Product quantization (PQ) is a technique that decomposes high-dimensional vector spaces into a Cartesian product of low-dimensional subspaces, with each subspace quantized independently. This approach represents each vector by a compact code, enabling efficient distance estimation while significantly reducing memory usage. [1]
It is commonly used in approximate nearest neighbor search, like the Hierarchical navigable small world data structure.
- ^ "Product Quantization for Nearest Neighbor Search". hal.science. Retrieved 2025-03-21.