Product Quantization
A vector-compression technique that splits vectors into subspaces and quantizes each subvector with a learned codebook to reduce memory and distance-computation cost.
Information-Retrieval Context
Product quantization partitions a vector into subspaces and represents each subvector by a learned codebook index. The compressed representation reduces memory footprint and can accelerate approximate distance evaluation in large vector collections, especially in IVF-PQ designs.
Quantization Boundary
This technique is distinct from neural-network weight quantization. Here the primary object being compressed is an embedding vector and the associated distance computation, with quantization error traded against recall.