ICLR 2026 Papers

Prevailing Dataset Distillation (DD) methods leveraging generative models confront two fundamental limitations. First, despite pioneering the use of diffusion models in DD and delivering impressive performance, the vast majority of approaches paradoxically require a diffusion model pre-trained on the full target dataset, undermining the very purpose of DD and incurring prohibitive training costs. Second, although some methods turn to general text-to-image models without relying on such target-specific training, they suffer from a significant distributional mismatch, as the web-scale priors encapsulated in these foundation models fail to faithfully capture the target-specific semantics, leading to suboptimal performance. To tackle these challenges, we propose Core Distribution Alignment (CoDA), a framework that enables effective DD using only an off-the-shelf text-to-image model. Our key idea is to first identify the ``intrinsic core distribution'' of the target dataset using a robust density-based discovery mechanism. We then steer the generative process to align the generated samples with this core distribution. By doing so, CoDA effectively bridges the gap between general-purpose generative priors and target semantics, yielding highly representative distilled datasets. Extensive experiments suggest that, without relying on a generative model specifically trained on the target dataset, CoDA achieves performance on par with or even superior to previous methods with such reliance across all benchmarks, including ImageNet-1K and its subsets. Notably, it establishes a new state-of-the-art accuracy of 60.4\% at the 50-images-per-class (IPC) setup on ImageNet-1K. Our code is available on the project webpage: https://github.com/zzzlt422/CoDA

Spatial reasoning is a key aspect of cognitive psychology and remains a bottleneck for current vision-language models (VLMs). While extensive research has aimed to evaluate or improve VLMs' understanding of basic spatial relations, such as distinguishing left from right, near from far, and object counting, these tasks cover only the most elementary layer of spatial reasoning and are largely approaching saturation in the latest reasoning models. In this work, we introduce OmniSpatial, a comprehensive and challenging benchmark for spatial reasoning, grounded in cognitive psychology. OmniSpatial covers four major categories: dynamic reasoning, complex spatial logic, spatial interaction, and perspective-taking, with 50 fine-grained subcategories. Through careful manual annotation, we construct over 8.4K question-answer pairs. Extensive experiments show that both open- and closed-source VLMs exhibit significant limitations in comprehensive spatial reasoning. We also explore two strategies—PointGraph (explicit scene graph cues) and SpatialCoT (novel-view chain-of-thought)—to bolster spatial reasoning.

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.

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.

Vision-language models (VLMs) have demonstrated strong performance in image geolocation, a capability further sharpened by frontier multimodal large reasoning models (MLRMs). This poses a significant privacy risk, as these widely accessible models can be exploited to infer sensitive locations from casually shared photos, often at street-level precision, potentially surpassing the level of detail the sharer consented or intended to disclose. While recent work has proposed applying a blanket restriction on geolocation disclosure to combat this risk, these measures fail to distinguish valid geolocation uses from malicious behavior. Instead, VLMs should maintain contextual integrity by reasoning about elements within an image to determine the appropriate level of information disclosure, balancing privacy and utility. To evaluate how well models respect contextual integrity, we introduce VLM-GEOPRIVACY, a benchmark that challenges VLMs to interpret latent social norms and contextual cues in real-world images and determine the appropriate level of location disclosure. Our evaluation of 14 leading VLMs shows that, despite their ability to precisely geolocate images, the models are poorly aligned with human privacy expectations. They often over-disclose in sensitive contexts and are vulnerable to prompt-based attacks. Our results call for new design principles in multimodal systems to incorporate context-conditioned privacy reasoning.

Functional data, representing curves or trajectories, are ubiquitous in fields like biomedicine and motion analysis. A fundamental challenge is phase variability—temporal misalignments that obscure underlying patterns and degrade model performance. Current methods often address registration (alignment) and classification as separate, sequential tasks. This paper introduces DeepFRC, an end-to-end deep learning framework that jointly learns diffeomorphic warping functions and a classifier within a unified architecture. DeepFRC combines a neural deformation operator for elastic alignment, a spectral representation using Fourier basis for smooth functional embedding, and a class-aware contrastive loss that promotes both intra-class coherence and inter-class separation. We provide the first theoretical guarantees for such a joint model, proving its ability to approximate optimal warpings and establishing a data-dependent generalization bound that formally links registration fidelity to classification performance. Extensive experiments on synthetic and real-world datasets demonstrate that DeepFRC consistently outperforms state-of-the-art methods in both alignment quality and classification accuracy, while ablation studies validate the synergy of its components. DeepFRC also shows notable robustness to noise, missing data, and varying dataset scales. Code is available at https://github.com/Drivergo-93589/DeepFRC.

