Pierre Fernandez

Cluster with Auctions for Vector Search

Swann Bessa, Pierre Fernandez, Gergely Szilvasy, Matthijs Douze, Hervé Jégou

Large-scale approximate nearest neighbor search commonly relies on partitions for indexing: database vectors are partitioned into clusters, and for each query a probing function selects the clusters to be scanned. The query probing function and the database partition are rarely t…Read more