CIIR@LiveRAG 2025: Optimizing Multi-Agent Retrieval Augmented Generation through Self-Training

TL;DR

This paper introduces mRAG, a multi-agent framework for retrieval-augmented generation that uses self-training and reward-guided sampling to improve collaboration and response quality in complex tasks.

Contribution

The paper proposes a novel multi-agent RAG system with self-training and reward-guided trajectory sampling, advancing the state-of-the-art in collaborative AI for complex information retrieval tasks.

Findings

mRAG outperforms baseline RAG models on SIGIR 2025 datasets

Self-training improves inter-agent collaboration and response quality

Framework demonstrates effectiveness in real-world complex tasks

Abstract

This paper presents mRAG, a multi-agent retrieval-augmented generation (RAG) framework composed of specialized agents for subtasks such as planning, searching, reasoning, and coordination. Our system uses a self-training paradigm with reward-guided trajectory sampling to optimize inter-agent collaboration and enhance response generation. Evaluated on DataMorgana-derived datasets during the SIGIR 2025 LiveRAG competition, mRAG outperforms conventional RAG baselines. We further analyze competition outcomes and showcase the framework's strengths with case studies, demonstrating its efficacy for complex, real-world RAG tasks.