ICLR 2026 Papers

Molecular understanding is central to advancing areas such as scientific and drug discovery, yet Large Language Models (LLMs) struggle to understand molecular graphs effectively. Existing graph–LLM bridges often adapt the Q-Former-style connector with fixed-length static tokens, which is originally designed for vision tasks. These designs overlook stereochemistry and substructural context and typically require costly LLM-backbone fine-tuning, limiting efficiency and generalization. We introduce **EDT-Former**, an **E**ntropy-guided **D**ynamic **T**oken Trans**former** that generates tokens aligned with informative molecular patches, thereby preserving both local and global structural features for molecular graph understanding. Beyond prior approaches, EDT-Former enables alignment between frozen graph encoders and LLMs without tuning the LLM backbone (excluding the embedding layer), resulting in computationally efficient finetuning, and achieves state-of-the-art results on MoleculeQA, Mol-Instructions, and property prediction benchmarks (TDC, MoleculeNet), underscoring its effectiveness for scalable and generalizable multimodal molecular understanding.

Multi-modal Large Language Models (MLLMs) have demonstrated strong capabilities in general-purpose perception and reasoning, but they still struggle with tasks that require spatial understanding of the 3D world. To address this, we introduce pySpatial, a visual programming framework that equips MLLMs with the ability to interface with spatial tools via Python code generation. Given an image sequence and a natural-language query, the model composes function calls to spatial tools including 3D reconstruction, camera-pose recovery, novel-view rendering, etc. These operations convert raw 2D inputs into an explorable 3D scene, enabling MLLMs to reason explicitly over structured spatial representations. Notably, pySpatial requires no gradient-based fine-tuning and operates in a fully zero-shot setting. Experimental evaluations on the challenging MindCube and Omni3D-Bench benchmarks demonstrate that our framework pySpatial consistently surpasses strong MLLM baselines; for instance, it outperforms GPT-4.1-mini by 12.94% on MindCube. Furthermore, we conduct real-world indoor navigation experiments where the robot can successfully traverse complex environments using route plans generated by pySpatial, highlighting the practical effectiveness of our approach. Our project website will be available at https://pySpatial.github.io.

Achieving generalization in robotic manipulation remains a critical challenge, particularly for unseen scenarios and novel tasks. Current Vision-Language-Action (VLA) models, while building on top of general Vision-Language Models (VLMs), still fall short of achieving robust zero-shot performance due to the scarcity and heterogeneity prevalent in embodied datasets. To address these limitations, we propose FSD (From Seeing to Doing), a novel vision-language model that generates intermediate representations through spatial relationship reasoning, providing fine-grained guidance for robotic manipulation. Our approach combines a hierarchical data construction pipeline for training with a self-consistency mechanism that aligns spatial coordinates with visual signals. Through extensive experiments, we comprehensively validated FSD’s capabilities in both “seeing” and “doing”, achieving outstanding performance across 8 benchmarks for general spatial reasoning and embodied reference abilities, as well as on our proposed more challenging benchmark VABench. We also verified zero-shot capabilities in robot manipulation, demonstrating significant performance improvements over baseline methods in both SimplerEnv and real robot settings. Experimental results show that FSD achieves 40.6% success rate in SimplerEnv and 72% success rate across 8 real-world tasks, outperforming the strongest baseline by 30%.

Large Language Models (LLMs) demonstrate strong performance on mathematical problems when prompted with Chain-of-Thought (CoT), yet it remains unclear whether this success stems from search, rote procedures, or rule-consistent reasoning. To address this, we propose modeling CoT as a certain rule-based stochastic process over directed acyclic graphs (DAGs), where nodes represent intermediate derivation states and edges encode rule applications. Within this framework, we introduce **logical closeness**, a metric that quantifies how well a model’s CoT trajectory (i.e., the LLM's output) adheres to the DAG structure, providing evaluation beyond classical PASS@$k$ metrics. Building on this, we introduce the *DAG-MATH* CoT format and construct a benchmark that guides LLMs to generate CoT trajectories in this format, thereby enabling the evaluation of their reasoning ability under our framework. Across standard mathematical reasoning datasets, our analysis uncovers statistically significant differences in reasoning fidelity among representative LLM families—even when PASS@$k$ is comparable—highlighting gaps between final-answer accuracy and rule-consistent derivation. Our framework provides a balance between free-form CoT and formal proofs systems, offering actionable diagnostics for LLMs reasoning evaluation.

