ICML 2026 论文列表

Even after decades of advances in neural network training, the inherent robustness challenge remains open. While the sensitivity to adversarial perturbations is understandable given their intentional learning, the most surprising fact is the vulnerability to natural corruptions. Surprisingly, not only is the cause of this inherent vulnerability unknown, but the concern extends beyond traditional CNNs; it also applies to current models, including transformers and large foundation models. For the first time, through this work, we observe that natural corruptions often collapse the network's internal feature space into a high-entropy state, causing predictions to rely on a small subset of fragile features. Inspired by this, we propose a simple yet effective entropy-guided fine-tuning framework, Dem-HEC, that strengthens corruption robustness while maintaining clean accuracy. Our method generates high-entropy samples within a bounded perturbation region and repairs the model using both clean and high-entropy samples. We further combine this objective with distilling knowledge from a teacher snapshot to maintain stable predictions. The proposed Dem-HEC is effective across datasets ranging from small to large-resolution, from pure CNNs to transformers, and to large foundation models, including DinoV3. The proposed approach outperforms the state-of-the-art (SOTA) models not only in improving robustness but also in retaining or boosting clean accuracy.

Large Language Models are typically benchmarked by evaluating every model on every test query. For practitioners seeking the best model to deploy, this is often wasteful: if a model clearly performs worse than others, there is no need to precisely estimate its performance. Best-arm identification algorithms can be naturally applied to drastically reduce costs by adaptively allocating evaluation budget. Further, language models often respond similarly to the same prompt—a property previous work has tried to leverage with mixed success in different use cases. We propose Synchronized Successive Rejects (SySRs), augmenting the classical Successive Reject algorithm with paired comparisons. Unlike prior attempts to leverage model similarity in best-model identification, our approach is hyperparameter-free and enjoys performance guarantees that improve with the degree of similarity between evaluated models. Empirically, our method outperforms all baselines in terms of average error rate across 15 standard benchmarks, and in terms of worst-case budget for reliably identifying the best model.

Despite strong performance, large language models (LLMs) still suffer from hallucinations. Most existing mitigation methods operate at inference time, without addressing the alignment limitation: LLMs are not trained to recognize their own lack of knowledge, and therefore tend to generate plausible responses even when the required knowledge is missing. Preference alignment approaches encourage uncertainty expression or refusal to improve truthfulness, but often consequently degrade helpfulness. To address this trade-off, existing preference alignment methods typically treat truthfulness and helpfulness as either universally collaborative or universally conflicting objectives across all samples. In contrast, we show that these objectives are consistent for most samples and conflict only in a small subset—where adaptive trade-off is truly needed. Based on this insight, we propose Conflict-Aware Adaptive Margin Preference Alignment (CAMP), which explicitly models when conflicts arise and adaptively regulates optimization strength. Experiments on UltraFeedback and representative hallucination benchmarks demonstrate that CAMP consistently improves truthfulness while maintaining a favorable helpfulness trade-off compared to strong hallucination mitigation and multi-objective alignment baselines.

Algorithmic fairness is often studied in static or single-agent settings, yet many real-world decision-making systems involve multiple interacting entities whose multi-stage actions jointly influence long-term outcomes. Existing fairness methods applied at isolated decision points frequently fail to mitigate disparities that accumulate over time. Although recent work has modeled fairness as a sequential decision-making problem, it typically assumes centralized agents or simplified dynamics, limiting its applicability to complex social systems. We introduce **MAFE**, a suite of *Multi-Agent Fair Environments* designed to simulate realistic, modular, and dynamic systems in which fairness emerges from the interplay of multiple agents. We demonstrate MAFEs in three domains—loan processing, healthcare, and higher education—supporting heterogeneous agents, configurable interventions, and fairness metrics. The environments are open-source and compatible with standard multi-agent reinforcement learning (MARL) libraries, enabling reproducible evaluation of fairness-aware policies. Through extensive experiments on cooperative use cases, we demonstrate how MAFE facilitates the design of equitable multi-agent algorithms and reveals critical trade-offs between fairness, performance, and coordination. MAFE provides a foundation for systematic progress in dynamic, multi-agent fairness research.

