ICLR 2026 Papers

Synthetic data generation for tabular datasets is shifting toward the use of large, general-purpose foundation models. TabPFN, a state-of-the-art example, uses in-context learning to generate probabilistic predictions conditioned on observed examples in a single forward pass. However, when variables are only weakly associated with others, the model's ability to generate realistic synthetic data deteriorates, as the context examples provide little predictive signal. To address this, we introduce the maximal information auxiliary variable (MIAV) strategy, which increases context information with auxiliary variables constructed by rank-matching random noise variables to real data. We establish theoretical properties of the approach which explain its good performance for weakly associated variables. Additional practical advantages of the MIAV approach include improved computational efficiency and invariance to variable order during the synthetic data generation process. Empirical evaluations, on simulated and real datasets, illustrate how the MIAV strategy improves data generation when compared to direct application of TabPFN, and is competitive against other baselines. To illustrate the generality of the MIAV approach we also present an implementation based on the TabICL model (a more scalable tabular foundation model restricted to classification tasks) for performing synthetic data generation on categorical datasets. Overall, MIAV offers an effective foundation model–based alternative to bespoke synthetic data generators.

Scaling foundation model training with Distributed Data Parallel~(DDP) methods is bandwidth-limited. Existing infrequent communication methods like Local SGD were designed to synchronize model parameters only and cannot be trivially applied to adaptive optimizers due to additional optimizer states. Heuristic approaches that keep states local or reset them lack guarantees and can be unstable in compute‑efficient batch regimes; conversely, Local Adam synchronizes all states uniformly and is provably convergent but triples communication costs. We propose Desynced Low Communication Adaptive Optimizers (DES-LOC), a family of optimizers assigning independent synchronization periods to parameters and momenta, enabling lower communication costs while preserving convergence. Our theoretical analysis shows that while parameter synchronization dominates the asymptotic rate in-expectation, high-probability convergence guarantees require at least infrequent synchronization of the second momentum. Furthermore, we prove that more frequent momentum sync permits larger stable step sizes. Experiments on language models of up to 1.7B show that DES-LOC can communicate 170x less than DDP and 2x less than the previous state-of-the-art Local Adam, enabling 1.3x–2.1x wall‑clock speedups over DDP for 1-13B models on 100Gb/s links. Furthermore, unlike previous heuristic methods, DES-LOC is robust to worker failures offering a scalable, efficient, and fault-tolerant solution for foundation model training.

Low-discrepancy point sets and digital sequences underpin quasi-Monte Carlo (QMC) methods for high-dimensional integration. We cast two long-standing QMC design problems as program synthesis and solve them with an LLM-guided evolutionary loop that mutates and selects code under task-specific fitness: (i) constructing finite 2D/3D point sets with low star discrepancy, and (ii) choosing Sobol’ direction numbers that minimize randomized quasi-Monte Carlo (rQMC) error on downstream integrands. Our two-phase procedure combines constructive code proposals with iterative numerical refinement. On finite sets, we rediscover known optima in small 2D cases and set new best-known 2D benchmarks for N ≥ 40, while matching known 3D optima up to the proven frontier (N ≤ 8) and reporting improved 3D benchmarks beyond. On digital sequences, evolving Sobol' parameters yields consistent reductions in rQMC mean-squared error for several 32-dimensional option-pricing tasks relative to widely used Joe–Kuo parameters, while preserving extensibility to any sample size and compatibility with standard randomizations. Taken together, the results demonstrate that LLM-driven evolutionary program synthesis can automate the discovery of high-quality QMC constructions, recovering classical designs where they are optimal and improving them where finite-N structure matters. Data and code are available at anonymous.

