Quantization-based fast inner product search
Topics: AI Mode, AIOverviews, Chunk Relevance, LLM Readability, LLMO / GEO, Passage based retrieval, Ranking, Scoring
This Google patent outlines an efficient system for approximating the inner product between a query and a large database of items, which are represented as vectors. The method addresses the issue of high computational and memory costs by quantizing the database vectors into smaller, compressed representations. This is achieved by dividing each vector into subspaces, creating a “codebook” for each subspace, and then assigning a codebook entry to each subspace, effectively reducing the amount of data that needs to be stored and processed.
More about Maximum Inner Product.
