ICLR 2026 Papers

Self-evolving Large Language Models (LLMs) offer a scalable path toward super-intelligence by autonomously generating, refining, and learning from their own experiences. However, existing methods for training such models still rely heavily on vast human-curated tasks and labels, typically via fine-tuning or reinforcement learning, which poses a fundamental bottleneck to advancing AI systems toward capabilities beyond human intelligence. To overcome this limitation, we introduce R-Zero, a fully autonomous framework that generates its own training data from scratch. Starting from a single base LLM, R-Zero initializes two independent models with distinct roles, a Challenger and a Solver. These models are optimized separately and co-evolve through interaction: the Challenger is rewarded for proposing tasks near the edge of the Solver capability, and the Solver is rewarded for solving increasingly challenging tasks posed by the Challenger. This process yields a targeted, self-improving curriculum without any pre-existing tasks and labels. Empirically, R-Zero substantially improves reasoning capability across different backbone LLMs, e.g., boosting the Qwen3-4B-Base by +6.49 on math-reasoning benchmarks and +7.54 on general-domain reasoning benchmarks.

Underwater 3D reconstruction remains challenging due to the intricate interplay between light scattering and environment dynamics. While existing methods yield plausible reconstruction with rigid scene assumptions, they struggle to capture temporal dynamics and remain sensitive to observation noise. In this work, we propose an Uncertainty-aware Dynamic Field (UDF) that jointly represents underwater structure and view-dependent medium over time. A canonical underwater representation is initialized using a set of 3D Gaussians embedded in a volumetric medium field. Then we map this representation into a 4D neural voxel space and encode spatial-temporal features by querying the voxels. Based on these features, a deformation network and a medium offset network are proposed to model transformations of Gaussians and time-conditioned updates to medium properties, respectively. To address input-dependent noise, we model per-pixel uncertainty guided by surface-view radiance ambiguity and inter-frame scene flow inconsistency. This uncertainty is incorporated into the rendering loss to suppress the noise from low-confidence observations during training. Experiments on both controlled and in-the-wild underwater datasets demonstrate our method achieves both high-quality reconstruction and novel view synthesis. Our code will be released.

Understanding long videos requires Multimodal Large Language Models (MLLMs) to grasp multi-timescale information, often organized in hierarchies. However, current long-video understanding benchmarks either overlook multi-timescale design or distribute questions targeting different timescales across different videos. This approach entangles timescales with video content, thereby hindering a clear assessment of MLLM multi-timescale performance. To address this, we introduce ScaleLong, the first benchmark to disentangle these factors by embedding questions targeting four hierarchical timescales\textemdash clip (seconds), shot (tens of seconds), event (minutes), and story (hours)\textemdash all within the same video content. This ``within-content'' multi-timescale questioning design enables direct comparison of model performance across timescales on identical videos. ScaleLong features 269 videos (avg. 86 min) from 5 main categories and 36 sub-categories, with 4–8 carefully designed questions, with at least one question targeting each timescale. Evaluating 22 MLLMs reveals a distinct U-shaped performance trend: higher accuracy at the shortest (clip) and longest (story) timescales, with a dip at intermediate levels. Furthermore, ablation studies demonstrate that increased visual token capacity consistently enhances reasoning across all timescales. ScaleLong offers a crucial fine-grained, multi-timescale benchmark for advancing MLLM capabilities in long-video understanding. The code and dataset are available at \url{https://anonymous.4open.science/r/ScaleLong-7717}.

Cardiovascular disease (CVD) is a leading cause of mortality worldwide. Electrocardiograms (ECGs) are the most widely used non-invasive tool for cardiac assessment, yet large, well-annotated ECG corpora are scarce due to cost, privacy, and workflow constraints. Generating ECGs can aid mechanistic understanding of cardiac electrical activity, enable the construction of large, heterogeneous, and unbiased datasets, and facilitate privacy-preserving data sharing. Generating realistic ECG signals from clinical context is important yet underexplored. Recent work has leveraged diffusion models for text-to-ECG generation, but two challenges remain: (i) existing methods often overlook physiological simulator knowledge of cardiac activity; and (ii) they ignore broader, experience-based clinical knowledge grounded in real-world practice. To address these gaps, we propose **SE-Diff**, a physiological simulator- and experience-enhanced diffusion model for comprehensive ECG generation. SE-Diff integrates a lightweight ordinary differential equation (ODE)–based ECG simulator into the diffusion process via a beat decoder and simulator-consistent constraints, injecting mechanistic priors that promote physiologically plausible waveforms. In parallel, we design an LLM-powered, experience retrieval–augmented strategy to inject clinical knowledge, providing stronger guidance for ECG generation. Extensive experiments on real-world ECG datasets demonstrate that SE-Diff improves both signal fidelity and text–ECG semantic alignment over baselines. We further show that simulator-based and experience-based knowledge benefit downstream ECG classification.

