NeurIPS 2026 论文列表

The task of visual reasoning relative pose identification (VRRPI) has exposed critical weaknesses in the multi-view 3D spatial reasoning capabilities of modern vision-language models (VLMs). We present a systematic analysis of VLMs behavior in VRRPI, uncovering stable output biases where VLMs disproportionately predict specific motion patterns regardless of visual evidence. To mitigate these inherent biases, we propose self knowledge re-expression with iterative debiasing (SKR-ID), an annotation-free pipeline that adapts VLMs for VRRPI using only randomly extracted unannotated image pairs. SKR-ID transitions VLMs' output mechanism from generic next-token prediction to task-optimized classification heads. The pipeline proceeds in two stages: (1) a cold-start initialization using debiasing object-centric prompts to anchor motion inference in 2D visual evidence; and (2) a recursive logit-ranking mechanism that enforces statistical parity by using the median logit value as a dynamic threshold. Experimental evaluations across frontier VLMs on standard benchmarks demonstrate that SKR-ID significantly mitigates systematic biases and unearths latent 3D reasoning ability. Our method achieves substantial gains, improving open-sourced VLMs' Macro-F1 scores by at least 8.3% and at most 23.0%, proving that VLMs possess significant untapped multi-view 3D reasoning potential.

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a key method for improving Large Language Models' reasoning capabilities, yet recent evidence suggests it may paradoxically shrink the reasoning boundary rather than expand it. This paper explains this shrinkage issue of RLVR by analyzing its learning dynamics and reveals two critical phenomena that account for this failure. First, we expose negative interference in RLVR, where learning to solve certain training problems actively reduces the likelihood of correct solutions for others, leading to the decline of Pass@$k$ performance, or the probability of generating a correct solution within $k$ attempts. Second, we uncover the winner-take-all phenomenon: RLVR disproportionately reinforces problems with high likelihood, correct solutions under the base model, while suppressing other initially low-likelihood ones. Through extensive theoretical and empirical analysis on multiple mathematical reasoning benchmarks, we show that this effect arises from the inherent on-policy sampling in standard RL objectives, causing the model to converge toward narrow solution strategies. These insights motivate us to design a \textit{simple yet effective} data curation algorithm that focuses RLVR learning on low-likelihood problems (the non-winners) and achieves notable improvement in Pass@$k$ performance, illustrating the potential of a broader class of data-curation mitigation strategies.

We envision a new era of AI, termed agentic organization, where agents solve complex problems by working collaboratively and concurrently, enabling outcomes beyond individual intelligence. To realize this vision, we introduce asynchronous thinking (AsyncThink) as a new paradigm of reasoning with large language models, which organizes the internal thinking process into concurrently executable structures. Specifically, we propose a thinking protocol where an organizer dynamically assigns sub-queries to workers, merges intermediate knowledge, and produces coherent solutions. More importantly, the thinking structure in this protocol can be further optimized through reinforcement learning. Experiments demonstrate that AsyncThink achieves 28% lower inference latency compared to parallel thinking while improving accuracy on mathematical reasoning. Moreover, AsyncThink generalizes its learned asynchronous thinking capabilities, effectively tackling unseen tasks without additional training.

Group Relative Policy Optimization (GRPO) is widely used to post-train reasoning language models, and its rollout group size $G$ is typically treated as a compute hyperparameter. We argue that $G$ also sets a difficulty threshold below which prompts produce no learning signal. Under binary verifier rewards, a rollout group only produces a learning signal when it contains both a success and a failure, and the rate at which a prompt yields such mixed groups turns on sharply near $p_c\simeq\ln 2/G$. Below that threshold the prompt is invisible to the relative-advantage estimator regardless of optimizer state. We verify this gate directly in pass@256 rollout traces from Qwen3-1.7B, Qwen3-8B, and Llama-3.1-8B, where empirical mixed-group rates match the binomial prediction within $1.4$–$1.9$ percentage points on average per model. Sweeping fixed-$G$ GRPO with $G\in\{8,16,32,64,128\}$ then shows that difficulty-binned per-prompt improvement after training tracks the frontier shape with binned $R^2$ between $0.962$ and $0.984$ across all three model families. Treating $p_c$ as a moving target, we instantiate a frontier-tracking group-size schedule that improves the best fixed-$G$ baseline by $4.2$–$5.3$ points on MATH-500 and $2.4$–$3.4$ points on GSM8K under matched rollout budget. Ablations attribute the gain primarily to frontier-aware switch timing.

