通过解耦多模态风险，提升大模型安全对齐能力。
DREAM: Disentangling Risks to Enhance Safety Alignment in Multimodal Large Language Models
- 将视觉与文本风险逐步解耦，增强模型风险感知能力。
- 在保持正常任务性能前提下，安全有效得分提升16.17%。
- 适合关注多模态模型安全的开发者与研究者使用。
多模态大语言模型（MLLM）因融合视觉与文本数据，面临独特的安全挑战，易引入新型攻击与复杂风险组合。本文通过逐步推理分析，系统性地对多模态输入中的风险进行解耦，发现该方法显著提升了模型的风险意识。基于此，提出新方法DREAM（Disentangling Risks to Enhance Safety Alignment in MLLMs），结合监督微调与迭代式从AI反馈中强化学习（RLAIF），增强模型安全对齐。实验表明，DREAM在推理与训练阶段均显著提升安全性，且未造成正常任务性能下降（即无过度安全问题），相比GPT-4V在SIUO安全&有效评分上提升16.17%。代码与数据已开源于https://github.com/Kizna1ver/DREAM。
原文摘要 · Abstract (English)
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.
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。