ICML 2025 论文列表

Multimodal retrieval-augmented generation (RAG) plays a crucial role in domains such as knowledge-based visual question answering (KB-VQA), where models should effectively integrate additional knowledge to generate a response. However, existing vision and language models (VLMs) are not inherently designed for context-augmented generation, limiting their effectiveness in such tasks. While synthetic data generation has recently gained attention for training large VLMs, its application for context-augmented generation remains underexplored. To address this gap, we introduce SKVQA, a large-scale synthetic multimodal dataset containing over 2 million visual question-answer pairs, each associated with external knowledge sources to determine the final answer. Compared to previous datasets, SKVQA exhibits 11× more unique questions, greater domain diversity, and a broader spectrum of image sources. Through human evaluations, we confirm the high quality of the generated question-answer pairs and their contextual relevance. Extensive experiments show that SKVQA serves both as a challenging benchmark for knowledge-based VQA and as an effective training resource for adapting generative multimodal models to context-augmented generation. Our results further indicate that models trained on SKVQA demonstrate enhanced generalization in both context-aware VQA and multimodal RAG settings.

The success of large language models (LLMs) in text generation has inspired their application to image generation. However, existing methods either rely on specialized designs with inductive biases or adopt LLMs without fully exploring their potential in vision tasks. In this work, we systematically investigate the design space of LLMs for image generation and demonstrate that LLMs can achieve near state-of-the-art performance without domain-specific designs, simply by making proper choices in tokenization methods, modeling approaches, scan patterns, vocabulary design, and sampling strategies. We further analyze autoregressive models' learning and scaling behavior, revealing how larger models effectively capture more useful information than the smaller ones. Additionally, we explore the inherent differences between text and image modalities, highlighting the potential of LLMs across domains. The exploration provides valuable insights to inspire more effective designs when applying LLMs to other domains. With extensive experiments, our proposed model, **ELM** achieves an FID of 1.54 on 256$\times$256 ImageNet and an FID of 3.29 on 512$\times$512 ImageNet, demonstrating the powerful generative potential of LLMs in vision tasks.

Large language model (LLM) unlearning has demonstrated its essential role in removing privacy and copyright-related responses, crucial for their legal and safe applications. However, the pursuit of complete unlearning often comes with substantial costs due to its compromises in their general functionality, leading to a notorious trade-off between unlearning and retention. It motivates this paper to explore enhanced unlearning schemes that can mitigate this trade-off. Specifically, we propose Gradient Rectified Unlearning (GRU), an improved framework that regulates the directions of gradient updates during the unlearning procedure such that their side impacts on other, unrelated responses can be minimized. GRU is easy and general to implement, demonstrating practical effectiveness across a variety of well-established unlearning benchmarks.

Incorporating geometric invariance into neural networks enhances parameter efficiency but typically increases computational costs. This paper introduces new equivariant neural networks that preserve symmetry while maintaining a comparable number of floating-point operations (FLOPs) per parameter to standard non-equivariant networks. We focus on horizontal mirroring (flopping) invariance, common in many computer vision tasks. The main idea is to parametrize the feature spaces in terms of mirror-symmetric and mirror-antisymmetric features, i.e., irreps of the flopping group. This decomposes the linear layers to be block-diagonal, requiring half the number of FLOPs. Our approach reduces both FLOPs and wall-clock time, providing a practical solution for efficient, scalable symmetry-aware architectures.

We introduce *Lavender*, a simple supervised fine-tuning (SFT) method that boosts the performance of advanced vision-language models (VLMs) by leveraging state-of-the-art image generation models such as Stable Diffusion. Specifically, Lavender aligns the text-vision attention in the VLM transformer with the equivalent used by Stable Diffusion during SFT, instead of adapting separate encoders. This alignment enriches the model’s visual understanding and significantly boosts performance across in- and out-of-distribution tasks. Lavender requires just 0.13 million training examples---2.5\% of typical large-scale SFT datasets---and fine-tunes on standard hardware (8 GPUs) in a single day. It consistently improves state-of-the-art open-source multimodal LLMs (e.g., Llama-3.2-11B, MiniCPM-Llama3-v2.5), achieving up to 30\% gains and a 68\% boost on challenging out-of-distribution medical QA tasks. By efficiently transferring the visual expertise of image generators with minimal supervision, Lavender offers a scalable solution for more accurate vision-language systems. Code, training data, and models are available on the [project page](https://astrazeneca.github.io/vlm/).