Despite significant advances in video synthesis, research into multi-shot video generation remains in its infancy. Even with scaled-up models and massive datasets, the shot transition capabilities remain rudimentary and unstable, largely confining generated videos to single-shot sequences. In this work, we introduce CineTrans, a novel framework for generating coherent multi-shot videos with cinematic, film-style transitions. To facilitate insights into the film editing style, we construct a multi-shot video-text dataset Cine250K with detailed shot annotations. Furthermore, our analysis of existing video diffusion models uncovers a correspondence between attention maps in the diffusion model and shot boundaries, which we leverage to design a mask-based control mechanism that enables transitions at arbitrary positions and transfers effectively in a training-free setting. After fine-tuning on our dataset with the mask mechanism, CineTrans produces cinematic multi-shot sequences while adhering to the film editing style, avoiding unstable transitions or naive concatenations. Finally, we propose specialized evaluation metrics for transition control, temporal consistency and overall quality, and demonstrate through extensive experiments that CineTrans significantly outperforms existing baselines across all criteria.

This paper studies a novel problem of out-of-distribution graph models merging, which aims to construct a generalized model from multiple graph models pre-trained on different domains with distribution discrepancy. This problem is challenging because of the difficulty in learning domain-invariant knowledge implicitly in model parameters and consolidating expertise from potentially heterogeneous GNN backbones. In this work, we propose a graph generation strategy that instantiates the mixture distribution of multiple domains. Then, we merge and fine-tune the pre-trained graph models via a MoE module and a masking mechanism for generalized adaptation. Our framework is architecture-agnostic and can operate without any source/target domain data. Both theoretical analysis and experimental results demonstrate the effectiveness of our approach in addressing the model generalization problem.

Watermarking provides an effective means for data governance. However, conventional post-editing graph watermarking approaches degrade the graph quality and involve NP-hard subroutines. Alternatively, recent approaches advocate for embedding watermarking patterns in the noisy latent during data generation from diffusion models, but remain uncharted for graph models due to the hardness of inverting the graph diffusion process. In this work, we propose CheckWate: the first watermarking framework for graph diffusion models embedding checkerboard watermark and providing polynomial time verification. To address NP-completeness due to graph isomorphism, CheckWate embeds the watermark into the latent eigenvalues, which are isomorphism-invariant. To detect the watermark through reversing the graph diffusion process, CheckWate leverages the graph eigenvectors to approximately dequantizes the discrete graph back to the continuous latent, with theoretical guarantees on the detectability and dequantization error. We further introduce a latent sparsification mechanism to enhance the robustness of CheckWate against graph modifications. We evaluate CheckWate on four datasets and four graph modification attacks, against three generation time watermark schemes. CheckWate achieves remarkable generation quality while being detectable under strong attacks such as isomorphism, whereas the baselines are unable to detect the watermark. Code available at: https://anonymous.4open.science/r/checkwate.

Proper initialisation strategy is of primary importance to mitigate gradient explosion or vanishing when training neural networks. Yet, the impact of initialisation parameters still lacks a precise theoretical understanding for several well-established architectures. Here, we propose a new initialisation for networks with sinusoidal activation functions such as \texttt{SIREN}, focusing on gradients control, their scaling with network depth, their impact on training and on generalization. To achieve this, we identify a closed-form expression for the initialisation of the parameters, differing from the original \texttt{SIREN} scheme. This expression is derived from fixed points obtained through the convergence of pre-activation distribution and the variance of Jacobian sequences. Controlling both gradients and targeting vanishing pre-activation helps preventing the emergence of inappropriate frequencies during estimation, thereby improving generalization. We further show that this initialisation strongly influences training dynamics through the Neural Tangent Kernel framework (NTK). Finally, we benchmark \texttt{SIREN} with the proposed initialisation against the original scheme and other baselines on function fitting and image reconstruction. The new initialisation consistently outperforms state-of-the-art methods across a wide range of reconstruction tasks, including those involving physics-informed neural networks.

