On Recent Advances in Graph-based Approximate Vector Search

Vector search, which returns the vectors most similar to a given query vector from a large vector dataset, underlies many important applications such as search, recommendation, and LLMs. To trade for efficiency, approximate vector search is usually used in practice, which returns most rather than all of the top-𝑘 nearest neighbors for each query. In this talk, an introduction to vector search and to existing methods in it will be made, with emphasis on graph-based algorithms. Additionally, KBEST, our vector search library, will be presented, tailored for the latest Huawei Kunpeng 920 CPUs. It up to 2x times outperforms known SOTA vector search libraries, running on x86 CPUs.
This talk is supported by Alfa Future Grants for academic staff — financial support programme by Alfa Bank.