HNSW is an effective approach for approximate nearest neighbor search. It uses small world graphs that connect local points and have short paths between distant points. HNSW improves upon NSW by using long-range hubs as entry points, limiting the number of connections checked at each hub, and introducing a hierarchy of layers with different connection lengths.
Table of contents
What’s The Story With HNSW?ContentsApproximate Nearest Neighbour SearchSmall WorldsNavigable Small WorldsHierarchical Navigable Small Worlds5 Impressions