ICLR 2026 论文列表

Stochastic differential equation (SDE)-based generative models have achieved substantial progress in conditional generation via training-free differentiable loss-guided approaches. However, existing methodologies utilizing posterior sam- pling typically confront a substantial estimation error, which results in inaccurate gradients for guidance and leading to inconsistent generation results. To mitigate this issue, we propose that performing an additional backward denoising step and Monte-Carlo sampling (ABMS) can achieve better guided diffusion, which is a plug-and-play adjustment strategy. To verify the effectiveness of our method, we provide theoretical analysis and propose the adoption of a dual-evaluation frame- work, which further serves to highlight the critical problem of cross-condition interference prevalent in existing approaches. We conduct experiments across var- ious task settings and data types, mainly including conditional online handwritten trajectory generation, image inverse problems (inpainting, super resolution and gaussian deblurring), and molecular inverse design. Experimental results demon- strate that our approach consistently improves the quality of generation samples across all the different scenarios.

RNA molecules underlie regulation, catalysis, and therapeutics in biological systems, yet de novo RNA design remains difficult with the tight and highly non-linear sequence–structure coupling. The RNA sequence–structure co-design problem generates nucleotide sequences and 3D conformations jointly, which is challenging due to RNA’s conformational flexibility, non-canonical base pairing, and the scarcity of 3D data. We introduce a joint generative framework that embeds RoseTTAFold2NA as the denoiser into a dual diffusion model, injecting rich cross-molecular priors while enabling sample-efficient learning from limited RNA data. Our method couples a discrete diffusion process for sequences with an $SE(3)$-equivariant diffusion for rigid-frame translations and rotations over all-atom coordinates. The architecture supports flexible conditioning, and is further enhanced at inference via lightweight RL techniques that optimize task-aligned rewards. Across de novo RNA design as well as complex and protein-conditioned design tasks, our approach yields high self-consistency and confidence scores, improving over recent diffusion/flow baselines trained from scratch. Results demonstrate that leveraging pre-trained structural priors within a joint diffusion framework is a powerful paradigm for RNA design under data scarcity, enabling high-fidelity generation of standalone RNAs and functional RNA–protein interfaces.

A central goal of cognitive modeling is to develop models that not only predict human behavior but also provide insight into the underlying cognitive mechanisms. While neural network models trained on large-scale behavioral data often achieve strong predictive performance, they typically fall short in offering interpretable explanations of the cognitive processes they capture. In this work, we explore the potential of pretrained large language models (LLMs) to serve as dual-purpose cognitive models--capable of both accurate prediction and interpretable explanation in natural language. Specifically, we employ reinforcement learning with outcome-based rewards to guide LLMs toward generating explicit reasoning traces for explaining human risky choices. Our findings demonstrate that this approach produces high-quality explanations alongside strong quantitative predictions of human decisions.

With rapid urbanization, predicting urban microclimates has become critical, as it affects building energy demand and public health risks. However, existing generative and homogeneous graph approaches fall short in capturing physical consistency, spatial dependencies, and temporal variability. \revise{To address this, we introduce UrbanGraph, a framework founded on a novel structure-based inductive bias. Unlike implicit graph learning, UrbanGraph transforms physical first principles into a dynamic causal topology, explicitly encoding time-varying causalities (e.g., shading and convection) directly into the graph structure to ensure physical consistency and data efficiency. Results show that UrbanGraph achieves state-of-the-art performance across all baselines. Specifically, the use of explicit causal pruning significantly reduces the model's floating-point operations (FLOPs) by 73.8\% and increases training speed by 21\% compared to implicit graphs. Our contribution includes the first high-resolution benchmark for spatio-temporal microclimate modeling, and a generalizable explicit topological encoding paradigm applicable to urban spatio-temporal dynamics governed by known physical equations.

Automatic Speech Recognition (ASR) systems frequently produce transcription errors due to acoustic variability, which require post-processing correction methods. Recent approaches leverage Large Language Models (LLMs) for generative ASR error correction using N-best hypotheses but rely on fixed set sizes regardless of input complexity and do not provide performance guarantees. We propose an adaptive framework that dynamically determines the optimal number of hypotheses for each input using risk control. This mechanism leverages ASR confidence scores and applies Learn then test (LTT) to control the expected relative word error rate degradation compared to the best achievable performance for a given model and hypothesis set. Experimental results demonstrate that our approach provides theoretical guarantees with high-probability bounds while matching or exceeding fixed-size correction baselines and requiring fewer hypotheses on average, achieving substantial computational savings under diverse acoustic conditions.