The advancements in Multimodal Large Language Models (MLLMs) have enabled various multimodal tasks to be addressed under a zero-shot paradigm. This paradigm sidesteps the cost of model fine-tuning, emerging as a dominant trend in practical application. Nevertheless, Multimodal Sentiment Analysis (MSA), a pivotal challenge in the quest for general artificial intelligence, fails to accommodate this convenience. The zero-shot paradigm exhibits undesirable performance on MSA, casting doubt on whether MLLMs can perceive sentiments as competent as supervised models. By extending the zero-shot paradigm to In-Context Learning (ICL) and conducting an in-depth study on configuring demonstrations, we validate that MLLMs indeed possess such capability. Specifically, three key factors that cover demonstrations' retrieval, presentation, and distribution are comprehensively investigated and optimized. A sentimental predictive bias inherent in MLLMs is also discovered and later effectively counteracted. By complementing each other, the devised strategies for three factors result in average accuracy improvements of 15.9% on six MSA datasets against the zero-shot paradigm and 11.2% against the random ICL baseline.

The rise of vision-language foundation models marks an advancement in bridging the gap between human and machine capabilities in 3D scene reasoning. Existing 3D reasoning benchmarks assume real-time scene accessibility, which is impractical due to the high cost of frequent scene updates. To this end, we introduce *Hypothetical 3D Reasoning*, namely Hypo3D, a benchmark designed to evaluate models' ability to reason without access to real-time scene data. Models need to imagine the scene state based on a provided change description before reasoning. Hypo3D is formulated as a 3D Visual Question Answering (VQA) benchmark, comprising 7,727 context changes across 700 indoor scenes, resulting in 14,885 question-answer pairs. An anchor-based world frame is established for all scenes, ensuring consistent reference to a global frame for directional terms in context changes and QAs. Extensive experiments show that state-of-the-art foundation models struggle to reason effectively in hypothetically changed scenes. This reveals a substantial performance gap compared to humans, particularly in scenarios involving movement changes and directional reasoning. Even when the change is irrelevant to the question, models often incorrectly adjust their answers. The code and dataset are publicly available at: https://matchlab-imperial.github.io/Hypo3D.

Recent advancements in Large Vision Language Models (LVLMs) have significantly improved performance in Visual Question Answering (VQA) tasks through multimodal Retrieval-Augmented Generation (RAG). However, existing methods still face challenges, such as the scarcity of knowledge with reasoning examples and erratic responses from retrieved knowledge. To address these issues, in this study, we propose a multimodal RAG framework, termed RCTS, which enhances LVLMs by constructing a Reasoning Context-enriched knowledge base and a Tree Search re-ranking method. Specifically, we introduce a self-consistent evaluation mechanism to enrich the knowledge base with intrinsic reasoning patterns. We further propose a Monte Carlo Tree Search with Heuristic Rewards (MCTS-HR) to prioritize the most relevant examples. This ensures that LVLMs can leverage high-quality contextual reasoning for better and more consistent responses. Extensive experiments demonstrate that our framework achieves state-of-the-art performance on multiple VQA datasets, significantly outperforming In-Context Learning (ICL) and Vanilla-RAG methods. It highlights the effectiveness of our knowledge base and re-ranking method in improving LVLMs.

Text-to-visual retrieval often struggles with semantic redundancy and granularity mismatches between textual queries and visual content. Unlike existing methods that address these challenges during training, we propose VISual Abstraction (VISA), a test-time approach that enhances retrieval by transforming visual content into textual descriptions using off-the-shelf large models. The generated text descriptions, with their dense semantics, naturally filter out low-level redundant visual information. To further address granularity issues, VISA incorporates a question-answering process, enhancing the text description with the specific granularity information requested by the user. Extensive experiments demonstrate that VISA brings substantial improvements in text-to-image and text-to-video retrieval for both short- and long-context queries, offering a plug-and-play enhancement to existing retrieval systems.

Code editing is a foundational task in software development, where its effectiveness depends on whether it introduces desired code property changes without changing the original code's intended functionality. Existing approaches often formulate code editing as an implicit end-to-end task, omitting the fact that code-editing procedures inherently consist of discrete and explicit steps, and thus suffer from suboptimal performance and lack of robustness and generalization. We introduce EditLord, a code editing framework that makes the code transformation steps explicit. Our key insight is to employ a language model (LM) as an inductive learner to extract code editing rules from the training code pairs as concise meta-rule sets. Such rule sets will be manifested for each training sample to augment them for finetuning or assist in prompting- and iterative-based code editing. EditLord outperforms the state-of-the-art by an average of 22.7% in editing performance and 58.1% in robustness while achieving 20.2% higher functional correctness, across critical software engineering and security applications, LM models, and editing modes.

