Article Source

• Towards a Learned Index Structure for Approximate Nearest Neighbor Search Query Processing

Towards a Learned Index Structure for Approximate Nearest Neighbor Search Query Processing

• Maximilian Hünemörder, Peer Kröger and Matthias Renz
• Similarity Search and Applications, SISAP 2021, Dortmund
• github: https://github.com/huenemoerder/kmean-lis

Abstract

In our short paper, we outline the idea of applying the concept of a learned index structure to approximate nearest neighbor query processing. We discuss different data partitioning approaches and show how the task of identifying the disc pages of potential hits for a given query can be solved by a predictive machine learning model.