Self-improvement has emerged as a state-of-the-art paradigm in Neural Combinatorial Optimization (NCO), where models iteratively refine their policies by generating and imitating high-quality solutions. Despite strong empirical performance, existing methods face key limitations. Training is computationally expensive, as policy updates require sampling numerous candidate solutions per instance to extract a single expert trajectory. More fundamentally, these approaches fail to exploit the structure of combinatorial problems involving the coordination of multiple agents, such as vehicles in min-max routing or machines in scheduling. By supervising on single-action trajectories, they fail to exploit agent-permutation symmetries, where distinct sequences of actions yield identical solutions, hindering generalization and the ability to learn coordinated behavior. We address these challenges by extending self-improvement to operate over joint multi-agent actions. Our model architecture predicts complete agent-task assignments jointly at each decision step. To explicitly leverage symmetries, we employ a set-prediction loss, which supervises the policy on multiple expert assignments for any given state. This approach enhances sample efficiency and the model's ability to learn coordinated behavior. Furthermore, by generating multi-agent actions in parallel, it drastically accelerates the solution generation phase of the self-improvement loop. Empirically, we validate our method on several combinatorial problems, demonstrating consistent improvements in the quality of the final solution and a reduced generation latency compared to standard self-improvement.

Video matting is a crucial task for many applications, but existing methods face significant limitations. They are often domain-specific, focusing primarily on human portraits, and rely on the mask of first frame that is challenging to acquire for transparent or intricate objects like fire or smoke. To address these challenges, we introduce Matting Anything 2 (MAM2), a versatile and robust video matting model that handles diverse objects using flexible user prompts such as points, boxes, or masks. We first propose Promptable Dual-mode Decoder (PDD), an effective structure that simultaneously predicts a segmentation mask and a corresponding high-quality trimap, leveraging trimap-based guidance to improve generalization. To tackle prediction instability for transparent objects across video frames, we further propose a Memory-Separable Siamese (MSS) mechanism. MSS employs a recurrent approach that isolates trimap prediction from potentially interfering mask memory, significantly enhancing temporal consistency. To validate our method's performance on diverse objects, we introduce the Natural Video Matting dataset, a new benchmark with substantially greater diversity. Extensive experiments show that MAM2 possesses exceptional matting accuracy and generalization capabilities. We believe MAM2 demonstrates a significant leap forward in creating a video matting method for anything.

Accurately predicting experimentally-realizable $3\textrm{D}$ molecular crystal structures from their $2\textrm{D}$ chemical graphs is a long-standing open challenge in computational chemistry called $\textit{crystal structure prediction}$ (CSP). Efficiently solving this problem has implications ranging from pharmaceuticals to organic semiconductors, as crystal packing directly governs the physical and chemical properties of organic solids. In this paper, we introduce $\textrm{OXtal}$, a large-scale $100\textrm{M}$ parameter all-atom diffusion model that directly learns the conditional joint distribution over intramolecular conformations and periodic packing. To efficiently scale $\textrm{OXtal}$, we abandon explicit equivariant architectures imposing inductive bias arising from crystal symmetries in favor of data augmentation strategies. We further propose a novel crystallization-inspired lattice-free training scheme, $\textit{Stoichiometric Stochastic Shell Sampling}$ ($S^4$), that efficiently captures long-range interactions while sidestepping explicit lattice parametrization---thus enabling more scalable architectural choices at all-atom resolution. Trained on $600 \text{K}$ experimentally validated crystal structures (including rigid and flexible molecules, co-crystals, and solvates), $\textrm{OXtal}$ achieves orders-of-magnitude improvements over prior $\textit{ab-initio}$ ML CSP methods, which remaining orders of magnitude cheaper than traditional quantum-chemical approaches. Specifically, $\textrm{OXtal}$ reproduces experimental structures with conformer $\mathrm{RMSD}_1<0.5$ Å and attains over 80\% lattice-match success, demonstrating its ability to model both thermodynamic and kinetic regularities of molecular crystallization.