Advances in Parameter-efficient Fine-tuning (PEFT) bridged the performance gap with Full Fine-Tuning (FFT) through sophisticated analysis of pre-trained parameter spaces. Starting from drawing insights from Neural Engrams (NE) in Biological Neural Networks (BNNs), we establish a connection between the low-rank property observed during PEFT's parameter space shifting and neurobiological mechanisms. This observation leads to our proposed method, **S**ynapse and **N**euron (**SAN**), which decomposes and propagates the scaling component from anterior feature adjustment vectors towards posterior weight matrices. Our approach is theoretically grounded in Long-Term Potentiation/Depression (LTP/D) phenomena, which govern synapse development through neurotransmitter release modulation. Extensive experiments demonstrate its effectiveness: on **vision tasks** across VTAB, FGVC, and GIC (25 datasets) using ViT, Swin-T and ConvNeXt architectures, SAN outperforms FFT up to *8.7%* and LoRA by *3.2%*; on **language tasks** using Commonsense Reasoning (8 datasets) with LLaMA models (all generations), surpassing ChatGPT up to *8.5%* and LoRA by *4.7%*; on **vision-language tasks** using Visual Instruction Tuning (7 datasets) with LLaVA models, it exceeds FFT up to *2.4%* and LoRA by *1.9%*. Our code and W&B log will be released

In this work, we argue that large language models (LLMs), though trained to predict only the next token, exhibit emergent planning behaviors: $\textbf{their hidden representations encode future outputs beyond the next token}$. Through simple probing, we demonstrate that LLM prompt representations encode global attributes of their entire responses, including $\textit{structure attributes}$ (e.g., response length, reasoning steps), $\textit{content attributes}$ (e.g., character choices in storywriting, multiple-choice answers at the end of response), and $\textit{behavior attributes}$ (e.g., answer confidence, factual consistency). In addition to identifying response planning, we explore how it scales with model size across tasks and how it evolves during generation. The findings that LLMs plan ahead for the future in their hidden representations suggest potential applications for improving transparency and generation control.

We challenge the dominant focus on neural scaling laws and advocate for a paradigm shift toward downscaling in the development of large language models (LLMs). While scaling laws have provided critical insights into performance improvements through increasing model and dataset size, we emphasize the significant limitations of this approach, particularly in terms of computational inefficiency, environmental impact, and deployment constraints. To address these challenges, we propose a holistic framework for downscaling LLMs that seeks to maintain performance while drastically reducing resource demands. This paper outlines practical strategies for transitioning away from traditional scaling paradigms, advocating for a more sustainable, efficient, and accessible approach to LLM development.

As large language models evolve, there is growing anticipation that they will emulate human-like Theory of Mind (ToM) to assist with routine tasks. However, existing methods for evaluating machine ToM focus primarily on unimodal models and largely treat these models as black boxes, lacking an interpretative exploration of their internal mechanisms. In response, this study adopts an approach based on internal mechanisms to provide an interpretability-driven assessment of ToM in multimodal large language models (MLLMs). Specifically, we first construct a multimodal ToM test dataset, GridToM, which incorporates diverse belief testing tasks and perceptual information from multiple perspectives. Next, our analysis shows that attention heads in multimodal large models can distinguish cognitive information across perspectives, providing evidence of ToM capabilities. Furthermore, we present a lightweight, training-free approach that significantly enhances the model’s exhibited ToM by adjusting in the direction of the attention head.

Recent studies have raised significant concerns regarding the vulnerability of Large Vision Language Models (LVLMs) to maliciously injected or perturbed input images, which can mislead their responses. Existing defense methods show that such vision attacks are sensitive to image modifications especially cropping, using majority voting across responses of modified images as corrected responses. However, these modifications often result in partial images and distort the semantics, which reduces response quality on clean images after voting. Instead of directly using responses from partial images for voting, we investigate using them to supervise (guide) the LVLM's responses to the original images at inference time. We propose a black-box, training-free method called **DPS (Defense through Partial-Perception Supervision)**. In this approach, the model is prompted using the responses generated by a model that perceives only a partial image. With DPS, the model can adjust its response based on partial image understanding when under attack, while confidently maintaining its original response for clean input. Empirical experiments show our method outperforms the baseline, cutting the average attack success rate by 76.3\% across six datasets on three popular models.

