ICML 2026 论文列表

Automatic verification is a critical component in building math-solving agents and reinforcement learning, yet it often falls short in generalizability, performance, and cost-efficiency. Identifying that the primary bottleneck of verification lies in error detection capability, we propose pessimistic verification, a paradigm of agentic workflows that rejects a solution if any of multiple parallel verifiers identifies a flaw. We further introduce progressive pessimistic verification, which employs fine-grained proof decomposition to significantly enhance verification accuracy and efficiency. Our approach surpasses the performance and token efficiency of extended long chain-of-thought (long CoT) and mainstream verification workflows, crucially, our analysis reveals that existing benchmarks underestimate its effectiveness on stronger models due to inherent annotation errors. To further validate the effectiveness of our method, we applied a verification-based solving workflow on the IMO 2025 and MathArena Apex 2025 datasets, where the workflow with progressive pessimistic verification exhibits remarkable improvements in both efficiency and accuracy on highly challenging contest-level math problems with state-of-the-art models.

Rapid progress in video models has largely focused on visual quality, leaving their reasoning capabilities underexplored. Video reasoning grounds intelligence in spatiotemporally consistent visual environments that go beyond what text can naturally capture, enabling intuitive reasoning over spatiotemporal structure, such as continuity, interaction, and causality. However, systematically studying video reasoning and its scaling behavior is hindered by the lack of large-scale video reasoning training data. To address this gap, we introduce the **Very Big Video Reasoning (VBVR) Dataset**, an unprecedentedly large-scale resource spanning *200* curated reasoning tasks following a principled taxonomy, and over *one million* video clips, making it approximately *three orders of magnitude* larger than existing datasets. We further present **VBVR-Bench**, a verifiable evaluation framework that moves beyond model-based judging by incorporating rule-based, human-aligned scorers, enabling reproducible and interpretable diagnosis of video reasoning capabilities. Leveraging the VBVR suite, we conduct one of the first video reasoning **scaling studies** and observe early signs of emergent generalization to unseen reasoning tasks. Together, VBVR lays a foundation for the next stage of research in generalizable video reasoning. The data, benchmark tool kit, and models are released publicly at **video-reason.com**.

Deep forecasting models often suffer from attenuated periodic perception and entangled trend–noise representations as network depth increases. Moreover, the widely adopted channel-independent paradigm, while improving training stability, disrupts intrinsic dynamic coordination among variables, hindering the modeling of cross-variable consistency in multivariate time series. To address these issues, we propose PESD-TSF, a physics-inspired structured decomposition framework for long-term time series forecasting that jointly emphasizes interpretability and predictive accuracy. PESD-TSF introduces three key designs. First, a Multiplicative Periodic Gating mechanism incorporates continuous-time priors to dynamically modulate signal amplitudes, preserving periodic structures across deep layers. Second, a multi-scale structured encoder integrates detrended attention with hierarchical sampling to explicitly decouple long-term trends from high-frequency variations while retaining fine-grained temporal semantics. Third, to recover disrupted inter-variable dependencies, we propose Cross-Scale Collaborative Attention (CSCA) together with an RLC regularization scheme, which reconstructs global inter-variable topology in deep feature spaces and enforces physically consistent collaboration through orthogonality and consistency constraints. Extensive experiments on benchmark datasets from multiple domains demonstrate that PESD-TSF consistently achieves state-of-the-art performance, with particularly strong gains on multivariate forecasting tasks involving complex inter-variable coupling, highlighting its superior structural modeling capability and generalization.

Reinforcement learning (RL) fine-tuning has become a key technique for enhancing large language models (LLMs) on reasoning-intensive tasks, motivating its extension to vision language models (VLMs). While RL-tuned VLMs improve on visual reasoning benchmarks, they remain vulnerable to weak visual grounding, hallucinations, and over-reliance on textual cues. We show that simple, controlled textual perturbations—misleading captions or incorrect chain-of-thought (CoT) traces—cause substantial drops in robustness, and that these effects are more pronounced when CoT consistency is taken into account across open-source multimodal reasoning models. In contrast, closed models exhibit similar failure modes but maintain markedly greater robustness and reasoning consistency, suggesting that the gap reflects a shortcoming in current RL fine-tuning methods rather than an inherent limitation of the task. To better understand these vulnerabilities, we further analyze RL fine-tuning dynamics and uncover an accuracy–faithfulness trade-off: fine-tuning raises benchmark accuracy, but can simultaneously erode the reliability of the accompanying CoT and its robustness to contextual shifts. Although adversarial augmentation improves robustness, it does not by itself prevent faithfulness drift. Incorporating a faithfulness-aware reward can restore alignment between answers and reasoning, but when paired with augmentation, training risks collapsing onto shortcut strategies. Together, these findings highlight the limitations of accuracy-only evaluations and motivate training and assessment protocols that jointly emphasize correctness, robustness, and consistency in visually grounded reasoning.