We introduce EvoLib, a test-time learning framework that enables large language models to accumulate, reuse, and evolve knowledge across problem instances without parameter updates or external supervision. Instead of adapting model parameters, our approach maintains a shared library of knowledge abstractions, including modular skills and reflective insights, automatically extracted from the model's own inference trajectories. To support continual improvement, we introduce a principled weighting and consolidation mechanism that jointly optimizes for immediate utility and long-term value. This allows simple, instance-specific abstractions to evolve into more general and reusable ones over time. Across challenging benchmarks in mathematical reasoning, code generation, and multi-turn agentic environments, EvoLib improves substantially over the top test-time scaling and learning methods without ground-truth feedback.

Chest X-ray visual question answering (CXR VQA) requires models not only to predict correct answers, but also to produce reliable medical reasoning. However, existing reinforcement-learning-based training typically relies on answer-level rewards, which are often too coarse to improve chain-of-thought (CoT) quality and can become ineffective when group-level advantages collapse to zero. We propose \textbf{Teach-to-Reason (T2R)}, a framework that introduces comparison-based supervision into CoT optimization through a self-improving \emph{Teacher} and a competition-guided \emph{Reasoner}. As the Teacher is iteratively strengthened via self-competition, the Reasoner is optimized against progressively stronger Teacher-generated references. We further introduce a case-wise reward design that preserves the original reward-induced positive/negative partition when it is informative, and restores supervision from competition scores when the original reward signal degenerates. Experiments on multiple CXR open-ended VQA benchmarks show that T2R consistently outperforms strong baselines, indicating that comparison-based supervision, when integrated in a controlled and principled manner, provides a more effective training signal for reasoning optimization.

In real home deployments, household agents must often operate from a complete household scene and a situated household request, rather than from a clean task specification. Such requests require agents to identify task-relevant entities, recover intended task conditions, and resolve ordering constraints from the surrounding scene context. We formalize this capability as full-scene household reasoning: given a complete household scene and a situated household request, an agent must infer executable task structure before producing a grounded skill-level action sequence. This setting is challenging because complete household scenes contain substantial task-irrelevant information, making direct complete-scene prompting inefficient and error-prone. In practical deployment, this challenge is further amplified by privacy and local compute constraints, which favor compact open-weight models with limited long-context reasoning ability. We propose TaskGround, a training-free and model-agnostic Ground-Infer-Execute framework that grounds complete scenes into compact task-relevant scene slices, infers executable task structure, and compiles it into grounded skill-level action sequences. To evaluate this setting, we introduce FullHome, a human-validated evaluation suite of 400 household tasks spanning diverse home-scale environments and both goal-oriented and process-constrained requirements. On FullHome, TaskGround improves task success rates by large margins across both proprietary and open-weight models. Notably, it makes Qwen3.5-9B competitive with GPT-5 under direct complete-scene prompting while reducing total input-token cost by up to 18x. Our results identify executable task-structure inference as a central bottleneck in full-scene household reasoning and show that structured grounding can make compact local models substantially more effective for practical household deployment.

Therapeutic target identification---determining which genes or proteins to modulate for a desired clinical effect---is a foundational step in drug discovery. However, existing computational methods that rank candidates at genome scale face three persistent challenges: (i) static feature design, which discards mechanistic knowledge locked in biomedical literature; (ii) lack of interpretability, offering no auditable gene-level rationale for individual predictions; and (iii) the false-negative problem, where unlabeled but potentially druggable genes are treated as definitive negatives. We propose TargetSage, an LLM reasoning agent for therapeutic target identification built on positive-unlabeled (PU) learning, with three modules. First, Agentic Profiling enriches each gene with LLM-gathered biomedical evidence, distilling 2,261 attributes into a 23-dimensional reasoning vocabulary. Second, Guided Reasoning produces gene-specific rationales and explicit reasoning attributes through a GRPO-trained policy rewarded by downstream Adjusted $F_1$. Third, PU-Aware Scoring integrates structured biological features, explicit reasoning attributes, and implicit reasoning embeddings via gated fusion under an LLM-informed class-prior to produce genome-wide druggability scores. Evaluated on 15 benchmark tasks spanning 19,032 human protein-coding genes, TargetSage consistently outperforms ten baselines, surpassing the strongest by 37% in macro-average Adjusted $F_1$ and generalizing to independent held-out benchmarks. Using only 2021 training data, TargetSage outperforms all ten baselines at ranking genes that gain clinical approval by 2025. On an independent validation set of Phase 2/3 clinical candidates, it further achieves a 51% higher enrichment odds ratio at the top-25 shortlist than the runner-up baseline. TargetSage provides an evidence-grounded, continuously updatable foundation for interpretable target prioritization at genome scale.

