Lucene will now intelligently merge HNSW graphs: elastic.co/search-labs/... Now indexing and merging is much cheaper, reducing the compute required and improving indexing throughput

Indexing and merging times are getting better for #Apache #Lucene vector search. Lucene has a read-only segment architecture. One of the drawbacks of this approach is throwing away previously completed work when merging HNSW graphs. Well, this got better :)

This this new algorithm, we have seen 3-5x fewer vector operations to achieve the same recall on previously horribly performing filter percentages.

We have implemented a variation of the ACORN-1. arxiv.org/abs/2403.04871 The key idea is expanding your HNSW neighborhood search, and only score candidates matching your filter criteria.

Filtered vector search is crazy important. So we made HNSW filtered search in Apache Lucene better. At similar recall, it can be 3-5x faster!

The number of improvements in Lucene here are crazy. Pretty much every count and boolean query gets a nice boost and some of the count improvements are hilarious 🚀🚀🚀.

But wait, you want push all those lower level retrievers combined with RRF through a cross-encoder for semantic reranking? Well, here you go: [🧵 cont.]

Want to combine any number of different query signals via RRF? We got you covered: [🧵 cont.]