当感知超越认知：在视觉-语言模型中解构知识冲突
When Seeing Overrides Knowing: Disentangling Knowledge Conflicts in Vision-Language Models
AI总结：
本文研究视觉-语言模型如何解决跨模态冲突，通过引入WHOOPS-AHA!数据集，发现特定注意力头可调节模型对内部知识或视觉信息的偏好，提升冲突解决的精确度。
摘要：
视觉-语言模型（VLMs）越来越多地结合视觉和文本信息以执行复杂任务。然而，其内部知识与外部视觉输入之间的冲突可能导致幻觉和不可靠的预测。本文通过引入WHOOPS-AHA!数据集，一种包含多模态反事实查询的集合，故意与内部常识知识矛盾，研究VLMs解决跨模态冲突的机制。通过logit检查，我们识别出一小组注意力头，这些头调解了这种冲突。通过干预这些头，我们可以引导模型朝向其内部参数知识或视觉信息。我们的结果表明，这些头上的注意力模式有效定位影响视觉覆盖的图像区域，相比梯度方法提供了更精确的归因。
英文摘要：
Vision-language models (VLMs) increasingly combine visual and textual information to perform complex tasks. However, conflicts between their internal knowledge and external visual input can lead to hallucinations and unreliable predictions. In this work, we investigate the mechanisms that VLMs use to resolve cross-modal conflicts by introducing WHOOPS-AHA!, a dataset of multimodal counterfactual queries that deliberately contradict internal commonsense knowledge. Through logit inspection, we identify a small set of attention heads that mediate this conflict. By intervening in these heads, we can steer the model towards its internal parametric knowledge or the visual information. Our results show that attention patterns on these heads effectively locate image regions that influence visual overrides, providing a more precise attribution compared to gradient-based methods.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
