ICLR 2026 论文列表

Current research in multimodal models faces a key challenge where enhancing generative capabilities often comes at the expense of understanding, and vice versa. We analyzed this trade-off and identify the primary cause might be the potential conflict between generation and understanding, which creates a competitive dynamic within the model. To address this, we propose the Reason-Reflect-Refine (R3) framework. This innovative algorithm re-frames the single-step generation task into a multi-step process of "generate-understand-regenerate". By explicitly leveraging the model's understanding capability during generation, we successfully mitigate the optimization dilemma, achieved stronger generation results and improved understanding ability which are related to the generation process. This offers valuable insights for designing next-generation unified multimodal models.

Large language model (LLM) agents exhibit strong mathematical problem-solving abilities and can even solve International Mathematical Olympiad (IMO) level problems with the assistance of formal proof systems. However, due to weak heuristics for auxiliary constructions, AI for geometry problem solving remains dominated by expert models such as AlphaGeometry 2, which rely heavily on large-scale data synthesis and search for both training and evaluation. In this work, we make the first attempt to build a medalist-level LLM agent for geometry and present InternGeometry. InternGeometry overcomes the heuristic limitations in geometry by iteratively proposing propositions and auxiliary constructions, verifying them with a symbolic engine, and reflecting on the engine’s feedback to guide subsequent proposals. A dynamic memory mechanism enables InternGeometry to conduct more than two hundred interactions with the symbolic engine per problem. To further accelerate learning, we introduce Complexity-Boosting Reinforcement Learning (CBRL), which gradually increases the complexity of synthesized problems across training stages. Built on InternThinker-32B, InternGeometry solves 44 of 50 IMO geometry problems (2000–2024), exceeding the average gold medalist score (40.9), using only 13K training examples, just $0.004\%$ of the data used by AlphaGeometry 2, demonstrating the potential of LLM agents on expert-level geometry tasks. InternGeometry can also propose novel auxiliary constructions for IMO problems that do not appear in human solutions. We will release the model, data, and symbolic engine to support future research.

Machine learning promises to revolutionize drug discovery, but its "black-box" nature and narrow focus limit adoption by experts. While Large Language Models (LLMs) offer a path forward with their broad knowledge and interactivity, existing methods remain data-intensive and lack transparent reasoning. To address these issues, we present DrugTrail, an LLM-based framework for explainable drug discovery that integrates structured reasoning trajectories with a Druggability‑Tailored Preference Optimization (DTPO) strategy. It not only introduces structured reasoning traces to articulate the "how" and "why" behind its conclusions but also serve to guide task-specific reasoning pathways within the LLM's vast knowledge space, thereby enhancing its interpretability and reliability of its final outputs. Furthermore, based on the fact that optimizing for binding affinity alone does not equate to optimizing for druggability, DTPO explicitly moves beyond single-metric optimization and opens up a broader search space that balances affinity with other essential factors. Extensive experiments demonstrate the effectiveness of our approach and its generalizability to a wider range of biomolecular optimization domains, bridging the gap between LLM reasoning capabilities and trustworthy AI-assisted drug discovery.

The Wasserstein–Fisher–Rao (WFR) metric extends dynamic optimal transport (OT) by coupling displacement with change of mass, providing a principled geometry for modeling unbalanced snapshot dynamics. Existing WFR solvers, however, are often unstable, computationally expensive, and difficult to scale. Here we introduce **WFR Flow Matching (WFR-FM)**, a simulation-free training algorithm that unifies flow matching with dynamic unbalanced OT. Unlike classical flow matching which regresses only a transport vector field, WFR-FM simultaneously regresses a vector field for displacement and a scalar growth rate function for birth–death dynamics, yielding continuous flows under the WFR geometry. Theoretically, we show that minimizing the WFR-FM loss exactly recovers WFR geodesics. Empirically, WFR-FM yields more accurate and robust trajectory inference in single-cell biology, reconstructing consistent dynamics with proliferation and apoptosis, estimating time-varying growth fields, and applying to generative dynamics under imbalanced data. It outperforms state-of-the-art baselines in efficiency, stability, and reconstruction accuracy. Overall, WFR-FM establishes a unified and efficient paradigm for learning dynamical systems from unbalanced snapshots, where not only states but also mass evolve over time.

Out-of-distribution (OOD) generalization is a defining hallmark of truly robust and reliable machine learning systems. Recently, it has been empirically observed that existing OOD generalization methods often underperform on real-world tabular data, where hidden confounding shifts drive distribution changes that boosting models handle more effectively. Part of boosting’s success is attributed to variance reduction, handling missing variables, feature selection, and connections to multicalibration. This paper uncovers a crucial reason behind its success in OOD generalization: boosting’s ability to infer stable environments robust to hidden confounding shifts and maximize predictive performance within those environments. This paper introduces an information-theoretic notion called $\alpha$-predictive sufficiency and formalizes its link to OOD generalization under hidden confounding. We show that boosting implicitly identifies suitable environments and produces an $\alpha$-predictive sufficient predictor. We validate our theoretical results through synthetic and real-world experiments and show that boosting achieves robust performance by identifying these environments and maximizing the association between predictions and true outcomes.

