ICML 2026 Papers

Mendelian Randomization (MR) is a prominent observational epidemiological research method, designed to address unobserved confounding when estimating causal effects. It is closely related to instrumental variable (IV) methods, where genetic variants serve as instruments to infer causal relationships from observational data. However, the core assumptions required for valid IV analysis---particularly the independence between instruments and unobserved confounders---are untestable and often violated in practice. In MR, such violations commonly arise when genetic variants are correlated with environmental factors (e.g., population stratification and assortive mating), leading to confounding between instruments and outcomes. At the same time, MR studies increasingly include data collected across multiple environments or populations, providing an opportunity to address these violations. Leveraging this setting, we propose a representation learning framework that exploits multi-environment data to recover latent exogenous components of genetic instruments suitable for causal inference. We provide theoretical insights into when and how the learned components can act as valid instruments, and we demonstrate the effectiveness of our approach through simulations and semi-synthetic experiments using genetic data from the All of Us Biobank.

Recent advances in Omni-Multimodal Large Language Models (Omni-MLLMs) have enabled strong integration of vision, audio, and language. However, their audio-visual intelligence (AVI) remains insufficiently evaluated due to the lack of systematic and comprehensive benchmarks. We introduce **AVI-Bench**, a cognitively inspired benchmark that evaluates Omni-MLLMs across three stages, perception, understanding, and reasoning, through cross-modal tasks requiring joint audio-visual interpretation. This design enables fine-grained diagnosis of model capabilities and failure modes. To further assess robustness beyond familiar domains, we propose **AVI-Bench-PriSe**, an extension that probes models' primitive audio-visual sensation using unfamiliar, low-semantic stimuli, testing generalization beyond common training distributions. Extensive experiments on both open-source and closed-source models reveal substantial limitations in current Omni-MLLMs. Based on these findings, we present a **four-level AVI taxonomy**. Overall, AVI-Bench provides a principled evaluation framework to guide the development of more robust and generalizable AVI. Project website: https://fudancvl.github.io/AVI-Bench

Optimization problems are fundamental across science and industry, including planning, scheduling, and resource allocation. While LLMs show promise in automating optimization, they struggle to bridge the gap between real-world requirements and both mathematical formulations and effective heuristic designs. Furthermore, the field lacks a unified framework that spans problem formulation and heuristic discovery for NP-hard settings. To address these challenges, we propose OptMaster, a unified framework that spans optimization from formulation to heuristic discovery, structuring the process as a Directed Acyclic Graph (DAG) where each node represents a candidate solution. The DAG architecture enables cross-branch knowledge transfer when search progress stagnates. Within each node, we further replace textual self-reflection with independently generated verification code, grounding the evaluation in deterministic computation to suppress hallucinations. OptMaster achieves competitive performance across two optimization paradigms. In Formulation Intelligence, OptMaster achieves state-of-the-art accuracy across the three most challenging benchmarks in the field. In Heuristic Discovery, OptMaster surpasses the best known solutions on Circle Packing ($n=26, 32$) and achieves a cut of 9,590 on Gset70 with significantly reduced time and search budgets.

Geometric analysis fundamentally distinguishes between extrinsic and intrinsic perspectives. The dominant paradigm in current 3D representation learning relies on either extrinsic spatial structures or high-level semantics, struggling to capture the essence of shape identity and underlying manifold topology. To bridge this gap, we introduce a novel 3D representation learning paradigm, namely PRISM, for Pre-training, which learns isometric embeddings by Recovering the Intrinsic Surface geodesic Metric. PRISM incorporates a topology-enforcing objective that explicitly constrains the structure of latent space, alongside a specialized two-stage training recipe mitigating sample imbalance inherent in the distribution of geodesic distances. Experiments demonstrate that our approach shows satisfactory accuracy, robustness, and high efficiency in geodesic distance prediction and achieves superior performance across diverse downstream tasks, including shape recognition, surface parameterization, and non-rigid correspondence. Our code will be made publicly available.