Reinforcement learning algorithms such as GRPO have driven recent advances in large language model (LLM) reasoning. While scaling the number of rollouts stabilizes training, existing approaches suffer from limited exploration on challenging prompts and leave informative feedback signals underexploited, due to context-independent rollout allocation across prompts (e.g., generating 16 rollouts per prompt) and relying heavily on sparse rewards. This paper presents XRPO (eXplore–eXploit GRPO), a unified framework that recasts policy optimization through the principled lens of rollout exploration–exploitation. To enhance exploration, XRPO introduces a mathematically grounded rollout allocator that adaptively prioritizes prompts with higher potential for uncertainty reduction. It further addresses stagnation on zero-reward prompts through an in-context seeding strategy that injects curated exemplars, steering the model into more difficult reasoning trajectories. To strengthen exploitation, XRPO develops a group-relative, novelty-aware advantage sharpening mechanism that leverages sequence likelihoods to amplify low-probability yet correct responses, thereby extending the policy’s reach beyond sparse rewards. Experiments across diverse math and coding benchmarks on both reasoning and non-reasoning models demonstrate that XRPO outperforms existing advances (e.g., GRPO and GSPO) up to 4% pass@1 and 6% cons@32, while accelerating training convergence by up to 2.7x.

As Large Reasoning Models (LRMs) are increasingly deployed, auditing their chain-of-thought (CoT) traces for safety becomes critical. Recent work has reported that monitorability—the degree to which CoT faithfully and informatively reflects internal computation—can appear as a "free gift" during the early stages of Reinforcement Learning with Verifiable Rewards (RLVR). We make this observation concrete through a systematic evaluation across model families and training domains. Our results show that this effect is not universal: monitorability improvements are strongly data-dependent. In particular, we demonstrate the critical role of data diversity and instruction-following data during RLVR training. We further show that monitorability is orthogonal to capability—improvements in reasoning performance do not imply increased transparency. Through mechanistic analysis, we attribute monitorability gains primarily to response distribution sharpening (entropy reduction) and increased attention to the prompt, rather than stronger causal reliance on reasoning traces. We also reveal how monitorability dynamics vary with controlled training and evaluation difficulty. Together, these findings provide a holistic view of how monitorability emerges under RLVR, clarifying when gains are likely to occur and when they are not.

Current approaches to enhance Large Language Model (LLM) reasoning, such as Chain-of-Thought and "Wait" prompts, primarily encourage models to think more, yet often fail to guide them toward Truth. While Representation Editing (RepE) offers a intrinsic control, its application to dynamic reasoning trajectories remains underexplored. In this work, we bridge this gap by investigating the geometry of truth within unfolding reasoning chains. We uncover three critical insights: (1) Truth is encoded at the sentence level and is entangled with latent reasoning patterns; (2) Effective intervention follows an Uncertainty Principle and a Decay Effect, requiring localization to early, high-entropy forks; (3) Naive steering vectors suffer from noise, risking collateral damage to correct trajectories. Based on these findings, we propose DynaSteer, a dynamic RepE framework. DynaSteer employs pattern clustering to disentangle reasoning manifolds and utilizes Fisher-LDA to project purified truth. By dynamically monitoring lookahead entropy, it selectively steers and rolls back trajectories only when necessary. Comprehensive experimental results on several MATH benchmark verify the effectiveness of DynaSteer, and experiments on out-of-domain coding tasks further confirm its generalization ability. Our code is publicly available at https://github.com/tianlwang/DynaSteer.

As a multimodal extension of Chain-of-Thought (CoT), Thinking with Images (TWI) has emerged as a promising avenue to enhance the reasoning capability of Multi-modal Large Language Models (MLLMs), which generates interleaved CoT by incorporating visual cues into the textual reasoning process. However, the success of existing TWI methods heavily relies on the assumption that interleaved CoTs are faultless, which is easily violated due to the complexity of multimodal understanding. In this paper, we reveal and study a highly-practical yet under-explored problem in TWI, termed Noisy Thinking (NT). Specifically, NT refers to the imperfect visual cues mining and answer reasoning process. As the saying goes, ``One mistake leads to another'', erroneous interleaved CoT would cause error accumulation, thus significantly degrading the performance. To solve NT, we propose a novel method dubbed Reliable Thinking with Images (RTWI). In brief, RTWI estimates the reliability of visual cues and textual CoT in a unified text-centric manner and then employs robust filtering and voting modules to prevent NT from contaminating final answer. Extensive experiments on seven benchmarks verify the effectiveness of RTWI against NT.

