GEAR: Graph-enhanced Agent for Retrieval-augmented Generation
Topics: AI (Deep Learning), Knowledge Graph, Retrieval Augmented Generation (RAG)
The document introduces GEAR, a Graph-enhanced Agent for Retrieval-augmented Generation designed to improve retrieval and reasoning in multi-hop question-answering tasks. By leveraging graph-based retrieval with an LLM-based agent, GEAR surpasses traditional retrieval methods in efficiency and accuracy, particularly for complex queries. The system uses a novel graph retriever (SyncGE) and a gist memory inspired by neurobiology to streamline multi-hop information retrieval and answer synthesis.