arXiv 2412.139492025-06-10cs.CLcs.CV

利用视觉感知头散度破解大型视觉语言模型中的幻觉密码

Cracking the Code of Hallucination in LVLMs with Vision-aware Head Divergence

Foundation Model Research Center, Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所基础模型研究中心）；School of Artificial Intelligence, University of Chinese Academy of Sciences（中国科学院大学人工智能学院）；National University of Singapore（新加坡国立大学）；Southeast University（东南大学）；Wuhan AI Research（武汉人工智能研究所）

Jinghan He, Kuan Zhu, Haiyun Guo, Junfeng Fang, Zhenglin Hua, Yuheng Jia, Ming Tang, Tat-Seng Chua, Jinqiao Wang

备注：: ACL2025

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出 VHD 度量识别 LVLM 中的视觉感知注意力头，并据此设计无需训练的 VHR 方法，以更低开销有效缓解多模态幻觉。

摘要：

大型视觉语言模型（LVLMs）在将大型语言模型（LLMs）与视觉输入相结合方面取得了实质性进展，实现了先进的多模态推理。尽管取得了成功，一个持续存在的挑战是幻觉——即生成的文本无法准确反映视觉内容——这会损害准确性和可靠性。现有方法聚焦于对齐训练或解码改进，但主要处理生成阶段的症状，而没有探究根本原因。在本研究中，我们调查了驱动 LVLMs 幻觉的内部机制，重点关注多头注意力模块。具体而言，我们提出了 Vision-aware Head Divergence（VHD，视觉感知头散度），这一度量用于量化注意力头输出对视觉上下文的敏感性。基于此，我们的发现揭示了更关注视觉信息的视觉感知注意力头的存在；然而，模型对其先验语言模式的过度依赖与幻觉密切相关。基于这些洞见，我们提出了 Vision-aware Head Reinforcement（VHR，视觉感知头强化），这是一种无需训练的方法，通过增强视觉感知注意力头的作用来缓解幻觉。大量实验表明，与最先进的幻觉缓解方法相比，我们的方法取得了更优性能，同时保持高效率，且额外时间开销可忽略不计。

英文摘要：

Large vision-language models (LVLMs) have made substantial progress in integrating large language models (LLMs) with visual inputs, enabling advanced multimodal reasoning. Despite their success, a persistent challenge is hallucination-where generated text fails to accurately reflect visual content-undermining both accuracy and reliability. Existing methods focus on alignment training or decoding refinements but primarily address symptoms at the generation stage without probing the underlying causes. In this work, we investigate the internal mechanisms driving hallucination in LVLMs, with an emphasis on the multi-head attention module. Specifically, we introduce Vision-aware Head Divergence (VHD), a metric that quantifies the sensitivity of attention head outputs to visual context. Based on this, our findings reveal the presence of vision-aware attention heads that are more attuned to visual information; however, the model's overreliance on its prior language patterns is closely related to hallucinations. Building on these insights, we propose Vision-aware Head Reinforcement (VHR), a training-free approach to mitigate hallucination by enhancing the role of vision-aware attention heads. Extensive experiments demonstrate that our method achieves superior performance compared to state-of-the-art approaches in mitigating hallucinations, while maintaining high efficiency with negligible additional time overhead.

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。