Soft Vector Quantization with Inverse Power-Function Distributions for Machine Learning Applications

摘要

Given a codebook of centroids~1; i.e. set of centers of classes/clusters c_1 ∈ ~sR~n; 1 ≤i≤L, vector quantization (VQ) is a fundamental signal processing operation that seeks to attribute a given point q ≤ ~sR~n to one of those centroids: c_(io) according to some optimization criterion (Duda and Hart 2000).