Large Vision-Language Models (LVLMs) excel in diverse cross-modal tasks. However, object hallucination, where models produce plausible but inaccurate object descriptions, remains a significant challenge. In contrast to previous work focusing on LLM components, this paper is the first to trace LVLM hallucinations to visual encoders and identifies three key issues: statistical bias, inherent bias, and vulnerability. To address these challenges, we propose SHIELD, a training-free framework that mitigates hallucinations through three strategies: re-weighting visual tokens to reduce statistical bias, introducing noise-derived tokens to counter inherent bias, and applying adversarial attacks with contrastive decoding to address vulnerability. Experiments demonstrate that SHIELD effectively mitigates object hallucinations across diverse benchmarks and LVLM families. Moreover, SHIELD achieves strong performance on the general LVLM benchmark, highlighting its broad applicability. Code is available at https://github.com/hukcc/SHIELD.

The key-value (KV) cache plays a vital role in accelerating autoregressive inference for large language models (LLMs). However, its linear memory growth with sequence length poses significant memory bottlenecks, especially in long-context scenarios. Quantization offers a promising solution for memory efficiency. While existing methods typically apply channel-wise quantization to the key cache and token-wise quantization to the value cache, they suffer from severe performance degradation under low-bit configurations. Our analysis reveals that quantization sensitivity varies across individual KV channels, presenting an opportunity for non-uniform bit allocation. Following this finding, we propose ChanMix, a mixed-precision quantization framework that supports channel-wise quantization on 2-bit setting with FP8 precision with a custom Triton kernel implementation. To improve low-bit quantization performance, we introduce a channel-aware bit reallocation strategy, which allocates bits across channel sensitivity. Through extensive evaluation, ChanMix demonstrates superior performance across the NIAH, RULER, and InfiniteBench benchmarks for the Llama, Mistral, and Qwen model families, achieving improvements of at least 5 absolute percentage points on RULER compared to all baseline methods. Additionally, ChanMix enables a 2.3× increase in batch size and supports a 1.5× longer context length during inference.

Sparse Mixture-of-Experts (MoE) allows scaling of language and vision models efficiently by activating only a small subset of experts per input. While this reduces computation, the large number of parameters still incurs substantial memory overhead during inference. Post-training quantization has been explored to address this issue. Because uniform quantization suffers from significant accuracy loss at low bit-widths, mixed-precision methods have been recently explored; however, they often require substantial computation for bit-width allocation and overlook the varying sensitivity of model performance to the quantization of different experts. We propose a theoretically grounded expert-wise mixed-precision strategy that assigns bit-width to each expert primarily based on their *change in router’s* $l_2$ *norm* during training. Experts with smaller changes are shown to capture less frequent but critical features, and model performance is more sensitive to the quantization of these experts, thus requiring higher precision. Furthermore, to avoid allocating experts to lower precision that inject high quantization noise, experts with large *maximum intra-neuron variance* are also allocated higher precision. Experiments on large-scale MoE models, including Switch Transformer and Mixtral, show that our method achieves higher accuracy than existing approaches, while also reducing inference cost and incurring only negligible overhead for bit-width assignment.

We develop a cost-efficient neurosymbolic agent to address challenging multi-turn image editing tasks such as "Detect the bench in the image while recoloring it to pink. Also, remove the cat for a clearer view and recolor the wall to yellow." It combines the fast, high-level subtask planning by large language models (LLMs) with the slow, accurate, tool-use, and local A* search per subtask to find a cost-efficient toolpath---a sequence of calls to AI tools. To save the cost of A* on similar subtasks, we perform inductive reasoning on previously successful toolpaths via LLMs to continuously extract/refine frequently used subroutines and reuse them as new tools for future tasks in an adaptive fast-slow planning, where the higher-level subroutines are explored first, and only when they fail, the low-level A* search is activated. The reusable symbolic subroutines considerably save exploration cost on the same types of subtasks applied to similar images, yielding a human-like fast-slow toolpath agent ``FaSTA*'': fast subtask planning followed by rule-based subroutine selection per subtask is attempted by LLMs at first, which is expected to cover most tasks, while slow A* search is only triggered for novel and challenging subtasks. By comparing with recent image editing approaches, we demonstrate FaSTA* is significantly more computationally efficient while remaining competitive with the state-of-the-art baseline in terms of success rate.

