ICLR 2026 论文列表

With the growing adoption of large language model (LLM) agents in persistent, real-world roles, they naturally encounter continuous streams of tasks and interactions. A key limitation, however, is their failure to learn from this accumulated experience, forcing them to discard valuable insights and repeat past errors. Unlike prior works that primarily store raw experience or successful routines, we propose ReasoningBank, a novel memory framework that allows an agent to self-curate generalizable reasoning strategies from both its successful and failed experiences for future leverage. This mechanism enables agents to generalize across tasks and become more capable over time. To accelerate and diversify this test-time learning process, we further propose memory-aware test-time scaling (MaTTS), which leverages a powerful synergy between memory and test-time scaling. On one hand, relevant memory from ReasoningBank guides the scaling process toward more effective exploration and improved reliability. On the other, scaling, in both parallel and sequential settings, generates abundant, diverse experiences that provide rich contrastive signals for synthesizing higher-quality memory. Experiments on web browsing and software engineering tasks show that ReasoningBank consistently outperforms existing memory mechanisms in both effectiveness and efficiency, with MaTTS further amplifying the gains. These findings position memory-driven experience as a new dimension of test-time scaling, where emergent behaviors naturally arise and agents acquire self-evolving capabilities.

Fluid-solid interaction (FSI) problems are fundamental in many scientific and engineering applications, yet effectively capturing the highly nonlinear two-way interactions remains a significant challenge. Most existing deep learning methods are limited to simplified one-way FSI scenarios, often assuming rigid and static solid to reduce complexity. Even in two-way setups, prevailing approaches struggle to capture dynamic, heterogeneous interactions due to the lack of cross-domain awareness. In this paper, we introduce \textbf{Fisale}, a data-driven framework for handling complex two-way \textbf{FSI} problems. It is inspired by classical numerical methods, namely the Arbitrary Lagrangian–Eulerian (\textbf{ALE}) method and the partitioned coupling algorithm. Fisale explicitly models the coupling interface as a distinct component and leverages multiscale latent ALE grids to provide unified, geometry-aware embeddings across domains. A partitioned coupling module (PCM) further decomposes the problem into structured substeps, enabling progressive modeling of nonlinear interdependencies. Compared to existing models, Fisale introduces a more flexible framework that iteratively handles complex dynamics of solid, fluid and their coupling interface on a unified representation, and enables scalable learning of complex two-way FSI behaviors. Experimentally, Fisale excels in three reality-related challenging FSI scenarios, covering 2D, 3D and various tasks. The code is included in the supplementary material for reproductivity.

Integrating Spiking Neural Networks (SNNs) with Transformer architectures offers a promising pathway to balance energy efficiency and performance, particularly for edge vision applications. However, existing Spiking Transformers face two critical challenges: i) a substantial performance gap relative to their Artificial Neural Network (ANN) counterparts, and ii) considerable memory overhead. Our theoretical analysis and empirical evidence indicate that these limitations arise from the unfocused global attention paradigm of Spiking Self Attention (SSA) and the storage cost of large attention matrices. Inspired by the localized receptive fields and membrane potential dynamics of biological visual neurons, we propose LRF-Dyn, which enables attention computation via spiking neurons endowed with localized receptive fields. Specifically, we integrate a LRF mechanism into SSA, enabling the model to allocate greater attention to neighboring regions and thereby enhance local modeling capacity. Moreover, LRF-Dyn approximates the charge–fire–reset dynamics of spiking neurons within the LRF-SSA, substantially reducing memory requirements during inference. Extensive experiments on visual tasks confirm that our method lowers memory overhead while delivering significant performance improvements. These results establish LRF-Dyn as a key component for achieving energy-efficient Spiking Transformers.