Autoregressive Large Language Models (LLMs) often fail in complex reasoning because early-stage errors remain uncorrectable in subsequent steps—a limitation fundamentally rooted in the inherent irreversibility of the Transformer architecture. In this paper, we propose HEdit, a lightweight reasoning enhancement paradigm that equips models with a "hindsight-like" capability for dynamic error correction during generation. Our core insight involves deconstructing reasoning failures into two pivotal stages: latent representational biases emerging at logical anchors, and the subsequent eruption of explicit cognitive dissonance at trigger points. Based on these observations, the HEdit framework detects internal inconsistency signals at trigger points in real-time, actively backtracks to critical anchors, and utilizes a lightweight trainable editor to precisely refine their Key-Value (KV) caches. This mechanism effectively breaks the unidirectional constraints of autoregressive inference. Empirical results demonstrate that HEdit significantly enhances the performance of various models on mathematical reasoning tasks—with average accuracy improvements ranging from 2.2\% to 10.8\%—while maintaining extremely low overhead (add parameters $<0.5\%$). HEdit provides a dynamic, pluggable and lightweight solution, making it particularly beneficial for users in low-resource environments.

Chain-of-Thought (CoT) prompting enables multi-step reasoning in large language models, yet long-horizon generation remains brittle under distribution shift and context interference: irrelevant cues persist, small deviations compound into inference drift, and late-stage corrections can destabilize the trajectory. We recast autoregressive decoding as a perturbed long-horizon dynamical system and introduce an *inference-time stabilization operator* that targets *trajectory-level* reliability rather than token-level fluency. Specifically, we propose *ODE-guided language models*, which augment a base Transformer with a persistent continuous-time *thought state* whose dynamics are explicitly designed to be dissipative, enabling stable evidence accumulation with controlled forgetting. Instantiating this framework, *Thinking in Flow* (TiF) equips the model with a lightweight Neural ODE controller and injects its output through post-norm residual updates to achieve numerically stable, low-intrusion steering. A demand--supply (uncertainty--capacity) gate determines *when* intervention is warranted, while a direction gate determines *how* to steer in representation space, yielding selective, do-no-harm corrections instead of persistent bias. We establish well-posedness, dissipativity, and incremental stability of the controlled thought dynamics, implying bounded interventions over arbitrarily long contexts, and empirically demonstrate improved robustness to distractions and semantic perturbations, while matching or improving accuracy on mathematical reasoning benchmarks across both the Llama and Qwen model families; we further observe gains on non-mathematical BBH reasoning tasks when training TiF on Llama.

Despite the central role of sensor-derived measurements such as imaging traits and plasma biomarkers in biomedical research and clinical practice, existing generative models for disease prediction largely depend on event-level representations from hospital and registry data. Given the multi-factorial nature of human disease, the absence of explicit modeling of social determinants of health (SDoH) limits the capacity for personalized disease modeling and clinical decision support. To address this limitation, we propose a generative model with ICD-coded proxies of SDoH for \textit{in silico} modeling of disease reasoning, a conditioned latent diffusion framework that establishes the connection between multi-organ sensor data with tokenized healthcare events. Specifically, we introduce a novel geometric diffusion model to characterize the temporal evolution of complex data representation such as brain networks (region-to-region connectivity encoded in a graph), in parallel with diffusion models for tabular data from other organ systems. Together, we integrate the generative model with digitalized SDoH proxies (coined **DiffDT**) for simulated intervention and reasoning of future disease trajectories. We conduct extensive experiments on the UK Biobank (UKB) dataset, which contains organ-specific imaging traits, including brain (44,834), heart (23,987), liver (28,722), and kidney (32,155), along with nearly 500k medical history sequences (age range: 25$\sim$89 years). Our **DiffDT** achieves significant improvements over state-of-the-art human disease autoregressive models and imaging trait generative baselines.