The rapid progress of large vision-language models (VLMs), such as CLIP, has spurred the development of a wide range of neural decoding frameworks. Nevertheless, most existing approaches still suffer from semantic mismatches during representational alignment. This challenge may stem from the fact that the human brain does not distribute attention uniformly across a visual scene, but rather selectively encodes salient or relevant regions. Moreover, such selectivity is closely related to individual interests and varies from person to person. To address this challenge, we propose Yo'Mind, a novel optimal transport (OT)-driven personalized multimodal semantic masking framework designed to bridge the semantic gap between brain and machines in interpreting visual scenes. Technically, Yo'Mind introduces a dynamic semantic pruning and allocation mechanism that adaptively masks redundant visual semantic components in stimulus images based on individual neural responses—without requiring extra human supervision or hyperparameter tuning. This strategy can be used to enhance semantic consensus between brain and machine representations during decoding. Furthermore, the inherent flexibility of OT theory enables Yo'Mind to perform brain-visual-linguistic alignment and cross-subject decoding within a unified end-to-end architecture. Extensive experiments demonstrate that our Yo'Mind offers several advantages, including state-of-the-art brain-to-text reconstruction performance and improved interpretability of the decoding process.

Text-to-image retrieval is a fundamental task in vision--language learning, yet in real-world scenarios it is often challenged by short and underspecified user queries. Such queries are typically only one or two words long, making them semantically ambiguous, prone to collisions across diverse visual interpretations, and lacking explicit control over the quality of retrieved images. To address these issues, we propose a new paradigm of quality-controllable retrieval, which enriches short queries with contextual details while incorporating explicit notions of image quality. Our key idea is to leverage a generative large language model as a query completion function, extending underspecified queries into descriptive forms that capture fine-grained visual attributes such as pose, scene, and aesthetics. We introduce a training framework that conditions query completion on discretized quality levels, derived from relevance and aesthetic scoring models, so that query enrichment is not only semantically meaningful but also quality-aware. The resulting system provides three key advantages: {1} flexibility, as it is compatible with any pretrained vision--language model without modification; {2} transparency, since enriched queries are explicitly interpretable by users; and {3} controllability, enabling retrieval results to be steered toward user-preferred quality levels. Extensive experiments demonstrate that our proposed approach significantly improves retrieval results and provides effective quality control, bridging the gap between the expressive capacity of modern vision--language models and the underspecified nature of short user queries.

The training of large language models (LLMs) is expensive. In this paper, we study data-efficient approaches for pre-training LLMs, \ie, techniques that aim to optimize the Pareto frontier of model quality and training resource/data consumption. We seek to understand the tradeoffs associated with data selection routines based on (i) expensive-to-compute data-quality estimates, and (ii) maximization of coverage and diversity-based measures in the feature space. Our first technique, AskLLM, leverages the zero-shot reasoning capabilities of instruction-tuned LLMs to directly assess the quality of a training example. To target coverage, we propose density sampling, which models the data distribution to select a diverse sample. Testing the effect of $22$ different data curation techniques on the pre-training of T5-style of models, involving hundreds of pre-training runs and post fine-tuning evaluation tasks, we find that AskLLM and density are the best methods in their respective categories. While coverage sampling techniques often recover the performance of training on the entire dataset, training on data curated via AskLLM consistently outperforms full-data training---even when we sample only $10$\% of the original dataset, while converging up to $70$\% faster.

Universal multimodal embedding models have achieved great success in capturing semantic relevance between queries and candidates. However, current methods either condense queries and candidates into a single vector, potentially limiting the expressiveness for fine-grained information, or produce too many vectors that are prohibitively expensive for multi-vector retrieval. In this work, we introduce MetaEmbed, a new framework for multimodal retrieval that rethinks how multimodal embeddings are constructed and interacted with at scale. During training, a fixed number of learnable Meta Tokens are appended to the input sequence. At test-time, their last-layer contextualized representations serve as compact yet expressive multi-vector embeddings. Through the proposed Matryoshka Multi-Vector Retrieval training, MetaEmbed learns to organize information by granularity across multiple vectors. As a result, we enable test-time scaling in multimodal retrieval where users can balance retrieval quality against efficiency demands by selecting the number of tokens used for indexing and retrieval interactions. Extensive evaluations on the Massive Multimodal Embedding Benchmark (MMEB) and the Visual Document Retrieval Benchmark (ViDoRe) confirm that MetaEmbed achieves state-of-the-art retrieval performance while scaling robustly to models with 32B parameters. Code is available at https://github.com/facebookresearch/MetaEmbed.