We consider the problem of sampling compositional and discrete objects from a given unnormalized posterior distribution. Notably, recent studies have shown that this problem can be efficiently solved by learning a deterministic Markov Decision Process (MDP) that progressively builds each object in proportion to the posterior. In this work, however, we demonstrate that the Markovian assumption can both hamper signal propagation during training and catastrophically reduce the learned sampler's expressivity due to state aliasing. To address these issues, we propose lifting the MDP with a learnable latent dynamics that allows the underlying policy to depend on the entire past trajectory---and not only on the current state. In view of this, we refer to the resulting method as \emph{path-dependent discrete amortized inference}. Importantly, we provably extend existing learning algorithms for amortized samplers to our setting. In experiments on standard benchmark problems, we also show that our approach often leads to faster learning convergence and improved state space exploration relatively to prior techniques.

Long-context large language model (LLM) inference has become the norm for today's AI applications. However, it is severely bottlenecked by the increasing memory demands of its KV cache. Previous works have shown that self-speculative decoding with sparse attention, where tokens are drafted using a subset of the KV cache and verified in parallel against the full KV cache, speeds up inference in a lossless manner. However, they rely on a standalone KV selection algorithm to select the KV entries used for drafting and overlook the fact that the criticality of each KV entry is inherently computed during verification. In this paper, we propose Vegas, a self-speculative decoding method with verification-guided sparse attention. Vegas identifies critical KV cache entries as a byproduct of verification and computes attention only over these entries when drafting subsequent tokens. This not only improves the draft token acceptance rate but also incurs low KV selection overhead, thereby improving decoding throughput. Vegas achieves a 1.25$\times$–2.81$\times$ speedup in decoding throughput over default vLLM and a 1.15$\times$–1.29$\times$ speedup over state-of-the-art sparse attention-based self-speculative decoding methods. Our code is available at https://github.com/platformxlab/vegas.

Real-world datasets across image and text domains are often characterized by skewed class distributions and noisy annotations, which jointly degrade model performance, particularly on minority classes. Among existing solutions, active learning offers an effective and efficient paradigm by selectively querying the most informative and balanced samples for annotation. We propose an innovative active learning framework that mitigates class imbalance and selects the most informative samples to annotate. Leveraging foundation model priors, our algorithm enables imbalance-aware co-decisions between foundation model and small model to tackle noisy and imbalanced labels across various domains. We introduce the first study to systematically explore active learning under the dual challenges of label noise and class imbalance across image and text domains. Extensive experiments on imbalanced datasets demonstrate that our method achieves substantial annotation savings—over 50\% compared to the best active learning baseline—while preserving performance and robustness to label noise.

The explosive growth of open-source model repositories has created a Model Jungle, where checkpoints are frequently shared without adequate documentation or metadata. While weight-space learning offers a pathway to identify and analyze these models directly from their parameters, processing full-scale weights is computationally prohibitive. Probing-based methods have emerged as a lightweight alternative, extracting permutation-equivariant representations via learnable probe vectors. However, existing probing methods are limited by a single-view design: they capture first-order structures but fail to encode the rich, higher-order correlation patterns inherent in row--column interactions. To bridge this gap, we introduce MVProbe, a multi-perspective probing framework that synthesizes first-order signals with interaction-aware (Gram-based) views. Our approach is theoretically grounded; we analyze the scaling laws of different probing orders to derive a principled standardization and fusion strategy that ensures balanced contributions from all branches. On the Model Jungle benchmark, MVProbe consistently outperforms the state-of-the-art ProbeX across diverse architectures, including ResNet, SupViT, MAE, and DINO.

Multi-Hop Fact Verification requires complex reasoning across disparate evidence, posing significant challenges for Large Language Models , which may suffer from hallucinations and fractured logical chains. Existing methods, while improving transparency via Chain-of-Thought , often lack explicit modeling of the structural dependencies between evidence and claims. In this work, we introduce an SCM-inspired framework that grounds reasoning in explicit directed dependency graphs, treating verification as a constructive structural reasoning process rather than full causal inference with interventions or counterfactual semantics. We empirically identify an ``inverted U-shaped'' correlation between reasoning-chain length and accuracy, revealing that excessive structural complexity can degrade performance. To address this, we propose a rule-based reinforcement learning strategy using Group Relative Policy Optimization. This approach dynamically optimizes the trade-off between structural depth and conciseness. Extensive experiments on HoVer and EX-FEVER demonstrate that our SCM-GRPO framework outperforms strong baselines while producing more traceable reasoning structures for complex fact verification.