Despite the growing use of large language models (LLMs) in subjective tasks such as role-playing, humor, emotional intelligence, and dialogue quality, their evaluation faces a pressing reproducibility crisis: even the same evaluator may contradict itself when re-judging the exact same sample. We attribute this instability to dimension drift, where free-form evaluation protocols (e.g., Chain-of-Thought reasoning) unpredictably shift their implicit criteria, undermining reliability. To address this fundamental challenge, we reformulate subjective evaluation as an information-theoretic optimization problem. Specifically, we propose an **Expected Information Gain (EIG)-based framework** that constructs a stable yet adaptive personalized rubric to eliminate dimension drift. Our two-stage “generate–then–score” design first produces a diverse pool of candidate evaluation questions and then selects the most informative subset via EIG, yielding explicit and repeatable criteria. Experiments on six benchmarks, including CharacterEval, The rJokes, and MT_bench, demonstrate that our approach substantially improves both evaluation consistency and alignment with human judgments, outperforming CoT-based and fixed-questionnaire baselines. These results highlight that information-theoretic questionnaire construction offers a principled and reliable path toward reproducible evaluation of subjective tasks.

Small language models (SLMs) offer efficient deployment, yet they often lag behind their larger counterparts (LLMs) in reasoning. Existing remedies either invoke an LLM at points of reasoning divergence, incurring substantial latency and cost, or rely on standard distillation, which is limited by the SLM's capacity to accurately mimic the LLM's complex generative distribution. We address this dilemma by identifying local sufficiency: at divergence points, the LLM's preferred token often resides within the SLM's top-$K$ next-token predictions, even when failing to emerge as the SLM top-1 choice. We therefore propose Select to Think (S2T), which reframes the LLM’s role from open-ended generation to selection among the SLM's proposals, simplifying the supervision signal to discrete candidate rankings. Leveraging this, we introduce S2T-Local, which distills the selection logic into the SLM, empowering it to perform autonomous re-ranking without inference-time LLM dependency. Empirically, a 1.5B SLM's top-8 candidates contain the 32B LLM's choice with a 95\% hit rate, and S2T-Local improves the 1.5B SLM's Math Avg. over greedy decoding by 24.1\% relative gain, matching the efficacy of 8-path self-consistency with single-trajectory efficiency. Code is available at https://github.com/YeRona/Select-to-Think.

Event sequences from complex systems, such as clinical workflows, are often sparse and incomplete. As a result, downstream models are trained on data that only partially captures the underlying dynamics. Synthetic sequence generation can augment real data by filling in missing structure and improving coverage of rare patterns, but generated trajectories must remain realistic, satisfy domain constraints, and allow control. We propose the Forward-Chaining Temporal Point Process (FC-TPP), a framework for constraint-aware and controllable sequence generation in continuous time. FC-TPP maintains an explicit latent symbolic state encoding high-level predicates, which evolves through a differentiable multi-hop forward-chaining operator. Logical rules update the latent state based on recent events, while a temporal point process decoder generates future event times and types conditioned on this evolving state. By tying the generative dynamics to multi-hop reasoning in latent space, FC-TPP incorporates symbolic structure throughout generation rather than relying directly on raw event histories. Experiments on synthetic data and four semi-synthetic/real-world benchmarks—LogiCity, MIMIC-IV, EPIC-100, and IKEA ASM—show that FC-TPP achieves higher generation quality under limited and incomplete data, with stronger constraint adherence and greater controllability than purely neural and prior neuro-symbolic baselines.

Example-based guidance is widely used to improve mathematical reasoning at inference time, yet its effectiveness is highly unstable across problems and models—even when the guidance is correct and problem-relevant. We show that this instability arises from a previously underexplored gap between *strategy usage*—whether a reasoning strategy appears in successful solutions—and *strategy executability*—whether the strategy remains effective when instantiated as guidance for a target model. Through a controlled analysis of paired human-written and model-generated solutions, we identify a systematic dissociation between usage and executability: human- and model-derived strategies differ in structured, domain-dependent ways, leading to complementary strengths and consistent source-dependent reversals under guidance. Building on this diagnosis, we propose *Selective Strategy Retrieval* (SSR), a test-time framework that explicitly models executability by selectively retrieving and combining strategies using empirical, multi-route, source-aware signals. Across multiple mathematical reasoning benchmarks, SSR yields reliable and consistent improvements over direct solving, in-context learning, and single-source guidance, improving accuracy by up to $+13$ points on AIME25 and $+5$ points on Apex for compact reasoning models. Code and benchmark are publicly available at: https://github.com/lwd17/strategy-execute-pipeline.