Generalized Category Discovery (GCD) requires models to categorize an unlabeled pool containing known and novel classes under sparse supervision. We identify a systemic confidence bias inherent in existing parametric methods: while entropy regularization prevents class collapse, it indiscriminately suppresses predictive certainty on all unlabeled instances. This bias drives a distributional wedge between labeled and unlabeled samples of the same category, forcing models to sacrifice their stability on known classes to achieve plasticity for new ones. To resolve this, we propose Reliable Confidence Alignment (RCA), a plug-and-play framework grounded in Evidential Deep Learning. RCA first establishes high certainty anchors on labeled data using a Reliable Anchor for Certainty (RAC) module. Then, we introduce Cross-view Confidence Alignment (CCA) to propagate this grounded reliability to the unlabeled discovery set. Thus, RCA captures the fine-grained geometry of the probability simplex, effectively calibrating the model's epistemic uncertainty. Extensive evaluations on coarse- and fine-grained benchmarks demonstrate that RCA effectively rectifies the confidence landscape, significantly mitigating performance decay on known classes without compromising novel-class discovery.

Traditional Mixture-of-Experts (MoE) training typically proceeds without any structural priors, effectively requiring the model to simultaneously train expert weights while searching for an optimal routing policy within a vast combinatorial space. This entanglement often leads to sluggish convergence and training instabilities. This paper introduces Grouter a preemptive routing method that by distilling high-quality structures from fully-trained MoE models and serving as a fixed router for target models. By decoupling structural optimization from weight updates, Grouter significantly accelerates both the speed and quality of model convergence. To ensure the framework's versatility, we also introduce expert folding to adapt Grouter across varying model configurations and expert tuning to rebalance workloads across different data distributions. Furthermore, by leveraging the structural priors provided by preemptive routing, we can implement targeted optimizations to further enhance training throughput. Experiments demonstrate that Grouter achieves superior performance and efficiency which boosts pre-training data utilization by $\textbf{4.28$\times$}$ and achieves up to $\textbf{33.5$\%$}$ throughput acceleration, establishing preemptive routing as a fundamental paradigm for scalable MoE training. We publicly release our code and pretrained Grouter checkpoints.

Synthetic data generation is increasingly used in machine learning for **training and data augmentation**. Yet, many current strategies rely on external foundation models or datasets, which can be restricted by policy or legal constraints, especially for sensitive modalities such as human face images and videos. We propose **ScoreMix**, a **self-contained data augmentation** method to boost recognition performance by leveraging score compositionality in class-conditioned diffusion models. ScoreMix mixes class-conditioned scores along reverse diffusion trajectories, yielding domain-specific hard augmentations without external resources. We systematically study class-selection strategies and find that mixing classes that are distant in the discriminator embedding space yields larger gains, providing **up to 3\% additional average improvement across benchmarks** over proximity-based selection. Interestingly, we observe that learned condition and embedding spaces are largely uncorrelated under standard alignment metrics, and that condition-space distances are weakly correlated to downstream gains. Across **8 public face recognition benchmarks**, ScoreMix improves accuracy by **up to 7 percentage points** without hyperparameter search, highlighting robustness and practicality. Project page: https://parsa-ra.github.io/scoremix/ .

We analyze the hardmax limit of self-attention dynamics for token embeddings in the zero-temperature regime $(\beta \to +\infty)$ and relate it to finite-$\beta$ behavior. In this limit, the update rule can be viewed as a Frank-Wolfe step for a quadratic objective over the convex hull of the current tokens. When the key-query matrix is negative semidefinite, the dynamics converge with the standard sublinear rate $\mathcal{O}(t^{-1})$ on the quadratic energy, whereas in the positive semidefinite case, extending the hardmax rule to the convex hull induces a Voronoi structure: vertices are stationary, interior points remain in their initial cells, and each token moves along a straight line toward its cell's vertex with exponential convergence under a step-size bounded away from zero. We additionally establish well-posedness of the associated ODE limit in this regime. For finite $\beta$, we model self-attention as a Markov chain and prove *dynamic metastability*: interior tokens reach near-vertex configurations in a constant number of steps and remain trapped for times exponential in $\beta$ with high probability, before eventual collapse to some point within the initial convex hull. Thus, hardmax dynamics accurately approximate the finite-$\beta$ process over exponentially long time horizons.