Assessing the quality of time series (TS) data is fundamental yet inherently challenging due to the multifaceted nature of quality dimensions. Recently, large language models (LLMs) have emerged as a promising paradigm for TS quality assessment via pairwise comparison and per-dimension evaluation. However, existing approaches rely on manually predefined quality dimensions and purely text-based reasoning, leaving it unknown whether LLMs can identify truly relevant quality dimensions or perform grounded and quantitative quality comparisons. To investigate this, we construct TSQBench, a dedicated benchmark for evaluating LLMs’ capabilities in TS quality assessment, which measures two progressive capabilities: (i) understanding and identifying relevant quality dimensions, and (ii) performing quality comparison under specific dimensions. Our analysis reveals that both open-source and proprietary LLMs consistently struggle with identifying critical quality dimensions and conducting precise, evidence-grounded reasoning over quality characteristics. To address these limitations, we propose TSQAgent, a novel agentic reasoning framework for dedicated TS data quality rating, comprising three collaborative roles: Perceiver for focused dimension selection, Inspector for dimension-wise quantitative analysis, and Adjudicator that aggregates and refines the final judgment. In particular, we introduce an agentic reasoning strategy that instills the ability to identify and prioritize the most relevant quality dimensions, and further propose an agent workflow equipped with external analytical tools to enable precise quantitative comparisons over selected dimensions. Experiments on both the proposed benchmark and eleven real-world datasets demonstrate that our framework not only substantially improves LLMs’ capabilities in quality understanding and quantitative comparison but also effectively translates these improvements into better quality-aware data selection, leading to enhanced downstream performance and data efficiency.

Multimodal Large Language Models (MLLMs) can exhibit strong text-based reasoning yet fail to apply the same capabilities to semantically equivalent visual inputs. We study this failure mode in a controlled setting by rendering text-only reasoning problems as images, preserving semantic content while changing only the input modality. Across multiple MLLM families, models solve the same problems substantially better from text than from images. We characterize this discrepancy as a reasoning-access gap, where models may correctly perceive visual content but still fail to route that content into the latent reasoning machinery used for text-based tasks. To bridge this gap, we propose TIGER (Text-to-Image Gap-targeted Training for Enhanced Reasoning), a framework that repurposes text-only reasoning corpora into effective multimodal training data. By mining "modality counterfactuals", instances where a model succeeds on a text problem but fails on its semantically-equivalent rendered image, TIGER provides targeted supervision for reasoning-access failures without requiring manually curated multimodal datasets. We implement TIGER using image-conditioned Group Relative Policy Optimization (GRPO), along with SFT and DPO variants, and show how it consistently narrows the modality gap and yields generalized visual reasoning performance gains in multimodal reasoning benchmarks such as MathVerse and EMMA. We show that Reinforcement Learning with Verifiable Reward-based models still exhibit modality-dependent reasoning gaps, and that our training recipe can further reduce these gaps. Further analysis using reasoning-subspace activation and activation patching shows that TIGER enables visual representations that better activate reasoning-relevant latent subspaces of the language backbone. Our results suggest that advancing multimodal intelligence requires moving beyond perceptual alignment toward explicit visual access to existing reasoning machinery.

Large Reasoning Models (LRMs) achieve impressive performance on complex reasoning tasks via Chain-of-Thought (CoT) reasoning, which enables them to generate intermediate thinking tokens before arriving at the final answer. However, LRMs often suffer from significant overthinking, spending excessive compute time even after the answer is generated early on. Prior work has identified the existence of an optimal reasoning length such that truncating reasoning at this point significantly shortens CoT outputs with virtually no change in performance. However, determining optimal CoT lengths for practical datasets is highly non-trivial as they are fully task and model-dependent. In this paper, we precisely address this and design Terminator, an early-exit strategy for LRMs at inference to mitigate overthinking. The central idea underpinning Terminator is that the first arrival of an LRM's final answer is often predictable, and we leverage these first answer positions to create a novel dataset of optimal reasoning lengths to train Terminator. Powered by this approach, Terminator achieves significant reductions in CoT lengths of 14%–55% on average across four challenging practical datasets: MATH-500, AIME 2025, HumanEval, and GPQA, while outperforming current state-of-the-art methods and reducing inference latency by more than 2$\times$ compared to the original LRM.