In retrieval-augmented generation, language models can generate incorrect responses if they fail to utilize query-relevant content from the retrieved evidence. This shifts the focus of uncertainty quantification (UQ) toward assessing contextual grounding, i.e., whether predictions are supported by query-relevant tokens. Recent UQ methods unpack language models to characterize how inputs are processed. Nevertheless, these methods focus on a few layers and overlook the whole progressive propagation within the model, thereby failing to fully capture the grounding dynamics essential for reliable uncertainty estimation. We use information flow to build a layer-wise trace that reveals each context token’s contribution to the output, providing an interpretable basis for assessing reliability. From this analysis, we introduce two measures to calibrate prediction confidence. The first, \textit{simulatability}, posits that a prediction is more likely to be correct when context token contributions align closely with their true relevance. The second, \textit{concentration}, asserts that a response is more likely to be correct when it is derived from a narrow, focused subset of tokens. Experiments show that our method achieves an average AUROC of 0.709, exceeding the runner-up performance of 0.676, while maintaining moderate computational cost.

Grading in large undergraduate STEM courses often yields minimal feedback due to heavy instructional workloads. We present a large-scale empirical study of AI grading on real, handwritten single-variable calculus work from the University of California, Irvine. Using OCR-conditioned large language models with structured, rubric-guided prompting, our system produces scores and formative feedback for thousands of handwritten, in-person proctored free-response quiz submissions from nearly 800 students included in the paper’s empirical analysis. In a setting with no single ground-truth label, we evaluate performance against official teaching-assistant grades, student surveys, and independent human review, finding strong alignment with TA scoring and a large majority of AI-generated feedback rated correct or acceptable across quizzes. Beyond calculus, this setting highlights core challenges in OCR-conditioned mathematical reasoning and partial-credit assessment. We analyze key failure modes, propose practical rubric- and prompt-design principles, and introduce a multi-perspective evaluation protocol for reliable real-course deployment. Building on the dataset and evaluation framework developed here, we outline a path toward a future standardized benchmark for AI grading of handwritten mathematics to support reproducible evaluation, transparent comparison, reliable deployment, and future research.

We give a very general and simple framework to incorporate predictions on requests for online covering problems in a rigorous and black-box manner. Our framework turns any online algorithm with competitive ratio $\rho(k, \cdot)$ depending on $k$, the number of arriving requests, into an algorithm with competitive ratio of $\rho(\eta, \cdot)$, where $\eta$ is the prediction error. With accurate enough prediction, the resulting competitive ratio breaks through the corresponding worst-case online lower bounds, and smoothly degrades as the prediction error grows. This framework directly applies to a wide range of well-studied online covering problems such as facility location, Steiner problems, set cover, parking permit, etc., and yields improved and novel bounds.

Large language models (LLMs) display strong comprehensive abilities, yet the internal mechanisms that support these behaviors remain insufficiently understood. In this work, we show that across a wide range of open-weight Transformers, a subset of neurons remains consistently highly activated during inference across tasks of multiple capability dimensions. By probing along the cross-task activation strength, an extremely sparse subset is isolated, whose removal causes a collapse in model behavior, which we term keystone neurons. Our analysis reveals that keystone neurons are a stable and intrinsic neuron subset of the model that is largely established during pretraining. The parameters associated with these neurons are tightly calibrated during the training process, and their precise values are critical for the capabilities of the model. Building on these insights, we propose a supervised fine-tuning approach that updates only keystone neurons, achieving task gains comparable to or even better than full-parameter fine-tuning while better preserving performance in other capability dimensions, despite modifying a much smaller number of parameters.

Motivated by discrete diffusion's success in language-vision modeling, we explore its potential for multi-view generation, a task dominated by continuous approaches. We introduce **ViewMask-1-to-3**, formulating multi-view generation as a discrete sequence modeling problem where each viewpoint is represented as visual tokens from MAGVIT-v2. Through **discrete diffusion** via masked token prediction, our approach **enables progressive multi-view generation via iterative token unmasking**, unifying language and vision in a shared token space. Importantly, simple random masking combined with self-attention naturally encourages cross-view consistency without specialized architectures or 3D geometric priors. Our method outperforms the baseline on the GSO and 3D-FUTURE benchmarks, ranking first on average across standard image metricsand, and achieving a 10.6% higher IoU than continuous diffusion models on 3D-FUTURE. Furthermore, the proposed framework can be naturally extended to support text-to-image generation and multimodal understanding, highlighting its potential toward a more unified paradigm for multimodal understanding and generation.