Puzzlehunts are a genre of complex, multi-step puzzles lacking well-defined problem definitions. In contrast to conventional reasoning benchmarks consisting of tasks with clear instructions and constrained environments, puzzlehunts requires discovering the underlying problem structure from multimodal evidence and iterative reasoning, mirroring real-world domains such as scientific discovery, exploratory data analysis, or investigative problem-solving. Despite progress in foundation models, their performance on open-ended settings remains largely untested. We introduce PuzzleWorld, a comprehensive benchmark of 667 puzzlehunt-style problems designed to assess step-by-step, open-ended, and creative multimodal reasoning. Each puzzle is annotated with the final solution, detailed reasoning traces, and cognitive skill labels, enabling holistic benchmarking and fine-grained diagnostic analysis. Most state-of-the-art models achieve only 1-4\% final answer accuracy. On PuzzleWorld, the best model solves only 14\% of puzzles and reaches 40\% stepwise accuracy, matching human puzzle novices but falling significantly behind puzzle enthusiasts. To demonstrate the value of our reasoning annotations, we show that fine-tuning a small model on reasoning traces boosts stepwise accuracy from 4\% to 11\%, which translates to improvements in downstream visual reasoning tasks. Our detailed error analysis reveals that current models exhibit myopic reasoning, are bottlenecked by the limitations of language-based inference, and lack sketching capabilities crucial for visual and spatial reasoning. We will publicly release PuzzleWorld to support future work on building more general, open-ended, and creative reasoning systems.

We revisit Proximal Policy Optimization (PPO) from a function-space perspective. Our analysis decouples policy evaluation and improvement in a reproducing kernel Hilbert space (RKHS): (i) A kernelized temporal-difference (TD) critic performs efficient RKHS-gradient updates using only one-step state–action transition samples. (ii) a KL-regularized, natural-gradient policy step exponentiates the evaluated action-value, recovering a PPO/TRPO-style proximal update in continuous state-action spaces. We provide non-asymptotic, instance-adaptive guarantees whose rates depend on RKHS entropy, unifying tabular, linear, Sobolev, Gaussian, and Neural Tangent Kernel (NTK) regimes, and we derive a sampling rule for the proximal update that ensures the optimal $k^{-1/2}$ convergence rate for stochastic optimization. Empirically, the theory-aligned schedule improves stability and sample efficiency on common control tasks (e.g., CartPole, Acrobot), while our TD-based critic attains favorable throughput versus a GAE baseline. Altogether, our results place PPO on a firmer theoretical footing beyond finite-dimensional assumptions and clarify when RKHS-proximal updates with kernel-TD critics yield global policy improvement with practical efficiency.

Based on the prevalent intermediate fusion (IF) and late fusion (LF) frameworks, multimodal representation learning (MML) demonstrates its superiority over unimodal representation learning. To investigate the intrinsic factors underlying the empirical success of MML, research grounded in theoretical justifications from the perspective of generalization error has emerged. However, these provable MML studies derive the theoretical findings based on LF, while theoretical exploration based on IF remains scarce. This naturally gives rise to a question: **Can we design a comprehensive MML approach supported by the sufficient theoretical analysis across fusion types?** To this end, we revisit the IF and LF paradigms from a fine-grained dimensional perspective. The derived theoretical evidence sufficiently establishes the superiority of IF over LF under a specific constraint. Based on a general $K$-Lipschitz continuity assumption, we derive the generalization error upper bound of the IF-based methods, indicating that eliminating the distribution incoherence can improve the generalizability of IF-based MML methods. Building upon these theoretical insights, we establish a novel IF-based MML method, which introduces the informatic constraint and performs distribution cohering. Extensive experimental results on multiple widely adopted datasets verify the effectiveness of the proposed method.