Collaborative perception emphasizes enhancing environmental understanding by enabling multiple agents to share visual information with limited bandwidth resources. While prior work has explored the empirical trade-off between task performance and communication volume, a significant gap remains in the theoretical foundation. To fill this gap, we draw on information theory and introduce a pragmatic rate-distortion theory for multi-agent collaboration, specifically formulated to analyze performance-communication trade-off in goal-oriented multi-agent systems. This theory concretizes two key conditions for designing optimal communication strategies: supplying pragmatically relevant information and transmitting redundancy-less messages. Guided by these two conditions, we propose RDcomm, a communication-efficient collaborative perception framework that introduces two key innovations: i) task entropy discrete coding, which assigns features with task-relevant codeword-lengths to maximize the efficiency in supplying pragmatic information; ii) mutual-information-driven message selection, which utilizes mutual information neural estimation to approach the optimal redundancy-less condition. Experiments on 3D detection and BEV segmentation show that RDcomm achieves state-of-the-art accuracy on datasets DAIR-V2X, OPV2V, V2XSeq, and V2V4Real, while reducing communication volume by up to 108×. The code will be released.

Large language models (LLMs) are increasingly expected to tackle complex tasks, driven by their expanding applications and users' growing proficiency in crafting sophisticated prompts. However, as the number of explicitly stated requirements increases (particularly more than $10$ constraints), LLMs often struggle to accurately follow such complex instructions, which limits their applicability in complex real-world scenarios. To the best of our knowledge, existing datasets do not exceed 10 constraints per instance. To address this challenge, we propose RECAST, an efficient and scalable framework for synthesizing datasets where each example incorporates far more constraints than those in existing benchmarks, aiming to challenge and extend the boundaries of models’ ability to follow complex instructions. These constraints are extracted from real-world prompt-response pairs to ensure practical relevance. Using this framework, we construct RECAST-$30$K, a large-scale, high-quality dataset comprising $30$k instances spanning $19$ constraint types. Experimental results demonstrate that models fine-tuned on RECAST-30K substantially improve in following complex instructions while maintaining their general capabilities without degradation. Moreover, RECAST enables automatic verification of constraint satisfaction via rule-based validators for quantitative constraints and LLM-based validators for qualitative ones, the verifiability provided by RECAST enables the design of reward functions for reinforcement learning, which further boosts model performance on complex and challenging tasks.

Anticipating traffic accidents in real time is critical for intelligent transportation systems, yet remains challenging under edge-device constraints. We propose a lightweight spatio-temporal framework that introduces a temporally shifted distillation strategy, enabling a student model to acquire predictive temporal dynamics from a frozen image-based teacher without requiring a video pre-trained teacher. The student combines a RepMixer spatial encoding with a RWKV-inspired recurrent module for efficient long-range temporal reasoning. To enhance robustness under partial observability, we design a masking memory strategy that leverages memory retention to reconstruct missing visual tokens, effectively simulating occlusions and future events. In addition, multi-modal vision-language supervision enriches semantic grounding. Our framework achieves state-of-the-art performance on multiple real-world dashcam benchmarks while sustaining real-time inference on resource-limited platforms such as the NVIDIA Jetson Orin Nano. Remarkably, it is 3-7$\times$ smaller than leading approaches yet delivers superior accuracy and earlier anticipation, underscoring its practicality for deployment in intelligent vehicles.