Reinforcement Learning (RL) has demonstrated significant success in optimizing complex control and planning problems. However, scaling RL to real-world applications with multiple, potentially conflicting requirements requires an effective handling of constraints. We propose a novel approach to constraint satisfaction in RL algorithms, focusing on incrementally improving policy safety rather than directly projecting the policy onto a feasible region. We accomplish this by first solving a nonparametric surrogate problem which is guaranteed to contract towards the feasible set, and then cloning that solution into a neural network policy. As a result, our approach improves stability, particularly during early training stages, when the policy lacks knowledge of constraint boundaries. We provide general theoretical results guaranteeing convergence to the safe set for this class of incremental systems. Notably, even the simplest algorithm produced by our theory produces comparable or superior performance when compared to highly tuned constrained RL baselines in challenging constrained environments.

Despite their fundamental role, it remains unclear what properties could make tokenizers more effective for generative modeling. We observe that modern generative models share a conceptually similar training objective---reconstructing clean signals from corrupted inputs, such as signals degraded by Gaussian noise or masking---a process we term \emph{denoising}. Motivated by this insight, we propose aligning tokenizer embeddings directly with the downstream denoising objective, encouraging latent embeddings that remain reconstructable even under significant corruption. To achieve this, we introduce the Latent Denoising Tokenizer (\method), a simple yet highly effective tokenizer trained to reconstruct clean images from latent embeddings corrupted via interpolative noise or random masking. Extensive experiments on class-conditioned (ImageNet $256\times256$ and $512\times512$) and text-conditioned (MSCOCO) image generation benchmarks demonstrate that our \method consistently improves generation quality across \textit{six} representative generative models compared to prior tokenizers. Our findings highlight denoising as a fundamental design principle for tokenizer development, and we hope it could motivate new perspectives for future tokenizer design. Code is available at: https://github.com/Jiawei-Yang/DeTok

This paper aims to develop a lower prediction bound (LPB) for survival time across different treatments in the general right-censored setting. Although previous methods have utilized conformal prediction to construct the LPB, their resulting prediction sets provide only probably approximately correct (PAC)–type miscoverage guarantees rather than exact ones. To address this problem, we propose a new calibration procedure under the potential outcome framework. Under the strong ignorability assumption, we propose a reweighting scheme that can transform the problem into a weighted conformal inference problem, allowing an LPB to be obtained via quantile regression with an exact miscoverage guarantee. Furthermore, our procedure is doubly robust against model misspecification. Empirical evaluations on synthetic and real-world clinical data demonstrate the validity and informativeness of our constructed LPBs, which indicate the potential of our analytical benchmark for comparing and selecting personalized treatments.

Decoding complex auditory experiences from non-invasive EEG is a rapidly emerging field that holds significant promise for advancing both fundamental neuroscience and human-machine interaction technologies. Recent developments in EEG foundation models have yielded powerful neural representations that are promising for auditory decoding. However, the effectiveness of these models remains fundamentally constrained by their limited integration with acoustic stimulus information. Specifically, the lack of deep coupling between neural signals and auditory inputs hampers the models’ ability to generalize effectively across diverse auditory tasks. To bridge this gap, we introduce MindMix, a multimodal foundation model designed to bridge the gap between unimodal EEG foundations and task-specific auditory decoders. MindMix employs a two-stage training strategy: first, a high-capacity EEG encoder is pre-trained on over 3,000 hours of EEG data to learn generalized EEG features that can transfer across tasks and subjects. Second, the model learns the neural-acoustic mapping using over 100 hours of paired data, facilitated by our novel Cross-Attention Low-Rank Alignment module, which facilitates fine-grained, cross-modal information integration. Experimental results demonstrate that MindMix substantially surpassing existing baselines across a range of auditory decoding tasks, including auditory attention decoding, auditory emotion recognition, and cross-modal retrieval. This work thus establishes a foundation for future research in multimodal brain decoding and auditory brain-computer interfaces. Our code is available at https://anonymous.4open.science/r/MindMix-654B/.