Foundation models have recently emerged as a promising approach for learning generalizable EEG representations for brain–computer interfaces (BCIs). Yet, their true advantages over traditional methods—particularly classical non-neural approaches—remain unclear. In this work, we present a comprehensive benchmark of state-of-the-art EEG foundation models, evaluated across diverse datasets, decoding tasks, and six evaluation protocols, with rigorous statistical testing. We introduce spatiotemporal EEGFormer (ST-EEGFormer), a simple yet effective Vision Transformer (ViT)-based baseline, pre-trained solely with masked autoencoding (MAE) on over 8M EEG segments. Our results show that while fine-tuned foundation models perform well in data-rich, population-level settings, they often fail to significantly outperform compact neural networks or even classical non-neural decoders in data-scarce scenarios. Furthermore, linear probing remains consistently weak, and performance varies greatly across downstream tasks, with no clear scaling law observed among neural network decoders. These findings expose a substantial gap between pre-training and downstream fine-tuning, often diminishing the benefits of complex pre-training tasks. We further identify hidden architectural factors that affect performance and emphasize the need for transparent, statistically rigorous evaluation. Overall, this study calls for community-wide efforts to construct large-scale EEG datasets and for fair, reproducible benchmarks to advance EEG foundation models.

While table understanding increasingly relies on pixel-only settings, current benchmarks predominantly use synthetic renderings that lack the complexity and visual diversity of real-world tables. Additionally, existing visual table understanding (VTU) datasets offer fixed examples with single visualizations and pre-defined instructions, providing no access to underlying serialized data for reformulation. We introduce TABLET, a large-scale VTU dataset with 4 million examples across 21 tasks, grounded in 2 million unique tables where 88% preserve original visualizations. To evaluate whether models are able to jointly reason over tabular and visual content, we also introduce VisualTableQA, a benchmark requiring both visual perception and table understanding. Fine-tuning vision-language models like Qwen2.5-VL-7B and Gemma 3-4B on TABLET improves performance on seen and unseen VTU tasks while increasing robustness on real-world table visualizations. By preserving original visualizations and maintaining example traceability in a unified large-scale collection, TABLET establishes a foundation for robust training and extensible evaluation of future VTU models.

Dynamic Text-Attributed Graphs (DyTAGs), which intricately integrate structural, temporal, and textual attributes, are crucial for modeling complex real-world systems. However, most existing DyTAG datasets exhibit poor textual quality, which severely limits their utility for generative DyTAG tasks requiring semantically rich inputs. Additionally, prior work mainly focuses on discriminative tasks on DyTAGs, resulting in a lack of standardized task formulations and evaluation protocols tailored for DyTAG generation. To address these critical issues, we propose \underline{G}enerative \underline{D}yTA\underline{G} \underline{B}enchmark (GDGB), which comprises eight meticulously curated DyTAG datasets with high-quality textual features for both nodes and edges, overcoming limitations of prior datasets. Building on GDGB, we define two novel DyTAG generation tasks: Transductive Dynamic Graph Generation (TDGG) and Inductive Dynamic Graph Generation (IDGG). TDGG transductively generates a target DyTAG based on the given source and destination node sets, while the more challenging IDGG introduces new node generation to inductively model the dynamic expansion of real-world graph data. To enable holistic evaluation, we design multifaceted metrics that assess the structural, temporal, and textual quality of the generated DyTAGs. We further propose GAG-General, an LLM-based multi-agent generative framework tailored for reproducible and robust benchmarking of DyTAG generation. Experimental results demonstrate that GDGB enables rigorous evaluation of TDGG and IDGG, with key insights revealing the critical interplay of structural and textual features in DyTAG generation. These findings establish GDGB as a foundational resource for advancing generative DyTAG research and unlocking further practical applications in DyTAG generation. The dataset and source code are available at \url{https://github.com/Lucas-PJ/GDGB-ALGO}.