Reinforcement learning (RL) has recently become the core paradigm for aligning and strengthening large language models (LLMs). Yet, applying RL in off-policy settings—where stale data from past policies are used for training—improves sample efficiency, but remains challenging: policy entropy declines sharply, optimization often becomes unstable and may even collapse. Through theoretical and empirical analysis, we identify two key insights: (i) an imbalance in optimization, where negative-advantage samples dominate the policy gradient, suppressing useful behaviors and risking gradient explosions; and (ii) the derived Entropy-Clip Rule, which reveals that the fixed clipping mechanism in PPO-like objectives systematically blocks entropy-increasing updates, thereby driving the policy toward over-exploitation at the expense of exploration. Building on these insights, we propose BAlanced Policy Optimization with Adaptive Clipping (BAPO), a simple yet effective method that dynamically adjusts clipping bounds to adaptively re-balance positive and negative contributions, preserve entropy, and stabilize RL optimization. Across diverse off-policy scenarios—including sample replay and partial rollout—BAPO achieves fast, stable, and data-efficient training. On AIME 2024 and AIME 2025 benchmarks, our 7B BAPO model surpasses open-source counterparts such as SkyWork-OR1-7B, while our 32B BAPO model not only achieves state-of-the-art results among models of the same scale but also outperforms leading proprietary systems like o3-mini and Gemini-2.5-Flash-Thinking.

While recent advances in Diffusion Transformers (DiTs) have significantly advanced text-to-image generation, text-driven image editing remains challenging. Existing approaches either struggle to balance structural preservation with flexible modifications or require costly fine-tuning of large models. To address this, We introduce W-Edit, a training-free framework for text-driven image editing based on wavelet-based frequency-aware feature decomposition. W-Edit employs wavelet transforms to decompose diffusion features into multi-scale frequency bands, disentangling structural anchors from editable details. A lightweight replacement module selectively injects these components into pretrained models, while an inversion-based frequency modulation strategy refines sampling trajectories using structural cues from attention features. Extensive experiments demonstrate that W-Edit achieves high-quality results across a wide range of editing scenarios, outperforming previous training-free approaches. Our method establishes frequency-based modulation as both a sound and efficient solution for controllable image editing.

Learning nonlinear dynamical models from data is central to control. Two fundamental challenges exist: (1) how to learn accurate models from limited data, and (2) how to ensure the learned models are suitable for control design of the nominal system. We address both by enforcing a critical \emph{a priori} property of the nominal system during learning: \emph{controllability}. Controllability guarantees the existence of control policies that can drive the learned model from any initial state to any desired state. From a modeling perspective, it captures key structural features of the nominal system, thereby improving data efficiency. For downstream control, it enables the use of modern techniques such as model predictive control (MPC). Our approach is based on controllability-preserving Koopman representation learning. Rather than learning dynamics directly in the nominal state space, we learn in a latent space where the system admits a linear representation. We prove that controllability of the learned latent model implies controllability in the nominal state space. To enforce this property, we introduce a novel canonical parameterization of the latent dynamics matrices. We further incorporate Gramian-based regularization to shape the degree of controllability, yielding well-conditioned models for control. Implemented as an end-to-end Neural ODE framework, our method learns models that are both predictive and controllable from limited data. Experiments on nonlinear benchmarks demonstrate accurate long-horizon prediction, reliable MPC performance, and substantially improved data efficiency.

In this paper, we present FSOD-VFM: Few-Shot Object Detectors with Vision Foundation Models, a framework that leverages vision foundation models to tackle the challenge of few-shot object detection. FSOD-VFM integrates three key components: a universal proposal network (UPN) for category-agnostic bounding box generation, SAM2 for accurate mask extraction, and DINOv2 features for efficient adaptation to new object categories. Despite the strong generalization capabilities of foundation models, the bounding boxes generated by UPN often suffer from overfragmentation, covering only partial object regions and leading to numerous small, false-positive proposals rather than accurate, complete object detections. To address this issue, we introduce a novel graph-based confidence reweighting method. In our approach, predicted bounding boxes are modeled as nodes in a directed graph, with graph diffusion operations applied to propagate confidence scores across the network. This reweighting process refines the scores of proposals, assigning higher confidence to whole objects and lower confidence to local, fragmented parts. This strategy improves detection granularity and effectively reduces the occurrence of false-positive bounding box proposals. Through extensive experiments on Pascal-5$^i$, COCO-20$^i$, and CD-FSOD datasets, we demonstrate that our method substantially outperforms existing approaches, achieving superior performance without requiring additional training. Notably, on the challenging CD-FSOD dataset, which spans multiple datasets and domains, our FSOD-VFM achieves 31.6 AP in the 10-shot setting, substantially outperforming previous training-free methods that reach only 21.4 AP. The source code will be released publicly upon publication.

