associative Memory that dynamically compresses streaming samples into fixed-capacity memory through gradient-based optimization and data-dependent retention, then retrieves sample-adaptive class prototypes for reliable inference. Extensive experiments demonstrate consistent improvements over state-of-the-art methods across 15 benchmarks.

Video action models are a promising foundation for Vision–Language–Action (VLA) because they can learn rich visual dynamics directly from video. However, likelihood-oriented training of diffusion predictors emphasizes globally plausible futures and does not guarantee precision-critical visual dynamics needed for manipulation, so small prediction errors can be amplified by downstream policies. We propose Dyn-VPP, a post-training framework that casts multi-step denoising as policy optimization and aligns predicted future latents with expert visual dynamics via verifiable terminal reward, without modifying any architecture. This enables explicit optimization of dynamics signals that are not captured by likelihood-only training. As a result, Dyn-VPP yields more accurate visual dynamics and improves downstream task execution. Experiments across diverse simulated and real-world manipulation settings show improved dynamics consistency and consistently higher task success.

Full-Reference image quality assessment (FR IQA) is important for image compression, restoration and generative modeling, yet current neural metrics remain slow and vulnerable to adversarial perturbations. We present BiRQA, a compact FR IQA metric model that processes four fast complementary features within a bidirectional multiscale pyramid. A bottom-up attention module injects fine-scale cues into coarse levels through an uncertainty-aware gate, while a top-down cross-gating block routes semantic context back to high resolution. To enhance robustness, we introduce Anchored Adversarial Training, a theoretically grounded strategy that uses clean "anchor" samples and a ranking loss to bound pointwise prediction error under attacks. On five public FR IQA benchmarks BiRQA outperforms or matches the previous state of the art (SOTA) while running $\sim 3 \times$ faster than previous SOTA models. Under unseen white-box attacks it lifts SROCC from 0.30-0.57 to 0.60-0.84 on KADID-10k, demonstrating substantial robustness gains. To our knowledge, BiRQA is the only FR IQA model combining competitive accuracy with real-time throughput and strong adversarial resilience.

We study the problems of approximate sampling from and distribution testing of \emph{mixture models}, where the modes satisfy a functional inequality called \emph{approximate tensorization of entropy} (ATE). While it is known that ATE makes these tasks more efficient in the unimodal setting, mixtures of few distributions satisfying ATE do not necessarily satisfy ATE overall, leading to a lack of theoretical guarantees for multimodal distributions, which are a key challenging case of modern generative models. We show this gap can be overcome by establishing the following pair of results for mixtures of ATE distributions: 1) We show fast mixing of Glauber dynamics from a \emph{data-based initialization}, with \emph{optimal} sample complexity, for mixtures of distributions satisfying modified log-Sobolev inequalities, building on similar results in (Koehler et al., 2024, Huang et al., 2024) for mixtures satisfying the weaker Poincaré inequality. 2) Answering an open question from (Blanca et al., 2023), we give efficient identity-testers for mixtures of ATE distributions in the coordinate-conditional sampling access model.

Computational pathology has made progress in diagnosis and prognosis prediction from whole slide images (WSIs), yet pipelines still rely on patch-level feature extraction and aggregation, departing from the cell-centric reasoning used by pathologists. This gap limits sensitivity to micro-lesions and subtle changes, and current methods rarely provide a unified solution that supports both local and global tasks with trustworthy evidence. We propose Cello, a universal cell-wise feature aggregation framework for reliable pathology image analysis. Cello integrates cell-level representations into WSI modeling via protein-signal–supervised cell-wise learning, preserving fine-grained cellular cues under gigapixel constraints. For local tasks, Cello introduces a flexible prototype-based contrastive module for scalable, task-adaptive representation learning. For global tasks, Cello adopts a weakly supervised gated aggregation that can widely leverage WSI labels. Finally, a cell–local–global decision-route consistency objective dynamically aggregates cellular evidence and aligns local predictions with global outcomes, improving reliability and faithfulness. Trained with only hundreds to thousands of samples, Cello achieves performance gains of 3.0%~7.6% and outperforms SOTA pathology foundation models pretrained on tens of thousands of samples. Code is available at https://github.com/HengruiLou/Cello.

