发现视觉语言模型幻觉源于注意力头的角色错位，提出无需微调的轻量检测方法。
Role-Break in Attention Heads: Understanding and Detecting Hallucinations in VLMs
- 从注意力头层面揭示幻觉是上下文行为的局部偏离，称为角色错位。
- 不需微调，检测器在6个模型4个基准上平均AUROC达93.23，特征维度低于5000。
- 可定位幻觉词元，适合用于模型可解释性与幻觉修正场景。
尽管视觉语言生成模型取得显著进展，仍易产生与输入图像不符的幻觉内容。现有方法多针对单一幻觉模式（如图文不平衡），但真实幻觉由多种模式混合导致，单一信号难以跨模型和任务保持稳定。我们从统一的头级视角发现，幻觉引发的变化表现为各注意力头对其忠实上下文行为的局部偏离，称为角色错位。详细分析显示，这种偏离在注意力头、上下文来源和偏离方向上系统分布，且在保留头身份后信号可线性读取。基于此，我们构建了一个无需微调的轻量线性检测器，特征维度低于5000，在六个VLM和四个基准上平均AUROC达93.23。小规模干预实验表明，检测出的词元可在判别设置中直接处理。
原文摘要 · Abstract (English)
Despite remarkable progress in vision-language generation, Vision-Language Models (VLMs) remain prone to hallucinations, producing content that is inconsistent with or unsupported by the input image. Existing works largely design detection or mitigation methods around one specific hallucination pattern, such as visual-textual imbalance, but real VLM hallucinations arise from a mixture of multiple patterns, so signals bound to a single pattern struggle to remain stable across models and tasks. Under a unified head-level view, we find that hallucination-induced changes manifest as localized deviations from each head's faithful contextual behavior, a phenomenon we term Role-Break. Detailed analysis reveals that these deviations are systematically organized across attention heads, contextual sources, and deviation directions, and that the resulting signal is linearly readable once head identity is preserved. Based on these findings, we build a lightweight linear detector on top of Role-Break that requires no fine-tuning of the VLM, whose feature dimension stays below 5,000 and reaches an average AUROC of 93.23 across six VLMs and four benchmarks. A small-scale intervention experiment further shows that the detected tokens can be directly acted upon in the discriminative setting.
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。