The sliced Wasserstein distance (SW) reduces optimal transport on $\mathbb{R}^d$ to a sum of one-dimensional projections, and thanks to this efficiency, it is widely used in geometry, generative modeling, and registration tasks. Recent work shows that quasi-Monte Carlo constructions for computing SW (QSW) yield direction sets with excellent approximation error. This paper presents an alternate, novel approach: learning directions with Bayesian optimization (BO), particularly in settings where SW appears inside an optimization loop (e.g., gradient flows). We introduce a family of drop-in selectors for projection directions: **BOSW**, a one-shot BO scheme on the unit sphere; **RBOSW**, a periodic-refresh variant; **ABOSW**, an adaptive hybrid that seeds from competitive QSW sets and performs a few lightweight BO refinements; and **ARBOSW**, a restarted hybrid that periodically relearns directions during optimization. Our BO approaches can be composed with QSW and its variants (demonstrated by ABOSW/ARBOSW) and require no changes to downstream losses or gradients. We provide numerical experiments where our methods achieve state-of-the-art performance, and on the experimental suite of the original QSW paper, we find that ABOSW and ARBOSW can achieve convergence comparable to the best QSW variants with modest runtime overhead. We release code with fixed seeds and configurations to support faithful replication (see supplementary material).

Automatic video polyp segmentation (VPS) is crucial for preventing and treating colorectal cancer by ensuring accurate identification of polyps in colonoscopy examinations. However, its clinical application is hampered by two key challenges: shape collapse, which compromises structural integrity, and episodic amnesia, which causes instability in challenging video sequences. To address these challenges, we present a novel video segmentation network, \emph{HFSTI-Net}, which integrates global perception with spatiotemporal consistency in spatial, temporal, and frequency domains. Specifically, to address shape collapse under low contrast or visual ambiguity, we design a Hierarchical Frequency-spatial Interaction (HFSI) module that fuses spatial and frequency cues for fine-grained boundary localization. Furthermore, we propose a recurrent mask-guided propagation (RMP) module that introduces a dual enhancement mechanism based on feature memory and mask alignment, effectively incorporating spatiotemporal information to alleviate inter-frame inconsistencies and ensuring long-term segmentation stability. Extensive experiments on the SUN-SEG and CVC-612 datasets demonstrate that our method achieves real-time inference and outperforms other state-of-the-art approaches. The codes will be made available upon publication.

Automated interpretability aims to translate large language model (LLM) features into human understandable descriptions. However, natural language feature descriptions are often vague, inconsistent, and require manual relabeling. In response, we introduce *semantic regexes*, structured language descriptions of LLM features. By combining primitives that capture linguistic and semantic patterns with modifiers for contextualization, composition, and quantification, semantic regexes produce precise and expressive feature descriptions. Across quantitative benchmarks and qualitative analyses, semantic regexes match the accuracy of natural language while yielding more concise and consistent feature descriptions. Their inherent structure affords new types of analyses, including quantifying feature complexity across layers, scaling automated interpretability from insights into individual features to model-wide patterns. Finally, in user studies, we find that semantic regexes help people build accurate mental models of LLM features.

Recent advances in video generation have produced powerful diffusion models capable of generating high-quality, temporally coherent videos. We ask whether space-time tracking capabilities emerge automatically within these generators, as a consequence of the close connection between synthesizing and estimating motion. We propose a simple but effective way to elicit point tracking capabilities in off-the-shelf image-conditioned video diffusion models. We simply place a colored marker in the first frame, then guide the model to propagate the marker across frames, following the underlying video’s motion. To ensure the marker remains visible despite the model’s natural priors, we use the unedited video's initial frame as a negative prompt. We evaluate our method on the TAP-Vid benchmark using several video diffusion models. We find that it outperforms prior zero-shot methods, often obtaining performance that is competitive with specialized self-supervised models, despite the fact that it does not require any additional training.

Diffusion models have transformed image generation, yet controlling their outputs for diverse applications, including content moderation and creative customization, remains challenging. Existing approaches usually require task-specific training and struggle to generalise across both concrete (e.g., objects) and abstract (e.g.,4 styles) concepts. We propose CASteer (Cross-Attention Steering), a training-free framework for controllable image generation using steering vectors to influence a diffusion model’s hidden representations dynamically. CASteer precomputes concept-specific steering vectors by averaging neural activations from images generated for each target concept. During inference, it dynamically applies these vectors to modify outputs only when necessary, either removing undesired concepts from images where they appear or adding desired concepts to images where they are absent. This selective activation ensures precise, context-aware adjustments without altering unaffected regions. This approach enables precise control over a wide range of tasks, including removing harmful content, interpolating between desired attributes, replacing objects, all without model retraining. CASteer outperforms state-of-the-art techniques while preserving unrelated content and minimising unintended effects. Code is provided in the supplementary