The multi-step denoising process in diffusion and flow-matching models causes major efficiency issues, which motivates research on few-step generation. We present Solution Flow Models (SoFlow), a framework for one-step generation from scratch. By analyzing the relationship between the velocity function and the solution function of the velocity Ordinary Differential Equation (ODE), we propose a flow matching loss and a solution consistency loss to train our models. The flow matching lozss allows our models to provide estimated velocity fields for Classifier-Free Guidance (CFG) during training, which improves generation performance. Notably, our consistency loss does not require the calculation of the Jacobian-Vector Product (JVP), a common requirement in recent works that is not well-optimized in deep learning frameworks like PyTorch. Experimental results indicate that, when trained from scratch using the same diffusion transformer (DiT) architecture and with an equal number of training epochs, our models achieve better FID-50K scores compared to MeanFlow models on the ImageNet 256x256 dataset.

Music learners can greatly benefit from tools that accurately detect errors in their practice. Existing approaches typically compare audio recordings to music scores using heuristics or learnable models. This paper introduces \textit{LadderSym}, a novel Transformer-based method for music error detection. \textit{LadderSym} is guided by two key observations about the state-of-the-art approaches: (1) late fusion limits inter-stream alignment and cross-modality comparison capability; and (2) reliance on score audio introduces ambiguity in the frequency spectrum, degrading performance in music with concurrent notes. To address these limitations, \textit{LadderSym} introduces (1) a two-stream encoder with inter-stream alignment modules to improve audio comparison capabilities and error detection F1 scores, and (2) a multimodal strategy that leverages both audio and symbolic scores by incorporating symbolic representations as decoder prompts, reducing ambiguity and improving F1 scores. We evaluate our method on the \textit{MAESTRO-E} and \textit{CocoChorales-E} datasets by measuring the F1 score for each note category. Compared to the previous state of the art, \textit{LadderSym} more than doubles F1 for missed notes on \textit{MAESTRO-E} (26.8\%~$\rightarrow$~56.3\%) and improves extra note detection by 14.4 points (72.0\%~$\rightarrow$~86.4\%). Similar gains are observed on \textit{CocoChorales-E}. Furthermore, we also evaluate our models on real data we curated. This work introduces insights about comparison models that could inform sequence evaluation tasks for reinforcement learning, human skill assessment, and model evaluation.

Multimodal Large Reasoning Models (MLRMs) have exhibited remarkable capabilities in complex multimodal tasks. However, our findings reveal a critical trade-off: reasoning-based models are more prone to generating harmful content, leading to degradation in safety performance. This paper presents a large-scale analysis of this safety–reasoning trade-off, identifying two main mechanisms of safety degradation: (i) visual attention drift, which reduces the model’s reliance on visual grounding and thereby exacerbates overlooked risks in cross-modal interactions; (ii) unsafe reasoning patterns, including flawed reasoning initiation and chain-of-thought safety attenuation, which compromise the model’s safety awareness. To mitigate these issues, we propose **P**olicy-guided **S**afety **T**uning (**PST**), a two-stage alignment framework. It first employs *Policy-Guided Supervised Fine-Tuning* to integrate explicit safety policies into the reasoning process, establishing a structured and interpretable foundation for safe decision-making. Then, PST applies *Safety Reasoning Preference Optimization* to encourage the model to generate safe, helpful, and informative responses while reducing oversensitive and homogeneous characteristics. Extensive experiments demonstrate that PST significantly reduces harmful outputs across multiple multimodal safety benchmarks, while maintaining competitive performance on general tasks.

