Weaviate ACORN: Filter-Agnostic Filtered Vector Search

AI Impact Summary

Weaviate’s ACORN implementation introduces a novel approach to filtered vector search by dynamically adjusting the HNSW graph index during query execution. This eliminates the need for pre-filtering or post-filtering, addressing the challenges of varying filter selectivity and correlation. By intelligently pruning the graph index based on filter criteria, ACORN significantly improves query performance and recall, particularly in scenarios with low-correlation filters, reducing unnecessary distance calculations and maintaining graph connectivity.