Prompt tuning adapts Vision-Language Models like CLIP to open-world tasks with minimal training costs. In this direction, one typical paradigm evaluates model performance **separately** on known classes (*i.e.*, base domain) and unseen classes (*i.e.*, new domain). However, real-world scenarios require models to handle inputs **without prior domain knowledge**. This practical challenge has spurred the development of **open-world prompt tuning**, which demands a unified evaluation of two stages: 1) detecting whether an input belongs to the base or new domain (**P1**), and 2) classifying the sample into its correct class (**P2**). What's more, as domain distributions are generally unknown, a proper metric should be insensitive to varying base/new sample ratios (**P3**). However, we find that current metrics, including HM, overall accuracy, and AUROC, fail to satisfy these three properties simultaneously. To bridge this gap, we propose $\mathsf{OpenworldAUC}$, a unified metric that jointly assesses detection and classification through pairwise instance comparisons. To optimize $\mathsf{OpenworldAUC}$ effectively, we introduce **Gated Mixture-of-Prompts (GMoP)**, which employs domain-specific prompts and a gating mechanism to dynamically balance detection and classification. Theoretical guarantees ensure generalization of GMoP under practical conditions. Experiments on 15 benchmarks in open-world scenarios show GMoP achieves SOTA performance on $\mathsf{OpenworldAUC}$ and other metrics.

Reasoning stands as a cornerstone of intelligence, enabling the synthesis of existing knowledge to solve complex problems. Despite remarkable progress, existing reasoning benchmarks often fail to rigorously evaluate the nuanced reasoning capabilities required for complex, real-world problemsolving, particularly in multi-disciplinary and multimodal contexts. In this paper, we introduce a graduate-level, multi-disciplinary, EnglishChinese benchmark, dubbed as Reasoning Bench (RBench), for assessing the reasoning capability of both language and multimodal models. RBench spans 1,094 questions across 108 subjects for language model evaluation and 665 questions across 83 subjects for multimodal model testing. These questions are meticulously curated to ensure rigorous difficulty calibration, subject balance, and cross-linguistic alignment, enabling the assessment to be an Olympiad-level multidisciplinary benchmark. We evaluate many models such as o1, GPT-4o, DeepSeek-R1, etc. Experimental results indicate that advanced models perform poorly on complex reasoning, especially multimodal reasoning. Even the top-performing model OpenAI o1 achieves only 53.2% accuracy on our multimodal evaluation. Data and code are made publicly available athttps://evalmodels.github.io/rbench/

Multimodal Machine Learning systems, particularly those aligning text and image data like CLIP/BLIP models, have become increasingly prevalent, yet remain susceptible to adversarial attacks. While substantial research has addressed adversarial robustness in unimodal contexts, defense strategies for multimodal systems are underexplored. This work investigates the topological signatures that arise between image and text embeddings and shows how adversarial attacks disrupt their alignment, introducing distinctive signatures. We specifically leverage persistent homology and introduce two novel Topological-Contrastive losses based on Total Persistence and Multi-scale kernel methods to analyze the topological signatures introduced by adversarial perturbations. We observe a pattern of monotonic changes in the proposed topological losses emerging in a wide range of attacks on image-text alignments, as more adversarial samples are introduced in the data. By designing an algorithm to back-propagate these signatures to input samples, we are able to integrate these signatures into Maximum Mean Discrepancy tests, creating a novel class of tests that leverage topological signatures for better adversarial detection.

It is difficult to anticipate the myriad challenges that a predictive model will encounter once deployed. Common practice entails a reactive, cyclical approach: model deployment, data mining, and retraining. We instead develop a proactive longtail discovery process by imagining additional data during training. In particular, we develop general model-based longtail signals, including a differentiable, single forward pass formulation of epistemic uncertainty that does not impact model parameters or predictive performance but can flag rare or hard inputs. We leverage these signals as guidance to generate additional training data from a latent diffusion model in a process we call Longtail Guidance (LTG). Crucially, we can perform LTG without retraining the diffusion model or the predictive model, and we do not need to expose the predictive model to intermediate diffusion states. Data generated by LTG exhibit semantically meaningful variation, yield significant generalization improvements on numerous image classification benchmarks, and can be analyzed by a VLM to proactively discover, textually explain, and address conceptual gaps in a deployed predictive model.