Recent progress in spatial reasoning with Multimodal Large Language Models (MLLMs) increasingly leverages geometric priors from 3D encoders. However, most existing integration strategies remain passive: geometry is exposed as a global stream and fused in an indiscriminate manner, which often induces semantic-geometry misalignment and redundant signals. We propose GeoThinker, a framework that shifts the paradigm from passive fusion to active perception. Instead of feature mixing, GeoThinker enables the model to selectively retrieve geometric evidence conditioned on its internal reasoning demands. GeoThinker achieves this through Spatial-Grounded Fusion applied at carefully selected VLM layers, where semantic visual priors selectively query and integrate task-relevant geometry via frame-strict cross-attention, further calibrated by Importance Gating that biases per-frame attention toward task-relevant structures. Comprehensive evaluation results show that GeoThinker sets a new state-of-the-art in spatial intelligence, achieving a peak score of 72.6 on the VSI-Bench. Furthermore, GeoThinker demonstrates robust generalization and significantly improved spatial perception across complex downstream scenarios, including embodied referring and autonomous driving. Our results indicate that the ability to actively integrate spatial structures is essential for next-generation spatial intelligence.

Vision-language models (VLMs) generate fluent causal explanations, but current evaluations cannot distinguish linguistic plausibility from faithful causal reasoning. We introduce a dual-probe methodology that isolates these properties. The Text-Only Probe measures linguistic quality. The Chain-Text Probe requires models to first generate explicit causal chains. The Abstraction Gap (AG) metric quantifies the normalized performance difference. Evaluating eight VLMs on CAGE (Causal Abstraction Gap Evaluation), a benchmark of 49,500 questions across 5,500 images spanning Pearl's causal hierarchy, we find seven models exhibit AG exceeding 0.50 with text scores of 6--8 but chain scores below 2.5. Fine-tuning on 45,000 chain-annotated examples fails to close the gap. However, one model achieves near-zero AG. The capability exists within current VLM architectures and depends on pretraining and architectural choices. CAGE provides a diagnostic tool for assessing faithful causal reasoning in VLMs.

Large Reasoning Models (LRMs) have recently achieved strong mathematical and code reasoning performance through Reinforcement Learning (RL) post-training. However, we show that modern reasoning post-training induces an unintended exploration collapse: temperature-based sampling no longer increases pass@$n$ accuracy. Empirically, the final-layer posterior of post-trained LRMs exhibit sharply reduced entropy, while the entropy of intermediate layers remains relatively high. Motivated by this entropy asymmetry, we propose Latent Exploration Decoding (LED), a depth-conditioned decoding strategy. LED aggregates intermediate posteriors via cumulative sum and selects depth configurations with maximal entropy as exploration candidates. Without additional training or parameters, LED consistently improves pass@1 and pass@16 accuracy by 0.61 and 1.03 percentage points across multiple reasoning benchmarks and models. Relevant code is included in the supplementary material and will made be fully public after this paper is accepted.

Agentic multimodal models have garnered significant attention for their ability to leverage external tools to tackle complex tasks. However, it is observed that such agents often meet premature interaction collapse, caused by two primary reasons: 1) the terminal reward often appending on the last token prevents the advantage from distinguishing trajectories with exploratory behavior; 2) excessively redundant context hinders the agent from absorbing useful feedback. To address these issues, we propose the Deepening Reasoning MMSearchAgent, the framework leverages the structural proximity to derive advantage signals from the whole rollout trajectories in an entire batch, such that trajectories of different lengths are further encouraged to be generated, even when containing the same correct answer. Additionally, differentiated gaussian rewards are employed to dynamically calibrate interaction tolerance, thereby ensuring information reliability and reducing redundancy. To support multi-turn interaction training, we have constructed a multi-step deep-reasoning dataset including 3602 high-quality QA pairs with at least 3 reasoning steps. Extensive experiments demonstrate that our method achieves state-of-the-art performance, outperforming the MMSearch-R1 by 8.4\% on FVQA-test.

Large Audio–Language Models (LALMs) are becoming essential as a powerful multimodal backbone for real-world applications. However, recent studies show that audio inputs can more easily elicit harmful responses than text, exposing new risks toward deployment. While safety alignment has made initial advances in LLMs and Large Vision–Language Models (LVLMs), we find that vanilla adaptation of these approaches to LALMs faces two key limitations: 1) LLM-based steering fails under audio input due to the large distributional gap between activations, and 2) prompt-based defenses induce over-refusals on benign-speech queries. To address these challenges, we propose **S**afe-**A**blated **R**efusal **Steer**ing (SARSteer), an effective inference-time defense framework for LALMs. Specifically, SARSteer leverages text-derived refusal steering to enforce rejection without manipulating audio inputs and introduces decomposed safe-space ablation to mitigate over-refusal. Extensive experiments demonstrate that SARSteer significantly improves harmful-query refusal while preserving benign responses, establishing a principled step toward safety alignment in LALMs. The codes and constructed datasets are released at https://github.com/linweiii/SARSteer.