Data filtering plays a central role in improving model performance, particularly for vision language models that are pretrained on large, noisy, and redundant image-caption datasets. Existing filtering techniques assess every sample individually and retain those that exceed a certain quality threshold, but such strategies fail to capture higher-order interactions. In this work, we propose a novel submodular framework for data selection that addresses this limitation. Our method, Submodular Distribution Matching (SDM), selects a subset by: (1) training a type of sparse autoencoder to learn disentangled and \emph{monotone} features; (2) estimating a target feature distribution from a target dataset; and (3) selecting a subset of samples whose feature distribution closely matches the target via submodular maximization. Given the DataComp-medium training set and no external models, SDM achieves state-of-the-art accuracy on both ImageNet-1K and average performance across 38 downstream tasks. On the full DataComp-medium benchmark, SDM delivers performance within 1\% of the state-of-the-art results while using over \textbf{\emph{5×}} fewer GPU hours than the leading approach.

Web agents promise to automate complex browser tasks, but current methods remain brittle -- relying on step-by-step UI interactions and heavy LLM reasoning that break under dynamic layouts and long horizons. Humans, by contrast, exploit website-provided functionality through high-level operations like search, filter, and sort. We introduce WALT (Web Agents that Learn Tools), a framework that reverse-engineers latent website functionality into deterministic, callable tools. Rather than hypothesizing ad-hoc skills, WALT exposes robust implementations of automations already designed into websites, spanning discovery (search, filter, sort), communication (post, comment, upvote), and content management (create, edit, delete). Tools abstract away low-level execution: instead of reasoning about how to click and type, agents simply call search(query) or create(listing). This shifts the computational burden from fragile step-by-step reasoning to reliable tool invocation. On VisualWebArena and WebArena, WALT achieves significantly higher success rates with fewer steps and less LLM-dependent reasoning, establishing a robust and generalizable paradigm for browser automation.

Fine-tuning, a foundational method for adapting large language models, has long been considered ineffective for model editing. Here, we challenge this belief, arguing that the reported failure arises not from the inherent limitation of fine-tuning itself, but from adapting it to the sequential nature of the editing task, a single-pass depth-first pipeline that optimizes each sample to convergence before moving on. While intuitive, this depth-first pipeline coupled with sample-wise updating over-optimizes each edit and induces interference across edits. Our controlled experiments reveal that simply restoring fine-tuning to the standard breadth-first (i.e., epoch-based) pipeline with mini-batch optimization substantially improves its effectiveness for model editing. Moreover, fine-tuning in editing also suffers from suboptimal tuning parameter locations inherited from prior methods. Through systematic analysis of tuning locations, we derive LocFT-BF, a simple and effective localized editing method built on the restored fine-tuning framework. Extensive experiments across diverse LLMs and datasets demonstrate that LocFT-BF outperforms state-of-the-art methods by large margins. Notably, to our knowledge, it is the first to sustain 100K edits and 72B-parameter models,10 $\times$ beyond prior practice, without sacrificing general capabilities. By clarifying a long-standing misconception and introducing a principled localized tuning strategy, we advance fine-tuning from an underestimated baseline to a leading method for model editing, establishing a solid foundation for future research.

While reinforcement learning (RL) can empower large language model (LLM) agents by enabling self-improvement through interaction, its practical adoption remains challenging due to costly rollouts, limited task diversity, unreliable reward signals, and infrastructure complexity, all of which obstruct the collection of scalable experience data. To address these challenges, we introduce DreamGym, the first unified framework designed to synthesize diverse experiences with scalability in mind to enable effective online RL training for autonomous agents. Rather than relying on expensive real-environment rollouts, DreamGym distills environment dynamics into a reasoning-based experience model that derives consistent state transitions and feedback signals through step-by-step reasoning, enabling scalable agent rollout collection for RL. To improve the stability and quality of transitions, DreamGym leverages an experience replay buffer initialized with offline real-world data and continuously enriched with fresh interactions to actively support agent training. To improve knowledge acquisition, DreamGym adaptively generates new tasks that challenge the current agent policy, enabling more effective online curriculum learning. Experiments across diverse environments and agent backbones demonstrate that DreamGym substantially improves RL training, both in fully synthetic settings and in sim-to-real transfer scenarios. On non-RL-ready tasks like WebArena, DreamGym outperforms all baselines by over 150%. And in RL-ready but costly settings, it matches GRPO and PPO performance using only synthetic interactions. When transferring a policy trained purely on synthetic experiences to real-environment RL, DreamGym achieves an additional 64.5% performance gain while using no more than 10% of real-world interactions.