We introduce MedAgentGym, a scalable and interactive training environment designed to enhance coding-based biomedical reasoning capabilities in large language model (LLM) agents. MedAgentGym comprises 72,413 task instances across 129 categories derived from 12 authentic real-world biomedical scenarios. Tasks are encapsulated within executable sandbox environments, each featuring detailed task specifications, interactive feedback mechanisms, verifiable ground truth annotations, and scalable training trajectory generation. Extensive benchmarking of 29 LLMs reveals substantial performance disparities in biomedical data science between commercial and open-source LLMs. Leveraging efficient multi-threaded and multi-turn trajectory sampling in MedAgentGym, Med-Copilot achieves performance gains of +43.02% and +45.28% from offline and online reinforcement learning, respectively, demonstrating MedAgentGym as an effective training ground while establishing itself as a cost-effective, privacy-preserving alternative competitive with proprietary LLMs (gpt-4o). By offering a unified execution environment with a comprehensive benchmark and accessible, extensible training resources, MedAgentGym delivers an integrated platform to develop LLM-based coding assistants for advanced biomedical data science.

Recent advancements in long chain-of-thought (CoT) reasoning, particularly through the Group Relative Policy Optimization algorithm used by DeepSeek-R1, have led to significant interest in the potential of Reinforcement Learning with Verifiable Rewards (RLVR) for Large Language Models (LLMs). While RLVR promises to improve reasoning by allowing models to learn from free exploration, there remains debate over whether it truly enhances reasoning abilities or simply boosts sampling efficiency. This paper systematically investigates the impact of RLVR on LLM reasoning. We revisit Pass@K experiments and demonstrate that RLVR can extend the reasoning boundary for both mathematical and coding tasks. This is supported by our introduction of a novel evaluation metric, CoT-Pass@K, which captures reasoning success by accounting for both the final answer and intermediate reasoning steps. Furthermore, we present a theoretical framework explaining RLVR’s incentive mechanism, demonstrating how it can encourage correct reasoning even when rewards are based solely on answer correctness. Our analysis of RLVR’s training dynamics reveals that it incentivizes correct reasoning early in the process, with substantial improvements in reasoning quality confirmed through extensive evaluations. These findings provide strong evidence of RLVR’s potential to enhance LLM reasoning, offering valuable insights into its mechanisms and performance improvements.

Data-Free Robustness Distillation (DFRD) aims to transfer the robustness from the teacher to the student without accessing the training data. While existing methods focus on overall robustness, they overlook the robust fairness issues, leading to severe disparity of robustness across different categories. In this paper, we find two key problems: (1) student model distilled with equal class proportion data behaves significantly different across distinct categories; and (2) the robustness of student model is not stable across different attacks target. To bridge these gaps, we present the first Fairness Enhanced data-free Robustness Distillation (FERD) framework to adjust the proportion and distribution of adversarial examples. For the proportion, FERD adopts a robustness guided class reweighting strategy to synthesize more samples for the less robust categories, thereby improving robustness of them. For the distribution, FERD generates complementary data samples for advanced robustness distillation. It generates Fairness-Aware Examples (FAEs) by enforcing a uniformity constraint on feature-level predictions, which suppress the dominance of class-specific non-robust features, providing a more balanced representation across all categories. Then, FERD constructs Uniform-Target Adversarial Examples (UTAEs) from FAEs by applying a uniform target class constraint to avoid biased attack directions, which distribute the attack targets across all categories and prevents overfitting to specific vulnerable categories. Extensive experiments on three public datasets show that FERD achieves state-of-the-art worst-class robustness under all adversarial attack (e.g., the worst-class robustness under FGSM and AutoAttack are improved by 15.1% and 6.4% using MobileNetV2 on CIFAR-10), demonstrating superior performance in both robustness and fairness aspects. Our code is available at: [https://anonymous.4open.science/r/FERD-2A48/](https://anonymous.4open.science/r/FERD-2A48/).

Vision-language-action models (VLAs) trained on large-scale robotic datasets have demonstrated strong performance on manipulation tasks, including bimanual tasks. However, because most public datasets focus on single-arm demonstrations, adapting VLAs for bimanual tasks typically requires substantial additional bimanual data and fine-tuning. To address this challenge, we introduce TwinVLA, a modular framework that composes two copies of a pretrained single-arm VLA into a coordinated bimanual VLA. Unlike monolithic cross-embodiment models trained on mixtures of single-arm and bimanual data, TwinVLA improves both data efficiency and performance by composing pretrained single-arm policies. Across diverse bimanual tasks in real-world and simulation settings, TwinVLA outperforms a comparably-sized monolithic RDT-1B model without requiring any bimanual pretraining. Furthermore, it narrows the gap to state-of-the-art model, $\pi_0$ which rely on extensive proprietary bimanual data and compute cost. These results establish our modular composition approach as a data-efficient and scalable path toward high-performance bimanual manipulation, leveraging public single-arm data.

