Quantisation Models | Awesome Learning to Hash Add your paper to Learning2Hash

Binary Quantisation Models

Quantisation Models play a crucial role in nearest neighbor search by converting real-valued data into binary hashcodes, making it easier and faster to retrieve similar items. Two primary types of quantization have been developed: scalar and vector quantisation. The difference lies in whether the input and output are treated as a single scalar or as a vector. This page focuses on scalar quantisation, a widely-used method in hashing.

How Scalar Quantisation Works:

In scalar quantization, the real-valued projections (resulting from the dot product between a data-point’s feature vector and the normal vectors of hyperplanes partitioning the feature space) are transformed into binary values. Each dot product produces a scalar value, which is then quantised into binary (0/1) through a thresholding process. The sequence of these binary values is concatenated to form a unique hashcode for each data point.

Below is a list of key publications on quantisation models, ordered by their bibliographic key: