Microsoft AI Introduces CoRAG (Chain-of-Retrieval Augmented Generation): An AI Framework for Iterative Retrieval and Reasoning in Knowledge-Intensive Tasks

We are a community of AI/ ML/Generative AI enthusiasts/researchers/journalists/writers who share interesting news and articles about the applications of AI. 

Machine Learning News

CoRAG (Chain-of-Retrieval Augmented Generation) is a new AI framework developed by Microsoft and Renmin University of China that enhances Retrieval-Augmented Generation (RAG) models. It trains models to iteratively retrieve and reason before generating answers, using rejection sampling to form intermediate retrieval chains. This approach addresses common issues in RAG systems, improves multi-hop reasoning tasks, and achieves state-of-the-art results on various benchmarks, including KILT. CoRAG allows for dynamic adjustment of test-time retrieval and supports diverse decoding strategies, optimizing the trade-off between performance and computational efficiency.