Large language models (LLMs) are increasingly applied in scientific research, offering new capabilities for knowledge discovery and reasoning. In single-cell biology, however, evaluation practices for both general and specialized LLMs remain inadequate: existing benchmarks are fragmented across tasks, adopt formats such as multiple-choice classification that diverge from real-world usage, and rely on metrics lacking interpretability and biological grounding. We present **SC-ARENA**, a natural language evaluation framework tailored to single-cell foundation models. SC-ARENA formalizes a *virtual cell* abstraction that unifies evaluation targets by representing both intrinsic attributes and gene-level interactions. Within this paradigm, we define five natural language tasks — cell type annotation, captioning, generation, perturbation prediction, and scientific QA — that probe core reasoning capabilities in cellular biology. To overcome the limitations of brittle string-matching metrics, we introduce **knowledge-augmented evaluation**, which incorporates external ontologies, marker databases, and scientific literature to support biologically faithful and interpretable judgments. Experiments and analysis across both general-purpose and domain-specialized LLMs demonstrate that: (i) under the *Virtual Cell* unified evaluation paradigm, current models achieve uneven performance on biologically complex tasks, particularly those demanding mechanistic or causal understanding; and (ii) our knowledge-augmented evaluation framework ensures biological correctness, provides interpretable, evidence-grounded rationales, and achieves high discriminative capacity, overcoming the brittleness and opacity of conventional metrics. **SC-ARENA** thus provides a unified and interpretable framework for assessing LLMs in single-cell biology, pointing toward the development of biology-aligned, generalizable foundation models.

Positive-unlabeled (PU) learning is a weakly supervised binary classification problem, in which the goal is to learn a binary classifier from only positive and unlabeled data, without access to negative data. In recent years, many PU learning algorithms have been developed to improve model performance. However, experimental settings are highly inconsistent, making it difficult to identify which algorithm performs better. In this paper, we propose the first PU learning benchmark to systematically compare PU learning algorithms. During our implementation, we identify subtle yet critical factors that affect the realistic and fair evaluation of PU learning algorithms. On the one hand, many PU learning algorithms rely on a validation set that includes negative data for model selection. This is unrealistic in traditional PU learning settings, where no negative data are available. To handle this problem, we systematically investigate model selection criteria for PU learning. On the other hand, the problem settings and solutions of PU learning have different families, i.e., the one-sample and two-sample settings. However, existing evaluation protocols are heavily biased towards the one-sample setting and neglect the significant difference between them. We identify the internal label shift problem of unlabeled training data for the one-sample setting and propose a simple yet effective calibration approach to ensure fair comparisons within and across families. We hope our framework will provide an accessible, realistic, and fair environment for evaluating PU learning algorithms in the future.

Correcting motion artifacts in scientific and medical imaging is important, as they significantly impact image quality. However, evaluating deep learning-based and classical motion correction methods remains fundamentally difficult due to the lack of accessible ground-truth target data. To address this challenge, we study three evaluation approaches: real-world evaluation based on reference scans, simulated motion, and reference-free evaluation, each with its merits and shortcomings. To enable evaluation with real-world motion artifacts, we release PMoC3D, a dataset consisting of unprocessed $\textbf{P}$aired $\textbf{Mo}$tion-$\textbf{C}$orrupted $\textbf{3D}$ brain MRI data. To advance evaluation quality, we introduce MoMRISim, a feature-space metric trained for evaluating motion reconstructions. We assess each evaluation approach and find real-world evaluation together with MoMRISim, while not perfect, to be most reliable. Evaluation based on simulated motion systematically exaggerates algorithm performance, and reference-free evaluation overrates oversmoothed deep learning outputs.

Information tasks such as writing surveys or analytical reports require complex search and reasoning, and have recently been grouped under the umbrella of _deep research_ --- a term also adopted by recent models targeting these capabilities. Despite growing interest, the scope of the deep research task remains underdefined and its distinction from other reasoning-intensive problems is poorly understood. In this paper, we propose a formal characterization of the deep research (DR) task and introduce a benchmark to evaluate the performance of DR systems. We argue that the core defining feature of deep research is not the production of lengthy report-style outputs, but rather the high fan-out over concepts required during the search process, i.e., broad and reasoning-intensive exploration.To enable objective evaluation, we define DR using an intermediate output representation that encodes key claims uncovered during search—separating the reasoning challenge from surface-level report generation. Based on this formulation, we propose a benchmark LiveDRBench with 100 challenging tasks over scientific topics (e.g., datasets, materials discovery, prior art search) and public interest events (e.g., flight incidents, movie awards). Across state-of-the-art DR systems, F1 score ranges between 0.02 and 0.72 for any sub-category. OpenAI's model performs the best with an overall F1 score of 0.55. Analysis of the reasoning traces reveals that systems cover only about half of the necessary search queries, with proprietary models issuing broader and and deeper queries than open source models, highlighting gaps in both coverage and reasoning depth.

