Additions to HNSW in Vespa: ACORN-1 and Adaptive Beam Search

Jan Böker, a software engineer, discusses recent enhancements to Vespa's HNSW (Hierarchical Navigable Small World) algorithm used for approximate nearest-neighbor (ANN) search in applications like search and recommender systems. The blog post highlights the addition of ACORN-1 and adaptive beam search to improve filtered ANN search, which allows for query-time constraints. ACORN-1 aims to optimize the search by exploring 2-hop neighbors first, reducing unnecessary computations, while adaptive beam search provides a distance-based termination condition to improve recall with fewer distance computations. These enhancements are tested and compared to previous implementations, showing significant improvements in search performance, with ACORN-1 particularly effective in reducing response time without compromising recall. The post also explores the balance between response time and recall when implementing multi-hop neighbor exploration and discusses the implications of these innovations for future search strategies in Vespa.