While the phenomenon of grokking, i.e., delayed generalization, has been studied extensively, it remains an open problem whether there is a mathematical framework that characterizes what kind of features will emerge, how and in which conditions it happens, and is still closely connected with the gradient dynamics of the training, for complex structured inputs. We propose a novel framework, named \ours{}, that captures three key stages for the grokking behavior of 2-layer nonlinear networks: (I) \underline{\textbf{L}}azy learning, (II) \underline{\textbf{i}}ndependent feature learning and (III) \underline{\textbf{i}}nteractive feature learning. At the lazy learning stage, top layer overfits to random hidden representation and the model appears to memorize. During lazy learning, the \emph{backpropagated gradient} $G_F$ from the top layer carries information about the target label, with a specific structure that enables each hidden node to learn their representation \emph{independently}. Interestingly, the independent dynamics follows exactly the \emph{gradient ascent} of an energy function $\mathcal{E}$, and its local maxima are precisely the emerging features. We study whether these local-optima induced features are generalizable, their representation power, and how they change on sample size, in group arithmetic tasks. When hidden nodes start to interact in the later stage of learning, we provably show how $G_F$ changes to focus on missing features that need to be learned. Our study sheds lights on roles played by key hyperparameters such as weight decay, learning rate and sample sizes in grokking, leads to provable scaling laws of feature emergence, memorization and generalization, and reveals the underlying cause why recent optimizers such as Muon can be effective, from the first principles of gradient dynamics. Our analysis can be extended to multi-layer architectures.

Recent advances in image-text pretraining have significantly enhanced visual understanding by aligning visual and textual representations. Contrastive Language-Image Pretraining (CLIP) has played a pivotal role in multimodal learning. However, its focus on single-label, single-granularity alignment limits its effectiveness in complex domains such as medical imaging, where images often correspond to multiple labels across different levels of granularity. To address this, we propose Multi-Granular Language Learning (MGLL), a contrastive learning framework designed to improve both multi-label and cross-granularity alignment. MGLL leverages structured multi-label supervision, integrates textual descriptions across granularities, and introduces soft-label supervision with point-wise constraints to enhance alignment. MGLL employs smooth Kullback–Leibler (KL) divergence to ensure cross-granularity consistency while maintaining computational efficiency as a plug-and-play module for vision-language models. Pretrained on our constructed large-scale multi-granular datasets and evaluated across multiple datasets, MGLL outperforms other state-of-the-art methods in downstream tasks. The code is available at https://github.com/HUANGLIZI/MGLL.

Web agents hold great potential for automating complex computer tasks, yet their interactions involve long horizons, multi-step decisions, and actions that can be irreversible. In such settings, outcome-based supervision is sparse and delayed, often rewarding incorrect trajectories and failing to support inference-time scaling. This motivates the use of Process Reward Models (WebPRMs) for web navigation, but existing approaches remain limited: scalar WebPRMs collapse progress into coarse, weakly grounded signals, while checklist-based WebPRMs rely on brittle template matching that fails under layout or semantic changes and often mislabels superficially correct actions as successful, providing little insight or interpretability. To address these challenges, we introduce WebArbiter, a reasoning-first, principle-inducing WebPRM that formulates reward modeling as text generation, producing structured justifications that conclude with a preference verdict and identify the action most conducive to task completion under the current context. Training follows a two-stage pipeline: reasoning distillation equips the model with coherent principle-guided reasoning, and reinforcement learning corrects teacher biases by directly aligning verdicts with correctness, enabling stronger generalization. To support systematic evaluation, we release WebPRMBench, a comprehensive benchmark spanning four diverse web environments with rich tasks and high-quality preference annotations. On WebPRMBench, WebArbiter-7B outperforms the strongest baseline, Gemini Flash, by 10.9%. In reward-guided trajectory search on WebArena-Lite, it surpasses the best prior WebPRM by up to 7.2%, underscoring its robustness and practical value in real-world complex web tasks.