Learning with guidance has proven effective across a wide range of machine learning systems. Guidance may, for example, come from annotated datasets in supervised learning, pseudo-labels in semi-supervised learning, and expert demonstration policies in reinforcement learning. However, guidance signals can be noisy due to domain shifts and limited data availability and may not generalize well. Blindly trusting such signals when they are noisy, incomplete, or misaligned with the target domain can lead to degraded performance. To address these challenges, we propose Adaptive Conformal Guidance (AdaConG), a simple yet effective approach that dynamically modulates the influence of guidance signals based on their associated uncertainty, quantified via split conformal prediction (CP). By adaptively adjusting to guidance uncertainty, AdaConG enables models to reduce reliance on potentially misleading signals and enhance learning performance. We validate AdaConG across diverse tasks, including knowledge distillation, semi-supervised image classification, gridworld navigation, and autonomous driving. Experimental results demonstrate that AdaConG improves performance and robustness under imperfect guidance, e.g., in gridworld navigation, it accelerates convergence and achieves over $\times 6$ higher rewards than the best-performing baseline. These results highlight AdaConG as a broadly applicable solution for learning under uncertainty.

Large Language Models (LLMs) have achieved remarkable success in software engineering tasks when trained with executable runtime environments, particularly in resolving GitHub issues. However, such runtime environments are often unavailable in other domains, especially cybersecurity, where challenge configurations and execution contexts are ephemeral or restricted. We present Cyber-Zero, the first runtime-free framework for synthesizing high-quality agent trajectories to train cybersecurity LLMs. Cyber-Zero leverages publicly available CTF writeups and employs persona-driven LLM simulation to reverse-engineer runtime behaviors and generate realistic, long-horizon interaction sequences without actual environments. Using trajectories synthesized by Cyber-Zero, we train LLM-based agents that achieve up to 13.1% absolute performance gains over baseline models on three prominent CTF benchmarks: InterCode-CTF, NYU CTF Bench, and Cybench. Our best model, Cyber-Zero-32B, establishes new state-of-the-art performance among open-weight models, matching the capabilities of proprietary systems like DeepSeek-V3-0324 and Claude-3.5-Sonnet while offering superior cost-effectiveness, and demonstrating that runtime-free trajectory synthesis can effectively democratize the development of state-of-the-art cybersecurity agents.

Generative models can generate photorealistic images at scale. This raises urgent concerns about the ability to detect synthetically generated images and attribute these images to specific sources. While watermarking has emerged as a possible solution, existing methods remain fragile to realistic distortions, susceptible to adaptive removal, and expensive to update when the underlying watermarking key changes. We propose a general watermarking framework that formulates the encoding problem as key-dependent perturbation of the parameters of a generative model. Within this framework, we introduce Mixture of LoRA Markers (MOLM), a routing-based instantiation in which binary keys activate lightweight LoRA adapters inside residual and attention blocks. This design avoids key-specific re-training and achieves the desired properties such as imperceptibility, fidelity, verifiability, and robustness. Experiments on Stable Diffusion and FLUX show that MOLM preserves image quality while achieving robust key recovery against distortions, compression and regeneration, averaging attacks, and black-box adversarial attacks on the extractor.

Humans rely on memory to perform tasks; our goal is to endow robot policies with the same ability. Naively conditioning on long observation histories is computationally expensive and brittle under covariate shift, while indiscriminate subsampling of history leads to irrelevant or redundant information. We propose a hierarchical policy framework, where the high-level policy is trained to select and track previous task-relevant keyframes from its experience. The high-level policy uses selected keyframes and the most recent frames when generating text instructions for a low-level policy to execute. This design is compatible with existing vision-language-action (VLA) models and enables the system to efficiently reason over long-horizon dependencies. In our experiments, we fine-tune Qwen2.5-VL-7B-Instruct and $\pi_{0.5}$ as the high-level and low-level policies respectively, using demonstrations supplemented with minimal language annotations. Our approach, MemER, outperforms prior methods on three real-world long-horizon robotic manipulation tasks that require minutes of memory. Videos and code can be found at https://memer-policy.github.io.

Large language models face challenges in long-context question answering, where key evidence of a query may be dispersed across millions of tokens. Existing works equip large language models with a memory corpus that is dynamically updated during a single-pass document scan, also known as the "memorize while reading" methods. While this approach scales efficiently, it suffers from irreversible forward-only processing, information loss through overwriting, and sparse reinforcement learning signals. To tackle these challenges, we present ReMemR1, a memory-augmented agent with callback-enhanced memory that allows selective retrieval from the entire memory history and allows non-linear reasoning and revisiting of early evidence. To further strengthen training, we propose Reinforcement Learning with Multi-Level Rewards (RLMLR), which combines final-answer rewards with dense, step-level signals that guide effective memory use. Together, these contributions mitigate information degradation, improve supervision, and support multi-hop memory utilizing. Experiments on long-document QA show significant gains over existing memory-based approaches, which validates ReMemR1 as an effective solution for long-context reasoning agents.

