Optimised quantisation method for approximate nearest neighbour search
We propose optimised group quantisation (OGQ) for approximate nearest neighbour (ANN) search. Specifically, we construct a group of codebooks and select a group of codewords from the codebooks to approximate the original data such that small quantisation error is obtained. We also propose an effective learning algorithm for optimisation. The experiments show that OGQ can significantly outperform several state-of-the-art ANN methods.