Speech inherently contains rich acoustic information that extends far beyond the textual language. In real-world spoken communication, effective interpretation often requires integrating semantic meaning (e.g., content), paralinguistic features (e.g., emotions, speed, pitch) and phonological characteristics (e.g., prosody, intonation, rhythm), which are embedded in speech. While recent multimodal Speech Large Language Models (SpeechLLMs) have demonstrated remarkable capabilities in processing audio, their ability to perform fine-grained perception and complex reasoning in natural speech remains largely unexplored. To address this gap, we introduce MMSU, a comprehensive benchmark designed specifically for understanding and reasoning in speech. MMSU comprises 5,000 meticulously curated audio-question-answer triplets across 47 distinct tasks. Notably, linguistic theory forms the foundation of speech language understanding (SLU), yet existing benchmarks have paid insufficient attention to this fundamental aspect and fail to capture the broader linguistic picture. To ground our benchmark in linguistic principles, we systematically incorporate a wide range of linguistic phenomena, including phonetics, prosody, rhetoric, syntactics, semantics, and paralinguistics. Through a rigorous evaluation of 22 advanced SpeechLLMs, we identify substantial room for improvement in existing models. MMSU establishes a new standard for comprehensive assessment of SLLU, providing valuable insights for developing more sophisticated human-AI speech interaction systems.

Key-Value (KV) cache plays a pivotal role in accelerating inference in large language models (LLMs) by storing intermediate attention outputs, thereby avoiding redundant computation during auto-regressive generation. However, the cache's memory footprint scales linearly with sequence length, often resulting in memory bottlenecks on constrained hardware. While prior work has explored offloading KV-cache to the CPU and maintaining a reduced subset on the GPU, these approaches frequently suffer from imprecise token prioritization and degraded performance in long-generation tasks such as multi-turn dialogues and chain-of-thought reasoning. In this paper, we propose a novel KV-cache management strategy called IceCache, that integrates semantic token clustering with PagedAttention, a memory-efficient paging mechanism. By clustering semantically related tokens and organizing them into a hierarchical, dynamically updateable structure, our method improves cache hit rates and memory bandwidth utilization during CPU-GPU transfers. Experimental results show that IceCache achieves over 99\% accuracy with a 256-token budget and still maintains 97\% accuracy with only a 64-token budget, compared to the full KV-cache model. It outperforms existing baselines even while using just 25\% of the KV-cache token budget, demonstrating its superior accuracy in long-sequence scenarios.

Implicit neural representations (INRs) have recently emerged as a powerful paradigm for modeling data, offering a continuous alternative to traditional discrete signal representations. Their ability to compactly encode complex signals has led to strong performance across a wide range of computer vision tasks. In previous studies, it has been repeatedly shown that INR performance has a strong correlation with the activation functions used in its multilayer perceptrons. Although numerous competitive activation functions for INRs have been proposed, the theoretical foundations underlying their effectiveness remain poorly understood. Moreover, key challenges persist, including spectral bias (the reduced sensitivity to high-frequency signal content), limited robustness to noise, and difficulties in jointly capturing both local and global features. In this paper, we explore the underlying mechanism of INR signal representation, leveraging harmonic analysis and Chebyshev Polynomials. Through a rigorous mathematical proof, we show that modulating activation functions using a complex sinusoidal term yields better and complete spectral support throughout the INR network. To support our theoretical framework, we present empirical results over a wide range of experiments using Chebyshev analysis. We further develop a new activation function, leveraging the new theoretical findings to highlight its feasibility in INRs. We also incorporate a regularized deep prior, extracted from the signal via a task-specific model, to adjust the activation functions. This integration further improves convergence speed and stability across tasks. Through a series of experiments which include image reconstruction (with an average PSNR improvement of +5.67 dB over the nearest counterpart across a diverse image dataset), denoising (with a +0.46 dB increase in PSNR), super-resolution (with a +0.64 dB improvement over the nearest State-Of-The-Art (SOTA) method for 6X super-resolution), inpainting, and 3D shape reconstruction we demonstrate the novel proposed activation over existing state of the art activation functions.

