<ul><li>Reinforcement Learning from Human Feedback (RLHF) has achieved success in fine-tuning large language models.</li><li>Existing RLHF frameworks assume homogeneous human preferences, limiting adaptability in personalized scenarios.</li><li>Low-Rank Adaptation (LoRA) is introduced to enable efficient learning of personalized reward models.</li><li>LoRA captures shared and individual-specific structures, addressing personalization requirements and data constraints.</li></ul>

A Shared Low-Rank Adaptation Approach to Personalized RLHF

Discover more