The reasoning large language model (RLLM) has been proven competitive in solving complex reasoning tasks such as mathematics, coding, compared to traditional LLM. However, the serving performance and behavior of RLLM remains \textit{unexplored}, which may undermine the deployment and utilization of RLLM in real-world scenario. To close this gap, in this paper, we conduct a comprehensive study of RLLM service. We first perform a pilot study on comparing the serving performance between RLLM and traditional LLM and reveal that there are several distinct differences regarding serving behavior: (1) \textit{significant memory usage and fluctuations}; (2) \textit{straggler requests}; (3) \textit{adaptive running time}; (4) \textit{domain preference}. Then we further investigate whether existing inference optimization techniques are valid for RLLM. Our main takeaways are that model quantization methods and speculative decoding can improve service system efficiency with small compromise to RLLM accuracy, while prefix caching, KV cache quantization may even degrade accuracy or serving performance for small RLLM. Lastly, we conduct evaluation under real world workload modeled by Gamma distribution to verify our findings. Empirical results for real world workload evaluation across different dataset are \textit{aligned} with our main findings regarding RLLM serving. We hope our work can provide the research community and industry with insights to advance RLLM inference serving.

We introduce Velocity-Regularized Adam (VRAdam), a physics-inspired optimizer for training deep neural networks that draws on ideas from quartic terms for kinetic energy with its stabilizing effects on various system dynamics. Previous algorithms, including the ubiquitous Adam, operate at the so-called adaptive edge of stability regime during training, leading to rapid oscillations and slowed convergence of loss. However, VRAdam adds a higher order penalty on the learning rate based on the velocity such that the algorithm automatically slows down whenever weight updates become large. In practice, we observe that the effective dynamic learning rate shrinks in high-velocity regimes, and damping oscillations. By combining this velocity‑based regularizer for global damping with Adam’s per‑parameter scaling, we create a powerful hybrid optimizer. For this optimizer, we provide rigorous theoretical analysis of operation at the edge of stability from a physical and control perspective for the momentum. Furthermore, we derive convergence bounds with the rate $\mathcal{O}(\ln(N)/\sqrt{N})$ for a stochastic non‑convex objective under mild assumptions. We demonstrate that VRAdam exceeds the performance against standard optimizers including AdamW. We benchmark various tasks such as image classification, language modeling, and generative modeling using diverse architectures and training methodologies including Convolutional Neural Networks (CNNs), Transformers, and GFlowNets.

Finetuning (pretrained) language models is a standard approach for updating their internal parametric knowledge and specializing them to new tasks and domains. However, the corresponding model weight changes ("weight diffs") are not generally interpretable. While inspecting the finetuning dataset can give a sense of how the model might have changed, these datasets are often not publicly available or are too large to work with directly. Towards the goal of broadly understanding model weight changes in natural language, we introduce Diff Interpretation Tuning (DIT), a method that trains models to describe their own finetuning-induced modifications. Our approach uses synthetic, labeled weight diffs to train an introspection adapter, which can be applied to a compatible finetuned model to make it self-describe the weight changes. We demonstrate in two proof-of-concept settings (reporting hidden behaviors and summarizing finetuned knowledge) that our method enables models to describe their finetuning-induced modifications using concise and accurate natural language descriptions.

We study the problem of learning to generate an answer (or completion) to a question (or prompt), where there could be multiple correct answers, any one of which is acceptable at test time. Learning is based on demonstrations of some correct answer to each training question, as in Supervised Fine Tuning (SFT). We formalize the problem as offline imitation learning in contextual bandits, with demonstrations from some optimal policy, without explicitly observed rewards. Prior work assumes that the demonstrator belongs to a low-complexity policy class, which motivates maximum likelihood estimation (i.e., log-loss minimization). In contrast, we propose relying only on the reward model (specifying which answers are correct) being in a low-cardinality class, which we argue is a weaker assumption. We show that likelihood maximization methods can fail in this case, and instead suggest an alternative novel approach that learns with sample complexity logarithmic in the cardinality of the reward class. Our work motivates looking beyond likelihood maximization when learning from demonstrations.

