AdaIAT: 适应性增加对生成文本的关注以减轻大型视觉-语言模型中的幻觉
AdaIAT: Adaptively Increasing Attention to Generated Text to Alleviate Hallucinations in LVLM
AI总结：
AdaIAT通过自适应增加对生成文本的关注，有效减轻大型视觉-语言模型中的幻觉问题，同时保持语言连贯性和预测能力。
摘要：
幻觉已成为当前大型视觉-语言模型（LVLM）的发展和应用中的重大障碍。为了缓解幻觉，一种直观且有效的方法是在推理过程中直接增加对图像标记的关注权重。虽然这种方法有效降低了幻觉率，但往往导致重复描述。为此，我们首先分析了注意力模式，发现真实物体标记倾向于将更高的注意力分配给生成的文本，而不是幻觉的那些。这促使我们利用包含与指令相关的视觉信息和上下文知识的生成文本，以减轻幻觉的同时保持语言连贯性。因此，我们提出了注意力生成文本（IAT），并证明它在减少幻觉率的同时避免了重复描述。为了防止简单放大影响LVLM的固有预测能力，我们进一步探索了自适应IAT（AdaIAT），它采用层间阈值来控制干预时间，并针对每个注意力头的特征进行精细化放大幅度。分析和实验均证明了AdaIAT的有效性。几种LVLM的结果显示，AdaIAT有效减轻了幻觉（在LLaVA-1.5上将幻觉率$C_S$和$C_I$分别降低了35.8%和37.1%），同时保持了语言表现和预测能力，实现了有吸引力的权衡。
英文摘要：
Hallucination has been a significant impediment to the development and application of current Large Vision-Language Models (LVLMs). To mitigate hallucinations, one intuitive and effective way is to directly increase attention weights to image tokens during inference. Although this effectively reduces the hallucination rate, it often induces repetitive descriptions. To address this, we first conduct an analysis of attention patterns and reveal that real object tokens tend to assign higher attention to the generated text than hallucinated ones. This inspires us to leverage the generated text, which contains instruction-related visual information and contextual knowledge, to alleviate hallucinations while maintaining linguistic coherence. We therefore propose Attention to Generated Text (IAT) and demonstrate that it significantly reduces the hallucination rate while avoiding repetitive descriptions. To prevent naive amplification from impairing the inherent prediction capabilities of LVLMs, we further explore Adaptive IAT (AdaIAT) that employs a layer-wise threshold to control intervention time and fine-grained amplification magnitude tailored to the characteristics of each attention head. Both analysis and experiments demonstrate the effectiveness of AdaIAT. Results of several LVLMs show that AdaIAT effectively alleviates hallucination (reducing hallucination rates $C_S$ and $C_I$ on LLaVA-1.5 by 35.8% and 37.1%, respectively) while preserving linguistic performance and prediction capability, achieving an attractive trade-off.
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。