Reinforcement learning (RL) is increasingly used in post-training to adapt large models to new reasoning tasks. Recent forgetting analyses, however, are mostly answer-centric: they report reward or accuracy on previously learned tasks and implicitly treat stable answers as evidence of stable reasoning. We show that this view is incomplete. In our sequential reasoning settings, externalized reasoning traces can change substantially in structure, strategy, and sample-wise behavior before answer-level metrics visibly collapse, and harder full-parameter staged runs can later degrade into catastrophic forgetting on earlier tasks. We refer to the trace-first regime as \emph{reasoning drift}. To address it, we propose \emph{Trace-Anchored Regularization Prior} (TARP), a trace-centric regularizer that mines reliable reasoning spans from successful rollouts and uses them to build a compact trace-derived quadratic prior for subsequent RL updates. Across multiple reasoning task sequences and model backbones, TARP reduces selected measured drift signals and improves earlier-task preservation in several evaluated settings, while also revealing a retention-verbosity trade-off that answer-only evaluation hides.

When distilling reasoning from large language models (LLMs) into smaller ones, teacher rationales for similar problems often vary wildly in structure and strategy. Like a chef who makes the same dish differently each time, this inconsistency burdens the student with noisy supervision that is hard to internalize. We propose Distillation through Reasoning Path Compression (D-RPC), which constrains the teacher to follow a compact, dynamically maintained bank of reusable high-level reasoning paths. For each training question, D-RPC retrieves the most relevant path and conditions the teacher to follow it, producing rationales that are consistent across similar problems yet diverse enough to cover different problem types. A PAC-Bayes analysis formalizes the resulting trade-off between bank size and coverage: smaller banks reduce supervision entropy but risk coverage gaps, and the generalization bound identifies an optimal intermediate size confirmed by our ablations. Across five math and commonsense reasoning benchmarks with two student models, D-RPC consistently outperforms chain-of-thought distillation, freeform rationale generation, direct distillation, and structured-supervision baselines, while using fewer tokens than template-heavy alternatives.

In practice, the reasoning strategy often matters more than the model. Rather than calling a fixed large model once per input, a well-designed strategy breaks the problem into steps and uses the right model for each, achieving better accuracy at lower cost. Existing methods have tried to automate strategy design by searching over a fixed set of strategy templates, but the templates themselves limit what reasoning patterns can be discovered. We propose Strategist, a meta-agent that designs a high-performing inference strategy for a given task. Each strategy is a structured executable topology that defines how agents plan, interact, verify, and route across models. Given a task and a small development set, Strategist proposes candidate strategies, runs them on real samples, and revises them against measured accuracy and cost. Every strategy and its components are reusable, forming an evolving library that grows with each new task and compounds what future tasks can build from. On 30 benchmarks, Strategist matches or beats the strongest baselines with an average of 8.5 absolute percentage points higher accuracy at 48% lower inference cost.

While large reasoning models (LRMs) can solve complex tasks via generating extended Chain-of-Thought traces (Long-CoT), longer traces often increase confidence without increasing correctness. In this work, we formalize such failure via a \emph{self-conditioned evidence} model: generated tokens provide information about the model's current working hypothesis rather than the ground-truth answer. This yields a simple diagnostic principle: entropy, fluency, and answer stability can certify within-trajectory self-consistency while remaining causally disconnected from correctness. Building on these findings, we propose \textbf{Faithful Pandora Inference (FPI)}, a minimal training-free controller that keeps the base LRM frozen and fits only a small calibration map on held-out data. At each checkpoint, FPI estimates calibrated correctness $r_t$ and the marginal evidence gain $m_t$ of continuing the current trajectory. It stops only when $r_t$ reaches a user-specified reliability target; if $r_t$ remains insufficient and $m_t$ falls below cost, it reallocates the remaining budget to a diversified restart. Across six reasoning benchmarks and multiple DeepSeek-R1 distilled models, FPI improves the accuracy-efficiency frontier, reduces expected calibration error, and improves selective accuracy on high-confidence outputs. Ablations on confidently wrong trajectories show that calibrated stopping and marginal-gain restarts address different failure modes and that the gains are not explained by additional samples alone.

Large reasoning models improve with more test-time computation, but often overthink, producing unnecessarily long chains-of-thought that raise cost without improving accuracy. Prior reinforcement learning approaches typically rely on a single outcome reward with trajectory-level length penalties, which cannot distinguish essential from redundant reasoning steps and therefore yield blunt compression. Although recent work incorporates step-level signals, such as offline pruning, supervised data construction, or verifier-based intermediate rewards, reasoning length is rarely treated as an explicit step-level optimization objective during RL. We propose Step-wise Adaptive Penalization (SWAP), a fine-grained framework that allocates length reduction across steps based on intrinsic contribution. We estimate step importance from the model's on-policy log-probability improvement toward the correct answer, then treat excess length as a penalty mass redistributed to penalize low-importance steps more heavily while preserving high-importance reasoning. We optimize with a unified outcome–process advantage within group-relative policy optimization. Extensive experiments demonstrate that SWAP reduces reasoning length by 56.3% on average while improving accuracy by 4.2% relative to the base model.

Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve the reasoning capabilities of large language models, but its reward is derived only from the correctness of the final answer and provides no supervision over intermediate reasoning steps. Recent rubric-based methods, such as Rubrics as Rewards (RaR), introduce finer-grained supervision by scoring rollouts against a structured set of evaluation criteria. However, the resulting rubric scores are still aggregated into a single scalar that is applied to the entire response, leading to three structural weaknesses, namely loss of the multi-criterion rubric structure, uniform supervision of correct and incorrect reasoning steps, and reward hacking in the trained model through unbounded self-correction. On a sample of 1{,}000 problems, we find that 18.2\% of steps within answer-correct responses are themselves wrong yet positively rewarded, while 49.9\% of steps within answer-incorrect responses are in fact correct yet penalized. Therefore, we introduce \textbf{Step-wise Rubrics as Rewards (SRaR)}, an RLVR framework that (i) uses an LLM judge to attribute each rubric item to a specific reasoning step, (ii) normalizes the per-step rubric scores across rollouts so that only steps whose quality varies produce a learning signal, and (iii) combines the resulting per-step reward with the standard outcome reward through a decoupled advantage estimator that keeps the outcome-driven baseline stable. To support training, we further build a 16K-problem rubric dataset by contrastively distilling rubric items from correct and flawed reasoning paths sampled from a strong model and verified against the ground-truth answer. Across six mathematical reasoning benchmarks spanning multiple difficulty levels, SRaR improves the average accuracy over RaR by \textbf{3.57} points on Qwen3-8B-Non-Thinking and by \textbf{2.75} points on Qwen3-32B-Non-Thinking, raises the Faithful Reasoning Rate on AIME~2025 from 34.5\% to 46.7\% (every correct-answer trajectory of SRaR uses entirely correct reasoning steps), and reduces the rate of self-correction looping from 48.1\% to 26.5\%.

Verifiable reward training has improved mathematical and coding reasoning, but these domains capture only part of step-by-step decision making. Many real-world tasks require finding a high-value feasible plan among many valid alternatives. We introduce $\text{OPT}\star$, a scalable family of optimization-style tasks for training and evaluating LLM step-by-step optimization-like reasoning along a complexity axis: each task provides a feasibility checker and evaluator, while a complexity parameter expands the search space without requiring new human labels. This motivates studying these tasks in two regimes: (i) solver-guided online policy optimization, which uses a solver as a value oracle for partial states and applies rank-based reward shaping to reinforce better next steps, and (ii) search-based offline RL when such solvers are unavailable. Theoretically, we relate success in large search spaces to the information a reasoner extracts per unit of search budget. Empirically, we ablate the ingredients that make search efficient on $\text{OPT}\star$ and show that training on $\text{OPT}\star$ improves step-by-step optimization-like reasoning.

Large language models deployed as personalized assistants must reason over long, evolving interaction histories. However, in long-term dialogue reasoning, relevant evidence is scattered across sessions, preferences may be revised over time, and standard long-context training fails to address these challenges under data scarcity and prohibitive computational costs. We propose StateTree, a data-driven RL pseudo-task that constructs a challenging auxiliary task from scarce dialogues with verifiable ground truth. StateTree augments multi-session dialogues with a tree-structured path-tracing task: key-value records are embedded across sessions to form a binary tree. Solving the task requires the model to traverse from root to leaf by retrieving records across sessions and comparing timestamps to resolve branches, then recover the hidden target question among distractor leaves. We apply curriculum RL training progressively increasing tree depth and introduce a compositional variant whose edges carry step-level reasoning fragments, training the model to compose partial cues into coherent queries. Trained on 10K-token contexts, StateTree generalizes to 128K tokens without full-length RL costs and exhibits capabilities including cross-session retrieval, temporal reasoning, knowledge update, and compositional multi-hop reasoning. StateTree outperforms both SFT and RL-based baselines while preserving short-context general reasoning. StateTree-7B achieves gains up to +23.60% on LongMemEval (128k), and StateTree-14B reaches 59.00% accuracy on LongMemEval, surpassing QwenLong-L1-32B (45.20%).