Diffusion LLMs have attracted growing interest, with plenty of recent work emphasizing their great potential in various downstream tasks; yet the long‑context behavior of diffusion LLMs remains largely uncharted. We present a case study of post‑training techniques for extending the context window of diffusion LLMs (i.e., LLaDA) without retraining from scratch. We show that a simple modification to the standard Rotary Positional Embeddings (RoPE) extension effectively accommodates the probabilistic modeling inherent in the diffusion process, enabling stable scaling to longer context ranges. We further compare masking strategies used during post‑training and analyze their impact on optimization stability and long‑range recall. Instantiating these insights, we introduce UltraLLaDA, a diffusion LLM with a 128K‑token context window that, in our empirical evaluation on long‑context tasks, significantly outperforms training‑free baselines. Our experimental results highlight the special positional extension as a key lever for scaling diffusion LLMs to extended contexts and offer practical guidance for practitioners seeking 128K‑scale context via efficient post‑training.

Electroencephalography (EEG) is a noninvasive technique for measuring brain electrical activity that supports a wide range of brain-computer interaction applications. Motivated by the breakthroughs of Large Language Models (LLMs), recent efforts have begun to explore Large EEG foundation Models trained on broad unlabeled corpora. However, most advances focus on improving the backbone while neglecting the classification head. Existing models often rely on a single class token, underutilizing the spatiotemporal structure and second-order statistics that are crucial for EEG decoding. We propose Riemannian High Order Pooling (RHOP), a plug-and-play module that injects principled Riemannian statistics into the classifier. RHOP maps each token to a quotient Gaussian jointly encoding mean and second-order information, yielding scale-invariant descriptors. Tokens are then aggregated by estimating a Riemannian Gaussian on the SPD manifold, where the Fréchet mean and covariance are embedded into an SPD descriptor. The resulting normalized vector is fused with the class token for prediction. RHOP is backbone-agnostic and integrates with modern EEG foundation models, e.g., BIOT and LaBraM. Across diverse EEG benchmarks, it improves accuracy, robustness, and efficiency under full fine-tuning, linear probing, and from-scratch training settings.

Diffusion models, well recognized for their strong generative priors, have recently been increasingly applied to super-resolution (SR) tasks. However, as diffusion models are trained on Gaussian-corrupted natural images, the distribution gap between low-resolution (LR) inputs and the model’s training distribution hinders direct inference. Prior works address this by conditioning on LR images, but their fine-tuning often weakens generative capability and requires multiple denoising steps. In this work, we present DM-SR, a novel framework that bridges this gap without modifying the pretrained diffusion model. We train an image encoder that maps LR inputs into the latent distribution aligned with the diffusion model’s training space, preserving its full generative power. Furthermore, DM-SR adaptively predicts the appropriate noise level based on the degradation of each input, ensuring optimal alignment with the diffusion model’s timestep-dependent distribution. Extensive experiments show that DM-SR achieves superior perceptual quality with a single-stage diffusion process, setting a new direction for efficient and high-fidelity SR with diffusion models.

Reflecting the greater significance of recent history over the distant past in non-stationary environments, $\lambda$-discounted regret has been introduced in online convex optimization (OCO) to gracefully forget past data as new information arrives. When the discount factor $\lambda$ is given, online gradient descent with an appropriate step size achieves an $O(1/\sqrt{1-\lambda})$ discounted regret. However, the value of $\lambda$ is often not predetermined in real-world scenarios. This gives rise to a significant \emph{open question}: is it possible to develop a discounted algorithm that adapts to an unknown discount factor. In this paper, we affirmatively answer this question by providing a novel analysis to demonstrate that smoothed OGD (SOGD) achieves a uniform $O(\sqrt{\log T/1-\lambda})$ discounted regret, holding for all values of $\lambda$ across a continuous interval simultaneously. The basic idea is to maintain multiple OGD instances to handle different discount factors, and aggregate their outputs sequentially by an online prediction algorithm named as Discounted-Normal-Predictor (DNP). Our analysis reveals that DNP can combine the decisions of two experts, even when they operate on discounted regret with different discount factors.