Diffusion models excel at generating high-likelihood samples but often require alignment with downstream objectives. Existing fine-tuning methods for diffusion models significantly suffer from reward over-optimization, resulting in high-reward but unnatural samples and degraded diversity. To mitigate over-optimization, we propose Soft Q-based Diffusion Finetuning (SQDF), a novel KL-regularized RL method for diffusion alignment that applies a reparameterized policy gradient of a training-free, differentiable estimation of the soft Q-function. SQDF is further enhanced with three innovations: a discount factor for proper credit assignment in the denoising process, the integration of consistency models to refine Q-function estimates, and the use of an off-policy replay buffer to improve mode coverage and manage the reward-diversity trade-off. Our experiments demonstrate that SQDF achieves superior target rewards while preserving diversity in text-to-image alignment. Furthermore, in online black-box optimization, SQDF attains high sample efficiency while maintaining naturalness and diversity. Our code is available at https://anonymous.4open.science/r/SQDF-B66C

Code localization is a fundamental challenge in repository-level software engineering tasks such as bug fixing. While existing methods equip language agents with comprehensive tools/interfaces to fetch information from the repository, they overlook the critical aspect of *memory*, where each instance is typically handled from scratch assuming no prior repository knowledge. In contrast, human developers naturally build long-term repository memory, such as the functionality of key modules and associations between various bug types and their likely fix locations. In this work, we augment language agents with such memory by leveraging a repository's *commit history* - a rich yet underutilized resource that chronicles the codebase's evolution. We introduce tools that allow the agent to retrieve from a non-parametric memory encompassing recent historical commits and linked issues, as well as functionality summaries of actively evolving parts of the codebase identified via commit patterns. We demonstrate that augmenting such a memory can significantly improve LocAgent, a state-of-the-art localization framework, on both SWE-bench-verified and the more recent SWE-bench-live benchmarks. Our research contributes towards developing agents that can accumulate and leverage past experience for long-horizon tasks, more closely emulating the expertise of human developers.

We study the problem of learning the utility functions of no-regret learning agents in a repeated normal-form game. Differing from most prior literature, we introduce a principal with the power to observe the agents playing the game, send agents signals, and give agents *payments* as a function of their actions. We show that the principal can, using a number of rounds polynomial in the size of the game, learn the utility functions of all agents to any desired precision $\varepsilon > 0$, for any no-regret learning algorithms of the agents. Our main technique is to formulate a zero-sum game between the principal and the agents, where the principal's strategy space is the set of all payment functions. Finally, we discuss implications for the problem of *steering* agents to a desired equilibrium: in particular, we introduce, using our utility-learning algorithm as a subroutine, the first algorithm for steering arbitrary no-regret learning agents without prior knowledge of their utilities.

While significant research has focused on developing embodied reasoning capabilities using Vision-Language Models (VLMs) or integrating advanced VLMs into Vision-Language-Action (VLA) models for end-to-end robot control, few studies directly address the critical gap between upstream VLM-based reasoning and downstream VLA policy learning. In this work, we take an initial step toward bridging embodied reasoning with VLA policy learning by introducing **Vlaser** - a **V**ision-**L**anguage-**A**ction Model with **s**ynergistic **e**mbodied **r**easoning capability, which is a foundational vision-language model designed to integrate high-level reasoning with low-level control for embodied agents. Built upon the high-quality Vlaser-6M dataset, Vlaser achieves state-of-the-art performance across a range of embodied reasoning benchmarks—including spatial reasoning, embodied grounding, embodied QA, and task planning. Furthermore, we systematically examine how different VLM initializations affect supervised VLA fine-tuning, offering novel insights into mitigating the domain shift between internet-scale pre-training data and embodied-specific policy learning data. Based on these insights, our approach achieves state-of-the-art results on the WidowX benchmark and competitive performance on the Google Robot benchmark. We will open-source the model weights, data generation pipelines, and the full dataset to support future research.

Recent advances in Large Reasoning Models (LRMs) have demonstrated strong performance on complex tasks through long Chain-of-Thought (CoT) reasoning. However, their lengthy outputs increase computational costs and may lead to overthinking, raising challenges in balancing reasoning effectiveness and efficiency. Current methods solving this problem often compromise reasoning quality or require extensive resources. In this paper, we investigate how to reduce the generation length of LRMs with limited tuning. We analyze generation path distributions and filter generated trajectories through difficulty estimation. Subsequently, we analyze the convergence behaviors of the objectives of various preference optimization methods under a Bradley-Terry loss based framework. Based on the analysis, we propose Length Controlled Preference Optimization (LCPO) that directly balances the implicit reward related to NLL loss. LCPO can effectively learn length preference with limited data and training. Extensive experiments demonstrate that our approach significantly reduces the average output length by over 50\% across multiple benchmarks while maintaining the reasoning performance. Our work highlights the potential for computationally efficient approaches in guiding LRMs toward efficient reasoning. Code available at https://anonymous.4open.science/r/anonymous\_code\_74CD.