The impressive performance of generalist large language models (LLMs) such as GPT and Claude in healthcare raises a critical question: will domain-specific medical specialist models become obsolete? We argue that the future of medical artificial intelligence (AI) lies not in building monolithic medical foundation models, nor in replacing human expertise, but in orchestrating collaboration among generalist LLMs, domain-specific specialist models, and clinicians. We propose HetMedAgent, a heterogeneous medical multi-agent framework that enables conflict-aware evidence fusion, uncertainty-based clinician intervention triggering, and adaptive threshold calibration. Experiments on three real-world clinical decision-making tasks demonstrate that the synergy between generalist LLMs and domain-specific specialist models significantly outperforms using either type of model alone, validating the irreplaceable value of specialist models in modality-specific analysis. HetMedAgent represents a shift from building medical LLMs or foundation models to multi-agent collaboration, achieving a balance between general reasoning capabilities and domain-specific precision.

In robotic imitation learning, accurately modeling the multimodality and temporal correlations of long-horizon action sequences remains challenging. Long-horizon tasks require preserving global task intent while executing precise low-level control; otherwise, local errors can accumulate and lead to failure. While recent coarse-to-fine autoregressive models have improved action generation, they struggle to maintain consistency across hierarchies, leading to suboptimal performance in long-horizon tasks. To address these shortcomings, we propose Probabilistically Consistent Multi-Scale Action Generation (MSP), a novel coarse-to-fine approach that promotes cross-scale consistency. MSP adopts a streamlined multi-scale design by directly downsampling in a continuous latent space. A scale-wise autoregressive Transformer is used to generate semantic conditions at each scale, which guide a lightweight MeanFlow model to capture multi-scale latent distributions, enabling probabilistically consistent refinement across scales. Through extensive simulation and real-world experiments, including long-horizon, multi-task, and few-shot generalization settings, we show that MSP outperforms existing coarse-to-fine methods, achieving state-of-the-art performance with high efficiency.

We present Reading Observed At Mindless Moments (ROAMM), a multimodal dataset comprising 50 hours of simultaneous EEG and eye-tracking recordings collected during naturalistic multi-page reading from 44 participants. ROAMM includes synchronized physiological recordings, eye-movement events, page-level comprehension scores, and span-level mind-wandering (MW) annotations obtained using a retrospective self-report paradigm. We introduce a standardized leave-one-subject-out benchmark for MW detection and achieve up to 0.609 AUROC using supervised models. We additionally evaluate EEG-to-text decoding on reading segments with and without MW labels, showing that decoding performance decreases during MW episodes. ROAMM enables research on MW detection, EEG-to-text decoding, multimodal representation learning, and attention-related degradation of language representations during naturalistic reading.

Progress in neural combinatorial optimization for Dynamic Flexible Job Shop Scheduling Problem (DFJSP) is currently hindered by a methodological tension: static benchmarks encourage benchmark overfitting, while uncalibrated generators obscure algorithmic capability with stochastic noise. To resolve this, we introduce \textbf{DynaSchedBench}, a diagnostic framework for DFJSP that rigorously controls the instance-generation process. Instead of relying on parameter sampling, our approach utilizes Sequential Event-Space Calibrator (SESC) that computes a novel Schedule Stress Index (SSI) to stratify instances by difficulty. We demonstrate that SESC is substantially more computationally efficient than evolutionary baselines while converging reliably to the target metrics. The framework integrates modular components for instance generation, snapshot-based simulation, agents, evaluation, and visualization, thereby enabling rigorous testing of reactive and lookahead-based policies. Leveraging this calibrated environment, we identify key limitations of LLM-based scheduling agents. Specifically, in step-wise online decision-making for dynamic scheduling, we identify an ``Observability Paradox'': providing agents with oracle access to full structural information can degrade policy performance, underperforming concise information. Furthermore, despite substantial token overhead, tool-augmented and refinement strategies fail to reliably improve performance, and most LLM agents fail to consistently surpass strong dispatching baselines—behaving more like robust heuristic approximators than superior optimizers.