Degradation-agnostic image restoration aims to handle diverse corruptions with one unified model, but faces fundamental challenges in balancing efficiency and performance across different degradation types. Existing approaches either sacrifice efficiency for versatility or fail to capture the distinct representational requirements of various degradations. We present MIRAGE, an efficient framework that addresses these challenges through two key innovations. First, we propose a channel-wise functional decomposition that systematically repurposes channel redundancy in attention mechanisms by assigning CNN, attention, and MLP branches to handle local textures, global context, and channel statistics, respectively. This principled decomposition enables degradation-agnostic learning while achieving superior efficiency-performance trade-offs. Second, we introduce manifold regularization that performs cross-layer contrastive alignment in Symmetric Positive Definite (SPD) space, which empirically improves feature consistency and generalization across degradation types. Extensive experiments across five degradation settings demonstrate that MIRAGE achieves state-of-the-art performance with remarkable efficiency, outperforming existing methods in both single and mixed degradation scenarios while showing strong zero-shot generalization to unseen domains.

Standard human preference-based alignment methods, such as Reinforcement Learning from Human Feedback (RLHF), are a cornerstone technology for aligning Large Language Models (LLMs) with human values. However, these methods are all underpinned by a strong assumption that the collected preference data is clean and that all observed labels are equally reliable. In reality, large-scale preference datasets contain substantial label noise due to annotator errors, inconsistent instructions, varying expertise, and even adversarial or low-effort feedback. This creates a discrepancy between the recorded data and the ground-truth preferences, which can misguide the model and degrade its performance. To address this challenge, we introduce Robust Preference Optimization (RPO). RPO employs an Expectation-Maximization algorithm to infer the posterior probability of each label’s correctness, which is used to adaptively re-weigh each data point in the training loss to mitigate noise. We further generalize this approach by establishing a theoretical link between arbitrary preference losses and their corresponding probabilistic models. This generalization enables the systematic transformation of existing alignment algorithms into their robust counterparts, elevating RPO from a specific algorithm to a general framework for robust preference alignment. Theoretically, we prove that under the condition of a perfectly calibrated model, RPO is guaranteed to converge to the true noise level of the dataset. Our experiments demonstrate RPO’s effectiveness as a general framework, consistently enhancing four state-of-the-art alignment algorithms (DPO, IPO, SimPO, and CPO). When applied to Mistral and Llama 3 models, the RPO-enhanced methods improve AlpacaEval 2 win rates by up to 7.0 percentage points over their respective baselines.

The success of large pretrained Transformers is closely tied to tokenizers, which convert raw input into discrete symbols. Extending these models to graph-structured data remains a significant challenge. In this work, we introduce a graph tokenization framework that generates sequential representations of graphs by combining reversible graph serialization, which preserves graph information, with Byte Pair Encoding (BPE), a widely adopted tokenizer in large language models (LLMs). To better capture structural information, the graph serialization process is guided by global statistics of graph substructures, ensuring that frequently occurring substructures appear more often in the sequence and can be merged by BPE into meaningful tokens. Empirical results demonstrate that the proposed tokenizer enables Transformers such as BERT to be directly applied to graph benchmarks without architectural modifications. The proposed approach achieves state-of-the-art results on 12 benchmark datasets and frequently outperforms both graph neural networks and specialized graph transformers. This work bridges the gap between graph-structured data and the ecosystem of sequence models.

Multimodal learning is of continued interest in artificial intelligence-based applications, motivated by the potential information gain from combining different types of data. However, modalities observed in the source environment may differ from the modalities observed in the target environment due to multiple factors, including cost, hardware failure, or the perceived *informativeness* of a given modality. This shift in missingness between the source and target environment has not been carefully studied. Naive estimation of the information gain associated with including an additional modality without accounting for missingness may result in improper estimates of that modality's value in the target environment. We formalize the problem of missingness, demonstrate its ubiquity, and show that the subsequent distribution shift results in bias when the missingness process is not explicitly accounted for. To address this issue, we introduce ICYM2I (In Case You Multimodal Missed It), a framework for the evaluation of predictive performance and information gain under missingness through inverse probability weighting-based correction. We demonstrate the importance of the proposed adjustment to estimate information gain under missingness on synthetic, semi-synthetic, and real-world datasets.

