Vector Quantization for Scalable Vector Search

The first post in this series introduced vector search, its relevance in today’s world, and the important metrics used to characterize it. We can achieve dramatic gains in vector search systems by improving their internal vector representations, as the majority of the search runtime is spent bringing vectors from memory to compute their similarity with the query. The focus of this post, Locally-adaptive Vector Quantization (LVQ), accelerates the search, lowers the memory footprint, and preserves the efficiency of the similarity computation.