Evaluation of socially unsafe content in spoken dialogues remains text-centric, missing prosody and transcription failures. We present LALM-as-a-Judge, which includes an open benchmark of 24,000 multi-turn spoken dialogues with one localized unsafe turn, generated out of 8 socially unsafe categories and 5 severity levels. We evaluate 6 large audio-language models (LALMs) as judges, open and closed-source, in text-only, audio-only, and multimodal setups by their sensitivity, severity-order specificity, and turn-position bias for socially harmful content in the dialogue. Results show that audio contributes non-lexical evidence beyond transcript semantics and that multimodal gains are not universal but can be text-anchored, balanced, conservative, and interfering, which we link to the audio pathway bottlenecks and fusion limits. We position the benchmark as diagnostic and derive practitioner guidance for model, modality, and prompts choices.

Optimizing pretraining data composition is pivotal for LLM generalization. While dynamic mixing outperforms static strategies by capturing evolving training dynamics, current methods fail to reconcile computational efficiency with sample efficiency and structural flexibility for diverse pipelines.We introduce \textbf{Actor--Critic Online Data Mixing (AC-ODM)}, which approaches data mixing from a reinforcement learning perspective with a parameterized policy that we theoretically prove to act as a dynamic linear surrogate maximizing the constructive interference of gradients. To enhance practical flexibility, AC-ODM supports two operational modes: (i) a \textbf{proxy mode} for fixed, pre-prepared corpora, where a policy learned on a small model is transferred to a larger target; and (ii) a \textbf{non-proxy mode} for direct end-to-end training from scratch without priors. Empirically, AC-ODM significantly outperforms prior methods in convergence speed and downstream accuracy across various architectures. On Pythia-1B, it reaches optimal validation perplexity using up to 66\% fewer training steps than competitive baselines, delivering a 27.5\% relative improvement in MMLU accuracy and a 2.23$\times$ higher pass@1 on HumanEval, all while incurring a virtually negligible ($~$0.4\%) per-step wall-clock increase and only 2\% additional memory overhead. Code is available at \url{https://github.com/DANG-ai/AC-ODM}.

While the ecosystem of Lean and Mathlib has enjoyed celebrated success in formal mathematical reasoning with the help of large language models (LLMs), the absence of many folklore lemmas in Mathlib remains a persistent barrier that limits Lean's usability as an everyday tool for mathematicians like LaTeX or Maple. To address this, we introduce MathlibLemma, a modular LLM-based pipeline for automated folklore-lemma mining: the discovery, formalization, and proving of reusable intermediate facts that mathematicians often take for granted but that are not always present in formal libraries. At its core, MathlibLemma proactively mines the missing connective tissue of mathematics. The pipeline produces a verified library of folklore-style lemmas, including 1,506 Lean-checked proofs that pass a proof-bypass screen; a small curated pilot subset has also been merged into Mathlib, providing external evidence that selected outputs can meet expert library standards. Leveraging this pipeline, we further construct the MathlibLemma benchmark, a suite of 4,028 non-trivial type-checked Lean statements spanning a broad range of mathematical domains. By transforming the role of LLMs from passive consumers to active contributors, this work takes a step toward AI-assisted expansion of formal mathematical libraries.

Best-of-N selection improves reasoning in large language models (LLMs) by allocating test-time compute to sample candidate trajectories, but it relies on reliable verification. Widely used proxies have complementary failure modes: logit-confidence signals can suffer from calibration collapse, where confidence becomes misaligned with correctness, while sample agreement can be costly or brittle. Instead, we analyze the model's latent dynamics during inference. Motivated by metastable dynamics in cognitive systems, we introduce Latent Velocity Entropy (LVE), a training-free metric that quantifies the temporal concentration of internal representation updates. Experiments on four reasoning benchmarks, AIME25, GPQA, MATH500, and BRUMO25, show that LVE-based selection remains informative when logit-confidence baselines fail, with the clearest gains on reasoning-dense mathematical tasks and more modest gains on knowledge-heavy GPQA. On MATH500, LogNorm-LVE reaches 92.2\% Pass@1 at $N=10$ and nearly matches 10-sample majority voting with only 3 samples. Cross-model, bootstrap, and length-controlled analyses indicate that LVE is not reducible to sequence length, while also clarifying the task-dependent limits of the signal. Code is available at https://github.com/ZopenDP/LatentVelocityEntropy.

