缩小视觉语言模型的模态差距，可显著提升其安全输出能力。
Bootstrapping LLM Robustness for VLM Safety via Reducing the Pretraining Modality Gap
- 通过正则化减少预训练时的模态差距
- 安全违规率降低最高达16.3%，性能不受影响
- 适合关注多模态模型安全性的研究者
确保视觉语言模型（VLMs）生成安全输出对其可靠部署至关重要。然而，相比于其语言模型（LLM）基线，大型视觉语言模型（LVLMs）的安全性急剧下降。即使输入空白或无关图像，LVLMs也可能对原本在纯文本场景中会被拒绝的提示生成有害回应。最近有研究推测，图像与文本表征之间的模态差距可能是导致LVLMs安全性下降的原因。但模态差距的大小如何影响安全性尚未被系统研究。本文发现，模态差距越大，VLMs的安全性越差，且该差距在预训练阶段引入并贯穿微调过程。受此启发，我们提出一种在预训练中减小模态差距的正则化方法。在LLaVA v1.5、ShareGPT4V和MiniGPT-4上的大量实验表明，该方法显著提升了LVLMs的安全对齐效果，使不安全响应率降低最高达16.3%，且不损害模型性能；还能进一步增强现有防御机制，提升上限达18.2%。
原文摘要 · Abstract (English)
Ensuring Vision-Language Models (VLMs) generate safe outputs is crucial for their reliable deployment. However, LVLMs suffer from drastic safety degradation compared to their LLM backbone. Even blank or irrelevant images can trigger LVLMs to generate harmful responses to prompts that would otherwise be refused in text-only contexts. The modality gap between image and text representations has been recently hypothesized to contribute to safety degradation of LVLMs. However, if and how the amount of modality gap affects LVLMs' safety is not studied. In this work, we show that the amount of modality gap is highly inversely correlated with VLMs' safety. Then, we show that this modality gap is introduced during pretraining LVLMs and persists through fine-tuning. Inspired by this observation, we propose a regularization to reduce the modality gap during pretraining. Our extensive experiments on LLaVA v1.5, ShareGPT4V, and MiniGPT-4 show that our method substantially improves safety alignment of LVLMs, reducing unsafe rate by up to 16.3% without compromising performance, and can further boost existing defenses by up to 18.2%.
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。