Natural-language explanations are widely used to interpret machine learning models, yet many prioritize human plausibility over accurately reflecting or predicting model behavior. Prior approaches often rely on human-written rationales, producing post-hoc explanations that neither align with the model’s decision function nor generalize. We introduce $\textbf{OPeX}$, a natural-language explanation model that directly optimizes for behavioral faithfulness: the ability of an explanation to reflect and predict a model’s observable input–output behavior. $\textbf{OPeX}$ is trained using reinforcement learning with Group Relative Policy Optimization (GRPO), optimizing two complementary metrics: recoverability, which measures whether explanations recover model predictions on seen examples, and simulatability, which measures prediction of model behavior on unseen inputs. Across structured and text-based tasks, $\textbf{OPeX}$ achieves high simulatability (∼0.85) and recoverability (∼0.99), outperforming GPT-4o, LLaMA-3.3-70B, MaNtLE, Chain-of-Thought (CoT)-based models, and human-written explanations, despite using an 8B-parameter backbone. Human user studies show a 15% improvement in classification accuracy over competent baselines.

Large language models are becoming increasingly significant in financial applications. Nevertheless, prevailing benchmarks are largely dependent on simulated or generic data, which leads to a significant gap between reported performance and actual efficacy in real-world scenarios. To tackle this challenge, we present BizFinBench.v2, the first integrated offline and online benchmark built upon authentic user query-response data from both Chinese and U.S. equity markets. It comprises 28,860 questions across eight offline and two online tasks. Experimental results show that GPT-5 achieves a mere 61.5\% accuracy, still failing to meet the practical business requirement (84.8\%). Among the evaluated commercial models, DeepSeek-R1 exhibits superior investment efficacy. Error analysis grounded in real financial practice reveals persistent limitations in existing models. By overcoming the constraints of prior benchmarks, BizFinBench.v2 provides a substantiated foundation for advancing LLM deployment in the financial sector. Our data and code are available at https://github.com/HiThink-Research/BizFinBench.v2.

Large Language Models (LLMs) employing Chain-of-Thought reasoning achieve strong performance but suffer from excessive token consumption that inflates inference costs. Existing efficiency methods—such as explicit length penalties, difficulty estimators, or multi-stage curricula—either degrade reasoning quality or require complex training pipelines. We introduce **Batched Contextual Reinforcement (BCR)**, a minimalist, single-stage training paradigm that unlocks efficient reasoning through a simple structural modification: training the model to solve N problems simultaneously within a shared context window, rewarded purely by per-instance accuracy. This formulation creates an implicit token budget that yields several key findings: (1) We identify a novel _task-scaling law_: as the number of concurrent problems N increases at inference time, per-problem token usage decreases monotonically---a phenomenon that arises purely at inference, holds for models both before and after training, and is unrelated to accuracy or the training procedure. BCR makes this regime practical by degrading accuracy far more gracefully than baselines as N grows, establishing N as a controllable throughput dimension. (2) BCR challenges the traditional accuracy-efficiency trade-off by demonstrating a "free lunch" phenomenon at standard single-problem (N=1) inference. Across both 1.5B and 4B model families, BCR reduces token usage by 15.8\% to 62.6\% while consistently maintaining or improving accuracy across five major mathematical benchmarks (e.g., +13.3\% on AIME25 for the 4B model). (3) Qualitative analyses reveal emergent self-regulated efficiency, where models autonomously eliminate redundant metacognitive loops without explicit length supervision. (4) Crucially, we empirically demonstrate that implicit budget constraints successfully circumvent the adversarial gradients and catastrophic optimization collapse inherent to explicit length penalties, offering a highly stable, constraint-based alternative for length control. These results establish BCR as a highly practical framework, demonstrating how simple structural training incentives can unlock latent high-density reasoning modes in LLMs.

We propose Re-FORC, an adaptive reward prediction method that, given a query, enables prediction of the expected future rewards as a function of the number of future thinking tokens. Re-FORC trains a lightweight adapter on reasoning models, demonstrating improved prediction with longer reasoning and larger models. Re-FORC enables: 1) early stopping of unpromising reasoning chains, reducing compute by up to 24\% compared to fixed-budget cutoffs, while maintaining accuracy, 2) optimized model and thinking length selection that outperforms the largest model alone---reaching 1.7 percentage points higher peak accuracy while needing up to 12\% less compute to match the largest model's accuracy, 3) adaptive test-time scaling, which increases accuracy by 8.8 percentage points (on average at maximum compute) over confidence-based baselines. Re-FORC allows dynamic reasoning with length control via cost-per-token thresholds while estimating computation time upfront.