DNA methylation (DNAm) is a key epigenetic modification that regulates gene expression and is pivotal in development and disease. However, profiling DNAm at genome scale is challenging: of $\textasciitilde$28 million CpG sites in the human genome, only about 1–3\% are typically assayed in common datasets due to technological limitations and cost. Recent deep learning approaches, including masking-based generative Transformer models, have shown promise in capturing DNAm–gene expression relationships, but they rely on partially observed DNAm values for unmeasured CpGs and cannot be applied to completely unmeasured samples. To overcome this barrier, we introduce MethylProphet, a gene-guided, context-aware Transformer model for whole-genome DNAm inference without any measured DNAm input. MethylProphet compresses comprehensive gene expression profiles ($\textasciitilde$25K genes) through an efficient bottleneck multilayer perceptron, and encodes local CpG sequence context with a specialized DNA tokenizer. These representations are integrated by a Transformer encoder to predict site-specific methylation levels. Trained on large-scale pan-tissue whole-genome bisulfite sequencing data from ENCODE (1.6 billion CpG–sample pairs, $\textasciitilde$322 billion tokens), MethylProphet demonstrates strong performance in hold-out evaluations, accurately inferring DNAm at unmeasured CpGs and generalizing to unseen samples. Furthermore, application to TCGA pan-cancer data (chromosome 1, 9,194 samples; $\textasciitilde$450 million training pairs, 91 billion tokens) highlights its potential for pan-cancer whole-genome methylome imputation. MethylProphet offers a powerful and scalable foundation model for epigenetics, providing high-resolution methylation landscape reconstruction and advancing both biological research and precision medicine.

Graph neural networks (GNNs) are the standard for learning on graphs, yet they have limited expressive power, often expressed in terms of the Weisfeiler-Leman (WL) hierarchy or within the framework of first-order logic. In this context, topological neural networks (TNNs) have recently emerged as a promising alternative for graph representation learning. By incorporating higher-order relational structures into message-passing schemes, TNNs offer higher representational power than traditional GNNs. However, a fundamental question remains open: _what is the logical expressiveness of TNNs?_ Answering this allows us to characterize precisely which binary classifiers TNNs can represent. In this paper, we address this question by analyzing isomorphism tests derived from the underlying mechanisms of general TNNs. We introduce and investigate the power of higher-order variants of WL-based tests for combinatorial complexes, called $k$-CCWL test. In addition, we introduce the topological counting logic $TC_{k}$, an extension of standard counting logic featuring a novel pairwise counting quantifier $\exists^{N}(x_i,x_j) \varphi(x_i,x_j),$ which explicitly quantifies pairs $(x_i, x_j)$ satisfying property $\varphi$. We rigorously prove the exact equivalence: $\text{k-CCWL} \equiv \text{TC}_{k{+}2} \equiv \text{Topological }(k{+}2)\text{-pebble game}.$ These results establish a logical expressiveness theory for TNNs.

Mixture-of-Experts (MoE) has demonstrated strong performance in video understanding tasks, yet its adversarial robustness remains underexplored. Existing attack methods often treat MoE as a unified architecture, overlooking the independent and collaborative weaknesses of key components such as routers and expert modules. To fill this gap, we propose Temporal Lipschitz-Guided Attacks (TLGA) to thoroughly investigate component-level vulnerabilities in video MoE models. We first design attacks on the router, revealing its independent weaknesses. Building on this, we introduce Joint Temporal Lipschitz-Guided Attacks (J-TLGA), which collaboratively perturb both routers and experts. This joint attack significantly amplifies adversarial effects and exposes the Achilles’ Heel (collaborative weaknesses) of the MoE architecture. Based on these insights, we further propose Joint Temporal Lipschitz Adversarial Training (J-TLAT). J-TLAT performs joint training to further defend against collaborative weaknesses, enhancing component-wise robustness. Our framework is plug-and-play and reduces inference cost by more than 60% compared with dense models. It consistently enhances adversarial robustness across diverse video datasets and model architectures, effectively mitigating both the independent and collaborative weaknesses of MoE.