Recent research has made growing efforts to leverage large language models (LLMs) for computer-aided design (CAD), a domain that demands advanced geometric and spatial reasoning across long operation sequence. However, existing studies remain limited in addressing complex modeling tasks that necessitate step-by-step reasoning, primarily due to the scarcity of high-quality CAD datasets and the absence of fine-grained evaluation frameworks. In response to these challenges, we introduce Op-CAD, the first large-scale, multi-modal dataset for operation-oriented CAD generation, encompassing four operation types and five modalities. Furthermore, we introduce a novel CAD parsing module together with a geometry-guided hierarchical annotation pipeline, which decomposes modeling sequences into discrete operations and substantially improves the annotation accuracy of Vision-Language Models (VLMs). Based on our dataset, we redefine the CAD modeling task by decoupling geometric and spatial perspectives and introduce a novel metric, Chamfer/Fillet Intersection over Union (CF-IoU), to fill the void in assessing chamfer and fillet operations. By comprehensively evaluating eight LLMs on Op-CAD, we establish a benchmark for current models on operation-oriented tasks. Finally, we investigate performance enhancement strategies through fine-tuning on Op-CAD and propose Chain-of-Operation (COOP), a novel prompting strategy that emulates human-engineer reasoning.

Post-training with explicit reasoning traces is common to improve the reasoning capabilities of Multimodal Large Language Models (MLLMs). However, acquiring high-quality reasoning traces is often costly and time-consuming. Hence, the self-improvement paradigm has emerged, enabling MLLMs to self-generate reasoning traces for training without external supervision. Despite its effectiveness, we reveal two shortcomings in the self-improvement training of MLLMs: 1) data imbalance, where simple samples are over-trained, but the challenging yet crucial samples are under-trained; 2) language prior bias, where MLLMs overly rely on linguistic priors while neglecting the visual cues. To this end, we propose VISTA, a VIsion-aware Self-improvement Training framework for enhancing the multimodal Reasoning of MLLMs. Specifically, VISTA first introduces a prefix resampling strategy to reuse the partial correct reasoning traces for efficient data collection, and then designs a vision-aware attention score to quantify the model’s focus on visual information. Extensive experiments show that VISTA can be applied to various post-training scenarios, i.e., supervised fine-tuning and preference learning, and effectively enhances the multimodal reasoning performance across various MLLMs and tasks, e.g., bringing up to +13.66% average performance gains for Qwen2.5-VL-3B-Instruct.

The annealing phase is a pivotal convergence stage in LLM pre-training that ultimately determines final model quality. However, effectively selecting training data during this phase remains a key challenge. Current strategies rely on empirical heuristics, such as domain filtering or context extension, which lack a principled grounding in optimization theory. In this work, we characterize the annealing phase through the lens of the loss landscape's spectral geometry. We argue that optimal convergence requires gradient updates to satisfy heterogeneous constraints across different eigen-directions. Building on this insight, we formulate data selection as a problem of satisfying these directional constraints. To this end, we propose **DiReCT** (**Di**rectionally-**Re**strained **C**onstrained **T**raining), a novel framework that reformulates sample selection in the annealing stage as a constrained optimization problem. By imposing explicit directional constraints on per-sample gradients based on the spectral properties of the Hessian, **DiReCT** identifies samples that align with the optimal curvature-aware descent path. Extensive experiments across various model scales demonstrate that **DiReCT** consistently achieves state-of-the-art performance. For future research, code is available at https://github.com/xuyj233/Direct.

Accurately modeling complex temporal and topological dependencies and depth information is critical for monocular 3D human pose estimation, yet existing Mamba-based approaches struggle to fulfill these demands, suffering from internal state update confusion induced by forced sequence flattening and depth modeling confusion inherent to single-view observations. To address this confusion, we propose a StructMamPose framework equipped with Identity Anchoring Mechanism (IAM) and View Transformation Hub (VTH). The IAM injects spatiotemporal identities into the parameter generation network to anchor the selectivity of state update matrices, suppressing spurious connections to enforce feature propagation along valid topological dependencies. The VTH performs internal coordinate rotation to transform implicit depth inference into observable planar features, endowing the model with explicit spatial understanding and multi-view constraints. Experimental results demonstrate that our framework achieves SOTA performance on most datasets.