用单一向量检测视觉语言模型的对抗图像，高效且跨模型通用。
Effective and Efficient Adversarial Detection for Vision-Language Models via A Single Vector
- 提取VLM隐藏状态中的攻击方向向量进行实时检测。
- 在LLaVA和MiniGPT-4上实现90%以上检测准确率，延迟低于10ms。
- 适用于提升多模型视觉语言系统的安全性，适合安全研究者使用。
视觉语言模型（VLMs）易受对抗攻击，尤其是来自对抗图像的攻击，但该问题在现有文献中尚未得到充分研究。为推动这一关键安全问题的研究，我们构建了一个大规模的对抗图像数据集RADAR，包含多样化的有害响应。基于该数据集，我们提出一种名为NEARSIDE的新方法，通过从VLM隐藏状态中提炼出一个单一向量（即攻击方向），实现对输入中对抗图像的实时检测。在两个目标VLM（LLaVA和MiniGPT-4）上的大量实验表明，该方法具有高有效性、高效率及良好的跨模型迁移能力。代码已公开于https://github.com/mob-scu/RADAR-NEARSIDE。
原文摘要 · Abstract (English)
Visual Language Models (VLMs) are vulnerable to adversarial attacks, especially those from adversarial images, which is however under-explored in literature. To facilitate research on this critical safety problem, we first construct a new laRge-scale Adervsarial images dataset with Diverse hArmful Responses (RADAR), given that existing datasets are either small-scale or only contain limited types of harmful responses. With the new RADAR dataset, we further develop a novel and effective iN-time Embedding-based AdveRSarial Image DEtection (NEARSIDE) method, which exploits a single vector that distilled from the hidden states of VLMs, which we call the attacking direction, to achieve the detection of adversarial images against benign ones in the input. Extensive experiments with two victim VLMs, LLaVA and MiniGPT-4, well demonstrate the effectiveness, efficiency, and cross-model transferrability of our proposed method. Our code is available at https://github.com/mob-scu/RADAR-NEARSIDE
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。