Image-to-video (I2V) diffusion models have recently made generative inbetweening a practical reality by synthesizing semantically plausible intermediate frames between two keyframes. Among them, inference-time sampling schemes that re-use large pre-trained I2V backbones without any additional training are especially attractive. Yet current methods frequently exhibit temporal inconsistency and artifacts such as ghosting or reverse motion. A key reason is that the two trajectories are driven by distinct motion priors, each inherited from its own conditioning frame, and are simply stitched together without explicitly reconciling these priors. We introduce Motion-Residual Conflict-Aware Time Reversal (MR‑CATR), an inference-time sampling framework that aligns conflicting motion priors instead of discarding one of them or collapsing to a single start-conditioned prior. MR‑CATR first derives a motion-residual–based direction from the forward path, combined with an end-conditioned residual to form a consensus motion axis. This design suppresses bidirectional motion conflicts while still allowing end-frame information to refine the trajectory and enforce endpoint consistency. MR‑CATR can be seamlessly integrated into existing time-reversal samplers without changing model parameters. Experiments on generative inbetweening benchmarks show that our method produces videos with smoother motion, fewer artifacts, and consistently better quantitative scores and user preferences than prior strategies.

Recently, large language models (LLMs) have shown remarkable reasoning abilities by producing long reasoning traces. However, as the sequence length grows, the key-value (KV) cache expands linearly, incurring significant memory and computation costs. Existing KV cache eviction methods mitigate this issue by discarding less important KV pairs, but often fail to capture complex KV dependencies, resulting in performance degradation. To better balance efficiency and performance, we introduce ForesightKV, a training-based KV cache eviction framework that learns to predict which KV pairs to evict during long-text generations. We first design the Golden Eviction algorithm, which identifies the optimal eviction KV pairs at each step using future attention scores. These traces and the scores at each step are then distilled via supervised training with a Pairwise Ranking Loss. Furthermore, we formulate cache eviction as a Markov Decision Process and apply the GRPO algorithm to mitigate the significant language modeling loss increase on low-entropy tokens. Experiments on AIME2024 and AIME2025 benchmarks of three reasoning models demonstrate that ForesightKV consistently outperforms prior methods under only half the cache budget, while benefiting synergistically from both supervised and reinforcement learning approaches. Code is available at \url{https://github.com/RUCAIBox/ForesightKV}.

A number of scientific conferences and journals have recently enacted policies that prohibit LLM usage by peer reviewers, except for polishing, paraphrasing, and grammar correction of otherwise human-written reviews. But, are these policies enforceable? To answer this question, we assemble a dataset of peer reviews simulating multiple levels of human-AI collaboration, and evaluate five state-of-the-art detectors, including two commercial systems. Our analysis shows that all detectors misclassify a non-trivial fraction of LLM-polished reviews as AI-generated, thereby risking false accusations of academic misconduct. We further investigate whether peer-review-specific signals, including access to the paper manuscript and the constrained domain of scientific writing, can be leveraged to improve detection. While incorporating such signals yields measurable gains in some settings, we identify limitations in each approach and find that none meets the accuracy standards required for identifying AI use in peer reviews. Importantly, our results suggest that recent public estimates of AI use in peer reviews through the use of current AI-text detectors should be interpreted with caution, as they misclassify mixed reviews (collaborative human-AI outputs) as fully AI generated, potentially overstating the extent of policy violations.

