通过语义级图像编辑与注意力调控，提升视觉语言模型的因果可解释性。
V-SEAM: Visual Semantic Editing and Attention Modulating for Causal Interpretability of Vision-Language Models
- 基于语义层级的视觉编辑与注意力头调控，实现概念级干预。
- 发现正向注意力头在同语义层内共享，跨层差异大；负向头则广泛通用。
- 自动调节关键注意力头嵌入，显著提升LLaVA和InstructBLIP在三类VQA任务表现。
近年来，因果可解释性研究从语言模型扩展至视觉语言模型（VLMs），旨在通过输入干预揭示其内部机制。尽管文本干预常聚焦语义，但视觉干预通常依赖粗粒度像素扰动，难以提供多模态融合的语义洞察。本文提出V-SEAM框架，结合视觉语义编辑与注意力调控，实现对VLMs的因果解释。该方法支持概念级视觉操作，并识别在物体、属性、关系三个语义层次上对预测有正/负贡献的注意力头。实验发现，正向注意力头在相同语义层级内具有共享性，但跨层级差异明显；而负向注意力头则表现出广泛泛化特性。此外，我们设计了一种自动调节关键头嵌入的方法，在三个不同VQA基准测试中显著提升了LLaVA与InstructBLIP的性能。相关数据与代码已开源：https://github.com/petergit1/V-SEAM。
原文摘要 · Abstract (English)
Recent advances in causal interpretability have extended from language models to vision-language models (VLMs), seeking to reveal their internal mechanisms through input interventions. While textual interventions often target semantics, visual interventions typically rely on coarse pixel-level perturbations, limiting semantic insights on multimodal integration. In this study, we introduce V-SEAM, a novel framework that combines Visual Semantic Editing and Attention Modulating for causal interpretation of VLMs. V-SEAM enables concept-level visual manipulations and identifies attention heads with positive or negative contributions to predictions across three semantic levels: objects, attributes, and relationships. We observe that positive heads are often shared within the same semantic level but vary across levels, while negative heads tend to generalize broadly. Finally, we introduce an automatic method to modulate key head embeddings, demonstrating enhanced performance for both LLaVA and InstructBLIP across three diverse VQA benchmarks. Our data and code are released at: https://github.com/petergit1/V-SEAM.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。