Improving RAG through Multi-Agent RL This work treats RAG as a multi-agent cooperative task to improve answer generation quality. It models RAG components like query rewriting, document selection, and answer generation as reinforcement learning agents working together toward generating accurate answers. Applies Multi-Agent Proximal Policy Optimization (MAPPO) to jointly optimize all agents with a shared reward based on answer quality. Besides improvements on popular benchmarks, the framework shows strong generalization capabilities in out-of-domain scenarios and maintains effectiveness across different RAG system configurations.

🤔How to align stronger LLMs with weaker LLMs without collapse? Thrilled to share that our paper, "MACPO: Weak-to-Strong Alignment via Multi-Agent Contrastive Preference Optimization" has been accepted at #ICLR2025, co-authored with @lingyongyan, @ZihanWa54274484, @yindawei, @jayren3, @mdr, and @zhaochun_ren Pape is here:

Announcing our 57 **Outstanding Area Chairs**!