Finding Relevant Items Using Approximate Nearest Neighbor Search Algorithms
Scalable Parallel Computing Lab, SPCL @ ETH Zurich via YouTube
Overview
Learn about nearest neighbor search algorithms in this 22-minute lecture from ETH Zurich's Scalable Parallel Computing Lab, exploring both exact and approximate solutions. Discover the fundamental concepts behind nearest neighbor search, starting with a comprehensive motivation for its importance. Delve into three key algorithms - KD-Tree for exact search, and HNSW (Hierarchical Navigable Small World) and IVF-PQ (Inverted File System with Product Quantization) for approximate search. Compare the performance and trade-offs between these different approaches, gaining practical insights into when to use each method for efficient similarity search in high-dimensional spaces.
Syllabus
: Introduction
: Motivation
: KD-Tree
: HNSW
: IVF-PQ
: Comparison
: Conclusion
Taught by
Scalable Parallel Computing Lab, SPCL @ ETH Zurich