Deep Research Agents (DRAs) can autonomously conduct complex investigations and generate comprehensive reports, demonstrating strong real-world potential. However, existing evaluations mostly rely on close-ended benchmarks, while open-ended deep research benchmarks remain scarce and typically neglect personalized scenarios. To bridge this gap, we introduce Personalized Deep Research Bench, the first benchmark for evaluating personalization in DRAs. It pairs 50 diverse research tasks across 10 domains with 25 authentic user profiles that combine structured persona attributes with dynamic real-world contexts, yielding 250 realistic user-task queries. To assess system performance, we propose the PQR Evaluation Framework, which jointly measures (P) Personalization Alignment, (Q) Content Quality, and (R) Factual Reliability. Our experiments on a range of systems highlight current capabilities and limitations in handling personalized deep research. This work establishes a rigorous foundation for developing and evaluating the next generation of truly personalized AI research assistants.

Large language models (LLMs) have recently enabled coding agents capable of generating, executing, and revising visualization code. However, existing models often fail in practical workflows due to limited language coverage, unreliable execution, and lack of iterative correction mechanisms. Progress has been constrained by narrow datasets and benchmarks that emphasize single-round generation and single-language tasks. To address these challenges, we introduce three complementary resources for advancing visualization coding agents. **VisCode-Multi-679K** is a large-scale, supervised dataset containing 679K validated and executable visualization samples with multi-turn correction dialogues across 12 programming languages. **VisPlotBench** is a benchmark for systematic evaluation, featuring executable tasks, rendered outputs, and protocols for both initial generation and multi-round self-debug. Finally, we present **VisCoder2**, a family of multi-language visualization models trained on VisCode-Multi-679K. Experiments show that VisCoder2 significantly outperforms strong open-source baselines and approaches the performance of proprietary models like GPT-4.1, with further gains from iterative self-debug, reaching **82.4%** overall execution pass rate at the 32B scale, particularly in symbolic or compiler-dependent languages.

The adoption of Large Language Models (LLMs) as automated evaluators (LLM-as-a-judge) has revealed critical inconsistencies in current evaluation frameworks. We identify two fundamental types of inconsistencies: (1) \textit{Score-Comparison Inconsistency}, where lower-rated responses outperform higher-scored ones in pairwise comparisons, and (2) \textit{Pairwise Transitivity Inconsistency}, manifested through circular preference chains ($A\!>\!B\!>\!C\!>\!A$) and equivalence contradictions ($A\!=\!B\!=\!C\!\neq\!A$). We argue that these issues come from information loss in discrete rating systems and ambiguous tie judgments during pairwise evaluation. We propose \textbf{TrustJudge}, a probabilistic framework that addresses these limitations through two key innovations: 1) \textit{distribution-sensitive scoring} that computes continuous expectations from discrete rating probabilities, preserving information entropy for more precise scoring, and 2) \textit{likelihood-aware aggregation} that resolves transitivity violations using bidirectional preference probabilities or perplexity. We also formalize the theoretical limitations of current LLM-as-a-judge frameworks and demonstrate how TrustJudge’s components overcome them. When evaluated with Llama-3.1-70B-Instruct as judge using our dataset, TrustJudge reduces Score-Comparison inconsistency by 8.43\% (from 23.32\% to 14.89\%) and Pairwise Transitivity inconsistency by 10.82\% (from 15.22\% to 4.40\%), while maintaining higher evaluation accuracy. Our work provides the first systematic analysis of evaluation framework inconsistencies in LLM-as-a-judge paradigms, offering both theoretical insights and practical solutions for reliable automated assessment. The framework demonstrates consistent improvements across various model architectures and scales, enabling more trustworthy LLM evaluation without requiring additional training or human annotations.