Chain-of-Retrieval Augmented Generation
Topics: AI (Deep Learning), LLMO, Microsoft, Retrieval Augmented Generation (RAG)
This Microsoft paper introduces CoRAG (Chain-of-Retrieval Augmented Generation), a novel approach for training language models to perform iterative information retrieval and reasoning. The system allows models to break down complex queries into smaller sub-queries, retrieve relevant information step by step, and generate more accurate answers. The research demonstrates significant improvements over existing methods, particularly on multi-hop question answering tasks, and establishes new state-of-the-art results on the KILT benchmark.