Diffusion model distillation has emerged as a powerful technique for creating efficient few-step and single-step generators. Among these, Distribution Matching Distillation (DMD) and its variants stand out for their impressive performance, which is widely attributed to their core mechanism of matching the student's output distribution to that of a pre-trained teacher model. In this work, we challenge this conventional understanding. Through a rigorous decomposition of the DMD training objective, we reveal that the primary driver of few-step generation is not the distribution matching term, but a previously overlooked component we identify as \textit{\textbf{C}FG \textbf{A}ugmentation} (\textbf{CA}). We demonstrate that this term acts as the core "engine" of distillation, while the \textbf{D}istribution \textbf{M}atching (\textbf{DM}) term functions as a "regularizer" that ensures training stability and mitigates artifacts. We further validate this decoupling by demonstrating that while the DM term is a highly effective regularizer, it is not unique; simpler non-parametric constraints or GAN-based objectives can serve the same stabilizing function, albeit with different trade-offs. This decoupling of labor between CA and DM also allows a more principled analysis of the properties of both terms, leading to a more systematic and in-depth understanding. This new understanding enables us to propose principled modifications to the distillation process, such as decoupling the noise schedules for the engine and the regularizer, leading to further performance gains.

Test-time domain generalization (TTDG) methods enhance the performance of neural networks on target domains by transferring the feature distribution of target samples to approximate that of the source domain, while avoiding the computational cost associated with fine-tuning on the target domain. However, existing TTDG methods primarily rely on style transfer strategies operating at a coarse granularity, which prove ineffective for pixel-level prediction tasks such as image super-resolution (SR). To address this limitation, we propose a multi-codebook based test-time domain generalization framework (MC-TTDG). Our method leverages both domain-specific and domain-invariant codebooks to achieve fine-grained representation learning on source domains, and performs pixel-level nearest-neighbor feature matching and transfer to accurately adjust target domain features. Furthermore, we introduce a voting-based strategy for optimal domain-specific codebook selection, which improves the precision of feature transfer through multi-party consensus. Extensive experiments across diverse data distributions, and network architectures demonstrate that the proposed method effectively transfers feature distributions for SR networks. Our code is available at https://github.com/ZaizuoTang/MC-TTDG.

Existing Image Manipulation Localization (IML) methods rely heavily on task-specific designs, making them perform well only on the target IML task, while joint training on multiple IML tasks causes significant performance degradation, hindering real applications. To this end, we propose Omni-IML, the first generalist model designed to unify IML across diverse tasks. Specifically, Omni-IML achieves generalization through three key components: (1) a Modal Gate Encoder, which adaptively selects the optimal encoding modality per sample, (2) a Dynamic Weight Decoder, which dynamically adjusts decoder filters to the task at hand, and (3) an Anomaly Enhancement module that leverages box supervision to highlight the tampered regions and facilitate the learning of task-agnostic features. Beyond localization, to support interpretation of the tampered images, we construct Omni-273k, a large high-quality dataset that includes natural language descriptions of tampered artifacts. It is annotated through our automatic, chain-of-thoughts annotation technique. We also design a simple-yet-effective interpretation module to better utilize these descriptive annotations. Our extensive experiments show that our single Omni-IML model achieves state-of-the-art performance across all four major IML tasks, providing a valuable solution for practical deployment and a promising direction of generalist models in image forensics. We will release our code and dataset.

The Successor Measure (SM), a powerful method in reinforcement learning (RL), describes discounted future state distributions under a policy, and it has recently been studied using generative modeling techniques. Although SM is a powerful predictive object, it lacks compact representations tailored for online RL. To address this, we introduce Successor Flow Features (SF2), a representation learning framework that bridges SM estimation with policy optimization. SF2 leverages flow-matching generative models to approximate successor measures, while enforcing a structured linear decomposition into a time-invariant embedding and a time-dependent projection. This yields compact, policy-aware state-action features that integrate readily into standard off-policy algorithms like TD3 and SAC. Experiments on DeepMind Control Suite tasks show that SF2 improves sample efficiency and training stability compared to strong successor feature baselines. We attribute these gains to the compact representation induced by flow matching, which reduces compounding errors in long-horizon predictions.