Retrieval Augmented Generation (RAG) is a popular method for adding information to Large Language Models (LLMs), and Vector DB is the core technology used for the retrieval part of RAG. In this presentation, I will focus on the mathematical background of the approximate nearest neighbor search algorithms that are central components of Vector DB. Specifically, I will provide details on graph-based search, the most widely used family of algorithms.

The 1st Japan-Korea Workshop on Artificial Intelligence, Dec. 27, 2023 @UTokyo

Yusuke Matsui (The University of Tokyo) https://yusukematsui.me/

December 26, 2023