We study the problem of zero-shot link prediction on knowledge graphs (KGs), which requires models to generalize to novel entities and novel relations. Knowledge graph foundation models (KGFMs) address this task by enforcing equivariance over both nodes and relations, which enables them to learn structural properties of nodes and relations that transfer to novel KGs with similar structure. However, the conventional notion of deterministic equivariance inherently limits the expressive power of KGFMs, as it prevents them from distinguishing relations that are structurally similar but semantically distinct. To overcome this limitation, we propose to leverage probabilistic node-relation equivariance, which preserves equivariance in distribution while using structured randomness to break symmetries at inference time. Building on this principle, we present Flock, a KGFM that iteratively samples random walks, encodes them into sequences, embeds them with a sequence model, and aggregates node and relation representations through learned pooling. Flock respects probabilistic node-relation equivariance and, crucially, is a universal approximator for isomorphism-invariant link-level functions over KGs. Empirically, Flock perfectly solves our new diagnostic dataset Petals on which current KGFMs fail, and achieves state-of-the-art performance on entity and relation prediction tasks across 54 KGs from diverse domains. Code is available at https://github.com/jw9730/flock.

Recent empirical studies have explored the idea of continuing to train a model at test-time for a given task, known as test-time training (TTT), and have found it to yield significant performance improvements. However, there is limited understanding of why and when TTT is effective. Earlier explanations mostly focused on the observation that TTT may help when applied to out-of-distribution adaptation or used with privileged data. However, the growing scale of foundation models with most test data being in-distribution questions these explanations. We instead posit that foundation models remain globally underparameterized, with TTT providing a mechanism for *specialization after generalization*—focusing capacity on concepts relevant to the test task. Specifically, under the linear representation hypothesis, we propose a model in which TTT achieves a substantially smaller *in-distribution* test error than global training. We empirically validate our model's key assumptions by training a sparse autoencoder on ImageNet, showing that semantically related data points are explained by only a few shared concepts. Finally, we perform scaling studies across image and language tasks that confirm the practical implications of our model, identifying the regimes where specialization is most effective. Beyond TTT, our results provide additional strong evidence in support of the linear representation hypothesis.

Uncovering the opacity of diffusion-based generative models is urgently needed, as their applications continue to expand while their underlying procedures largely remain a black box. With a critical question -- how can the diffusion generation process be interpreted and understood? -- we proposed \textit{Patronus}, an interpretable diffusion model that incorporates a prototypical network to encode semantics in visual patches, revealing \textit{what} visual patterns are learned and \textit{where} and \textit{when} they emerge throughout denoising. This interpretability of Patronus provides deeper insights into the generative mechanism, enabling the detection of shortcut learning via unwanted correlations and the tracing of semantic emergence across timesteps. We evaluate \textit{Patronus} on four natural image datasets and one medical imaging dataset, demonstrating both faithful interpretability and strong generative performance. With this work, we open new avenues for understanding and steering diffusion models through prototype-based interpretability.

Diffusion-based samplers learn to sample complex, high-dimensional distributions using energies or log densities alone, without training data. Yet, they remain impractical for molecular sampling because they are often slower than molecular dynamics and miss thermodynamically relevant modes. Inspired by enhanced sampling, we encourage exploration by introducing a sequential bias along bespoke, information-rich, low-dimensional projections of atomic coordinates known as collective variables (CVs). We introduce a repulsive potential centered on the CVs from recent samples, which pushes future samples towards novel CV regions and effectively increases the temperature in the projected space. Our resulting method improves efficiency, mode discovery, enables the estimation of free energy differences, and retains independent sampling from the approximate Boltzmann distribution via reweighting by the bias. On standard peptide conformational sampling benchmarks, the method recovers diverse conformational states and accurate free energy profiles. We are the first to demonstrate reactive sampling using a diffusion-based sampler, capturing bond breaking and formation with universal interatomic potentials at near-first-principles accuracy. The approach resolves reactive energy landscapes at a fraction of the wall-clock time of standard sampling methods, advancing diffusion-based sampling towards practical use in molecular sciences.

Small influential data subsets can dramatically impact model conclusions, with a few data points overturning key findings. While recent work identifies these most influential sets, there is no formal way to tell when maximum influence is excessive rather than expected under natural random sampling variation. We address this gap by developing a principled framework for most influential sets. Focusing on linear least-squares, we derive a convenient exact influence formula and identify the extreme value distributions of maximal influence – the heavy-tailed Fréchet for constant‑size sets and heavy tailed data, and the well-behaved Gumbel for growing sets or light tails. This allows us to conduct rigorous hypothesis tests for excessive influence. We demonstrate through applications across economics, biology, and machine learning benchmarks, resolving contested findings and replacing ad‑hoc heuristics with rigorous inference.