Accurate and efficient Video Quality Assessment (VQA) has long been a key research challenge. Current mainstream VQA methods typically improve performance by pretraining on large-scale classification datasets, followed by fine-tuning on VQA datasets. However, this strategy presents two significant challenges: (1) merely transferring semantic knowledge learned from pretraining is insufficient for VQA, as video quality depends on multiple factors (e.g., semantics, distortion, motion); (2) pretraining on large-scale datasets demands enormous computational resources, often dozens to hundreds of times more than training on VQA datasets. Recently, Contrastive Vision-Language Models (CVLMs) have shown strong generalization across visual tasks and promising potential for quality assessment. In this work, we propose Q-CLIP, the first fully CVLMs-based framework for VQA. Q-CLIP enhances both visual and textual representations through a Shared Cross-Modal Adapter (SCMA), which contains only a minimal number of trainable parameters and is the only component that requires training. This design significantly reduces computational cost. In addition, we introduce a set of five learnable quality-level prompts to guide the CVLMs in perceiving subtle quality variations. Furthermore, we investigate the impact of different frame sampling strategies on VQA performance. Extensive experiments demonstrate that Q-CLIP exhibits excellent performance on several VQA datasets. Code: https://github.com/xiao-mi-d/Q-CLIP

Post-training alignment often reduces LLM diversity, leading to a phenomenon known as mode collapse. Unlike prior work that attributes this effect to algorithmic limitations, we identify a fundamental, pervasive data-level driver: typicality bias in preference data, whereby annotators systematically favor familiar text as a result of well-established findings in cognitive psychology. We formalize this bias theoretically, verify it empirically on preference datasets, and show that it plays a central role in mode collapse. Motivated by this analysis, we introduce Verbalized Sampling (VS), a simple, training-free prompting strategy to circumvent mode collapse. VS prompts the model to verbalize a probability distribution over a set of responses (e.g., "Generate 5 jokes about coffee and their corresponding probabilities"), which relieves the pressure to produce a single "typical" answer. Experiments show that VS significantly improves performance across creative writing (poems, stories, jokes), social dialogue simulation, synthetic data generation, and open-ended QA, without sacrificing safety and factual accuracy. For instance, in creative writing, VS increases diversity by 1.6-2.1x compared to direct prompting. We further observe an emergent trend that more capable models benefit more from VS. In sum, our work provides a new data-centric perspective on mode collapse and a practical inference-time remedy that helps unlock pre-trained generative diversity.

Bayesian Optimization is critically vulnerable to extreme outliers. Existing provably robust methods typically assume a bounded cumulative corruption budget, which makes them defenseless against even a single corruption of sufficient magnitude. To address this, we introduce a new adversary whose budget is only bounded in the frequency of corruptions, not in their magnitude. We then derive RCGP-UCB, an algorithm coupling the upper confidence bound (UCB) approach with a Robust Conjugate Gaussian Process (RCGP). We present stable and adaptive versions of RCGP-UCB, and prove that they achieve sublinear regret in the presence of up to $O(T^{1/4})$ and $O(T^{1/7})$ corruptions with possibly infinite magnitude. This robustness comes at near zero cost: without outliers, RCGP-UCB's regret bounds match those of the standard GP-UCB algorithm.

Uncertainty quantification (UQ) in graph neural networks (GNNs) is crucial in high-stakes domains but remains a significant challenge. In graph settings, message passing often relies on strong assumptions such as exchangeability, which are rarely satisfied in practice. Moreover, achieving reliable UQ typically requires costly resampling or post-hoc calibration. To address these issues, we introduce Quantile-free Prediction Interval GNN (QpiGNN), a framework that builds on quantile regression (QR) to enable GNN-based UQ by directly optimizing coverage and interval width without requiring quantile inputs or post-processing. QpiGNN employs a dual-head architecture that decouples prediction and uncertainty, and is trained with label-only supervision through a quantile-free joint loss. This design allows efficient training and yields robust prediction intervals, with theoretical guarantees of asymptotic coverage and near-optimal width under mild assumptions. Experiments on 19 synthetic and real-world benchmarks show QpiGNN achieves average 22% higher coverage and 50% narrower intervals than baselines, while ensuring efficiency and robustness to noise and structural shifts.