注意力劫持者：检测并解耦LVLM中的注意力劫持以缓解幻觉
Attention Hijackers: Detect and Disentangle Attention Hijacking in LVLMs for Hallucination Mitigation
AI总结：
本文揭示LVLM解码中指令token导致的注意力劫持现象，并提出免训练的AID方法，通过检测、解耦与再解耦显著缓解多基准幻觉。
摘要：
尽管大型视觉语言模型（LVLMs）取得了成功，但它们仍容易产生幻觉。现有研究将幻觉原因归结为对图像token的视觉注意力不足，而我们的发现表明，幻觉也源于解码过程中指令token的干扰。直观而言，某些指令token在解码期间持续扭曲LVLMs的视觉感知，将其视觉注意力劫持到区分度较低的视觉区域。这种扭曲使其无法整合来自图像的更广泛上下文信息，最终导致幻觉。我们将该现象称为“注意力劫持”，其中具有破坏性的指令token充当“注意力劫持者”。为解决此问题，我们提出了一种新颖的免训练策略，即注意力劫持者检测与解耦（AID），旨在隔离劫持者的影响，使LVLMs能够依赖其具备上下文感知能力的内在注意力图。具体而言，AID包含三个组件：首先，注意力劫持者检测通过计算指令驱动的视觉显著性来识别注意力劫持者。其次，提出注意力解耦机制以遮蔽这些已识别劫持者的视觉注意力，从而减轻其对后续token的破坏性影响。最后，再解耦重新计算指令驱动与图像驱动视觉显著性之间的平衡，以避免过度遮蔽效应。大量实验表明，AID在多个基准上显著降低了各类LVLMs的幻觉。
英文摘要：
Despite their success, Large Vision-Language Models (LVLMs) remain vulnerable to hallucinations. While existing studies attribute the cause of hallucinations to insufficient visual attention to image tokens, our findings indicate that hallucinations also arise from interference from instruction tokens during decoding. Intuitively, certain instruction tokens continuously distort LVLMs' visual perception during decoding, hijacking their visual attention toward less discriminative visual regions. This distortion prevents them integrating broader contextual information from images, ultimately leading to hallucinations. We term this phenomenon 'Attention Hijacking', where disruptive instruction tokens act as 'Attention Hijackers'. To address this, we propose a novel, training-free strategy namely Attention HIjackers Detection and Disentanglement (AID), designed to isolate the influence of Hijackers, enabling LVLMs to rely on their context-aware intrinsic attention map. Specifically, AID consists of three components: First, Attention Hijackers Detection identifies Attention Hijackers by calculating instruction-driven visual salience. Next, Attention Disentanglement mechanism is proposed to mask the visual attention of these identified Hijackers, and thereby mitigate their disruptive influence on subsequent tokens. Finally, Re-Disentanglement recalculates the balance between instruction-driven and image-driven visual salience to avoid over-masking effects. Extensive experiments demonstrate that AID significantly reduces hallucination across various LVLMs on several benchmarks.
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。