The development of multimodal large language models (MLLMs) enables the evaluation of image quality through natural language descriptions. This advancement allows for more detailed assessments. However, these MLLM-based IQA methods primarily rely on general contextual descriptions, sometimes limiting fine-grained quality assessment. To address this limitation, we introduce a new image quality assessment (IQA) task paradigm, **grounding-IQA**. This paradigm integrates multimodal referring and grounding with IQA to realize more fine-grained quality perception, thereby extending existing IQA. Specifically, grounding-IQA comprises two subtasks: grounding-IQA-description (GIQA-DES) and visual question answering (GIQA-VQA). GIQA-DES involves detailed descriptions with precise locations (e.g., bounding boxes), while GIQA-VQA focuses on quality QA for local regions. To realize grounding-IQA, we construct a corresponding dataset, GIQA-160K, through our proposed automated annotation pipeline. Furthermore, we develop a well-designed benchmark, GIQA-Bench. The benchmark comprehensively evaluates the model grounding-IQA performance from three perspectives: description quality, VQA accuracy, and grounding precision. Experiments demonstrate that our proposed task paradigm, dataset, and benchmark facilitate the more fine-grained IQA application. Code will be made public.

Agentic multimodal models should not only comprehend text and images, but also actively invoke external tools, such as code execution environments and web search, and integrate these operations into reasoning. In this work, we introduce DeepEyesV2 and explore how to build an agentic multimodal model from the perspectives of data construction, training methods, and model evaluation. We observe that direct reinforcement learning alone fails to induce robust tool-use behavior. This phenomenon motivates a two-stage training pipeline: a cold-start stage to establish tool-use patterns, and reinforcement learning stage to further refine tool invocation. We curate a diverse, moderately challenging training dataset, specifically including examples where tool use is beneficial. We validate DeepEyesV2 across real-world understanding, mathematical reasoning, and search-intensive benchmarks, demonstrating that systematic tool integration enables reliable and extensible multimodal reasoning behaviour. Moreover, DeepEyesV2 exhibits task-adaptive tool invocation, tending to use image operations for perception tasks and numerical computations for reasoning tasks. Reinforcement learning further enable complex tool combinations and allowing model to selectively invoke tools based on problem context. We hope our study can provide guidance for community in developing agentic multimodal models.

Large language models (LLMs) have shown remarkable capabilities in language understanding and generation. However, such impressive capability typically comes with a substantial model size, which presents significant challenges in deployment and inference. While structured pruning of model parameters offers a promising way to reduce computational costs at deployment time, current methods primarily focus on single model pruning. In this work, we develop a novel strategy to compress models by strategically combining or merging layers from finetuned model variants, which preserves the original model's abilities by aggregating capabilities accentuated in different finetunes. We pose the optimal tailoring of these LLMs as a zero-order optimization problem, adopting a search space that supports three different operations: (1) Layer removal, (2) Layer selection from different candidate models, and (3) Layer merging. Our experiments demonstrate that this approach leads to competitive model pruning, for example, for the Llama2-13B model families, our compressed models maintain approximately 97.3\% of the original performance while removing ～25\% of parameters, significantly outperforming previous state-of-the-art methods.

Vision-Language Models (VLMs) have enabled computer use agents (CUAs) that operate GUIs autonomously, showing great potential, yet progress is limited by the lack of large-scale, open-source computer use data and foundation models. In this work, we introduce ScaleCUA, a step toward scaling open-source CUAs. It offers a large-scale dataset spanning 6 operating systems and 3 task domains, built via a closed-loop pipeline uniting automated agents with human experts. Trained on this scaled-up data, ScaleCUA can operate seamlessly across platforms. Specifically, it delivers strong gains over baselines (+26.6 on WebArena-Lite-v2, +10.7 on ScreenSpot-Pro) and sets new state-of-the-art results (94.4% on MMBench-GUI L1-Hard, 60.6% on OSWorld-G, 47.4% on WebArena-Lite-v2). These findings underscore the power of data-driven scaling for general-purpose computer use agents. We will release data, models, and code to advance future research.