DREAM：解耦风险以增强多模态大语言模型的安全对齐
DREAM: Disentangling Risks to Enhance Safety Alignment in Multimodal Large Language Models
AI总结：
本文提出DREAM方法，通过多模态风险解耦、监督微调和迭代RLAIF增强多模态大语言模型的安全对齐，在保持正常任务性能的同时显著提升安全性。
摘要：
多模态大语言模型（MLLMs）由于融合了视觉和文本数据，带来了独特的安全挑战，从而引入了新的潜在攻击维度和复杂的风险组合。在本文中，我们首先进行详细分析，旨在通过多模态输入中的逐步推理来解耦风险。我们发现，系统化的多模态风险解耦能够显著增强MLLMs的风险意识。通过利用多模态风险解耦的强大判别能力，我们进一步提出DREAM（解耦风险以增强MLLMs安全对齐），这是一种通过监督微调和迭代式基于AI反馈的强化学习（RLAIF）来增强MLLMs安全对齐的新方法。实验结果表明，DREAM在推理和训练阶段均显著提升了安全性，同时不影响正常任务上的性能（即避免过度安全），与GPT-4V相比，在SIUO安全与有效得分上实现了16.17%的提升。数据和代码可在https://github.com/Kizna1ver/DREAM获取。
英文摘要：
Multimodal Large Language Models (MLLMs) pose unique safety challenges due to their integration of visual and textual data, thereby introducing new dimensions of potential attacks and complex risk combinations. In this paper, we begin with a detailed analysis aimed at disentangling risks through step-by-step reasoning within multimodal inputs. We find that systematic multimodal risk disentanglement substantially enhances the risk awareness of MLLMs. Via leveraging the strong discriminative abilities of multimodal risk disentanglement, we further introduce \textbf{DREAM} (\textit{\textbf{D}isentangling \textbf{R}isks to \textbf{E}nhance Safety \textbf{A}lignment in \textbf{M}LLMs}), a novel approach that enhances safety alignment in MLLMs through supervised fine-tuning and iterative Reinforcement Learning from AI Feedback (RLAIF). Experimental results show that DREAM significantly boosts safety during both inference and training phases without compromising performance on normal tasks (namely oversafety), achieving a 16.17\% improvement in the SIUO safe\&effective score compared to GPT-4V. The data and code are available at https://github.com/Kizna1ver/DREAM.
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。