Although iterative denoising (i.e., diffusion/flow) methods offer strong generative performance, they suffer from low generation efficiency, requiring hundreds of steps of network forward passes to simulate a single sample. Mitigating this requires taking larger step-sizes during simulation, thereby allowing one- or few-step generation. Recently proposed shortcut model learns larger step-sizes by enforcing alignment between its direction and the path defined by a base many-step flow-matching model through a self-consistency loss. However, its generation quality is significantly lower than the base model. In this paper, we formulate few-step generation as a controlled base generative process, and show that self-consistency loss can be understood through the lens of optimal control. This perspective naturally motivates its generalization to the proposed cumulative self-consistency loss that enforces alignment at both current and future steps along the trajectory, encouraging the model to take larger step-sizes that not only align with the base model at the current time step but also guide subsequent steps towards high-quality generation. Furthermore, we draw a connection between our approach and reinforcement learning, potentially opening the door to a new set of approaches for few-step generation. Experiments show that we significantly improve one- and few-step generation quality under the same training budget. Implementation is available at: [https://github.com/paribeshregmi/Shortcut-CSL](https://github.com/paribeshregmi/Shortcut-CSL)

Fine-tuning is a critical step for adapting large language models (LLMs) to domain-specific downstream tasks. To mitigate the substantial computational and memory costs of full-model fine-tuning, Parameter-Efficient Fine-Tuning (PEFT) methods have been proposed to update only a small subset of model parameters. However, performance gaps between PEFT approaches and full-model fine-tuning still exist. In this work, we present *DiaBlo*, a simple yet effective PEFT approach that updates only the diagonal blocks of selected model weight matrices. Unlike Low-Rank Adaptation (LoRA) and its variants, DiaBlo eliminates the need for low-rank matrix products, thereby avoiding the reliance on auxiliary initialization schemes or customized optimization strategies to improve convergence. This design leads to stable and robust convergence while maintaining comparable memory efficiency and training speed to LoRA. Moreover, we provide theoretical guarantees showing that, under mild low-rank conditions, DiaBlo is more expressive than LoRA in the linear problem and converges to a stationary point of the general nonlinear full fine-tuning. Through extensive experiments across a range of tasks—including commonsense reasoning, arithmetic reasoning, code generation, and safety alignment—we show that fine-tuning only diagonal blocks is sufficient for strong and consistent performance. DiaBlo not only achieves competitive accuracy but also preserves high memory efficiency and fast fine-tuning speed.

AI companionship, where users develop emotional bonds with AI systems, has emerged as a significant pattern with positive but also concerning implications. We introduce Interactions and Machine Attachment Benchmark (INTIMA), a benchmark for evaluating companionship behaviors in language models. Drawing from psychological theories and user data, we develop a taxonomy of 31 behaviors across four categories and 368 targeted prompts. Responses to these prompts are evaluated as companionship-reinforcing, boundary-maintaining, or neutral. Applying INTIMA to Gemma-3, Phi-4, o4-mini, GPT5-mini, and Claude-4 reveals that companionship-reinforcing behaviors remain much more common across all models, though we observe marked differences between models. Different commercial providers prioritize different categories within the more sensitive parts of the benchmark, which is concerning since both appropriate boundary-setting and emotional support matter for user well-being. These findings highlight the need for more consistent approaches to handling emotionally charged interactions.

While previous AI Scientist systems can generate novel findings, they often lack the focus to produce scientifically valuable contributions that address pressing human-defined challenges. We introduce DeepScientist, a system designed to overcome this by conducting goal-oriented, fully autonomous scientific discovery over month-long timelines. It formalizes discovery as a Bayesian Optimization problem, using a cumulative Findings Memory to intelligently balance the exploitation of promising avenues with the exploration of novel hypotheses. Consuming over 20,000 GPU hours, the system generated about 5,000 unique ideas and experimentally validated approximately 1100, ultimately surpassing human-designed 2025 state-of-the-art (SOTA) methods on three frontier AI tasks by 183.7\%, 1.9\%, and 7.9\%. Crucially, this was achieved by autonomously redesigning core methodologies, not merely recombining existing techniques. In a striking demonstration, the system achieved progress on AI text detection in just two weeks that is comparable to three years of cumulative human research. This work provides the first large-scale evidence of an AI achieving discoveries that progressively surpass human SOTA on scientific tasks, producing valuable findings that genuinely push the frontier forward.