Modern segmentation models achieve strong predictive performance but remain largely opaque, limiting our ability to diagnose failures, understand dataset shift, or intervene in a principled manner. We introduce $\textbf{Med-SegLens}$, a model-diffing framework that decomposes segmentation model activations into interpretable latent features using sparse autoencoders trained on SegFormer and U-Net. Through cross-architecture and cross-dataset latent alignment across healthy, adult, pediatric, and sub-Saharan African glioma cohorts, we identify a stable backbone of shared representations, while dataset shift is driven by differential reliance on population-specific latents. We show that these latents act as causal bottlenecks for segmentation failures, and that targeted latent-level interventions can correct errors and improve cross-dataset adaption without retraining, recovering performance in 70% of failure cases and improving Dice score from 39.4% to 74.2%. Our results demonstrate that latent-level model diffing provides a practical and mechanistic tool for diagnosing failures and mitigating dataset shift in segmentation models.

Evaluating conditional coverage remains one of the most persistent challenges in assessing the reliability of predictive systems. Although conformal methods can give guarantees on marginal coverage, no method can guarantee to produce sets with correct conditional coverage, leaving practitioners without a clear way to interpret local deviations. To overcome sample-inefficiency and overfitting issues of existing metrics, we cast conditional coverage estimation as a classification problem. Conditional coverage is violated if and only if some classifier can achieve lower risk than the target coverage. Through the choice of a (proper) loss function, the resulting risk difference gives a conservative estimate of natural miscoverage measures such as L1 and L2 distance, and can even separate the effects of over- and under-coverage, as well as handle non-constant target coverages. We call the resulting family of metrics excess risk of the target coverage (ERT). We show experimentally that the use of modern classifiers provides much higher statistical power than simple classifiers underlying established metrics like CovGap. Additionally, we use our metric to benchmark different conformal prediction methods. Finally, we release an open-source package for ERT as well as previous conditional coverage metrics. Together, these contributions provide a new lens for understanding, diagnosing, and improving the conditional reliability of predictive systems.

Large vision–language models (LVLMs) perform well on multimodal tasks, but their ability to reason and precisely align visual and textual information still has room for improvement. In this study, we show that external visual cues, such as symbols or grid lines, help LVLMs form more accurate connections between visual components, such as objects, and their corresponding textual descriptions, improving their grounding and reasoning abilities. We introduce the concept of Grounding IDs, which are latent identifiers that arise within the model as a result of external cues structuring both visual and textual modalities. Our analysis reveals that partition-inducing external cues lead to Grounding IDs that make better alignment between corresponding visual and text representations, helping the model focus on relevant information. We find that Grounding IDs enhance attention between related components, improving cross-modal grounding and reducing hallucinations. Overall, our results show that Grounding IDs are a key mechanism that enables external cues to improve cross-modal alignment, reduce errors, and enhance the overall performance of LVLMs across a range of multimodal tasks.

Why do neural networks generalize well on natural data? Natural data originates from processes subject to specific physical constraints, such as temporal and spatial invariance, that make it easier to learn. We investigate the sufficiency of these properties using 2D cellular automata as a controlled testbed: systems that are perfectly local, symmetric, and deterministic. We find that these conditions alone are \textit{not sufficient} to predict the $k$-step evolution of a cellular automaton. We then examine smoothness (average sensitivity) as an additional criterion and find it predictive but still incomplete. Finally, we introduce a circuit complexity perspective, hypothesizing that natural functions are computable by small circuits. Junta coefficients, measuring the concentration of Fourier weight by interaction degree, provide a tighter predictor of learnability and a correspondence to combinatorial complexity. Across architectures (CNNs, transformers, MLPs), learnable functions are predominantly those with spectral weight concentrated at low degrees and therefore low complexity. These results would be consistent with the hypothesis that natural data is learnable because natural dynamics filters out complex, high-degree interactions.

Diffusion Transformers (DiTs) are increasingly adopted for long-video generation, yet inference is dominated by the quadratic cost of 3D full attention. Sparse attention mitigates this bottleneck by exploiting *token-pair redundancy* and pruning query-key interactions. Nevertheless, its effectiveness on video generation is often constrained by non-sparse attention heads, making it hard to strike a good balance between inference speed and generation quality. To address this, we identify another pervasive but overlooked redundancy specific to video DiTs: *frame-block similarity*, where frame-blocks in attention weights exhibit highly similar distributions and can be well approximated by lightweight linear calibration. Motivated by this observation, we propose **EchoAttention**, which jointly leverages *token-pair redundancy* (*Sparse* operator) and *frame-block similarity* (*Echo* operator), together with a fine-grained routing policy learned via three-stage distillation. This design enables efficient handling of both sparse and non-sparse heads, overcoming the inherent ceiling of purely sparse attention and yielding a better speed-quality trade-off. Across public video DiTs, EchoAttention consistently improves the speed-quality frontier over SOTA sparse-attention baselines, reducing end-to-end latency up to 2.42$\times$ with minimal quality loss.