BYY harmony enforcing regularization for gaussian mixture learning

Abstract

In this paper, a Bayesian Ying-Yang (BYY) harmony enforcing regularization (BYY-HER) algorithm is proposed for Gaussian mixture learning with a sample dataset on both parameter estimation and model selection, i.e., selecting an appropriate number of Gaussians in the mixture, through a regularization process from the BYY harmony learning to the maximum likelihood learning. It has been demonstrated by experiments on synthetical and real sample datasets that our proposed BYY-HER algorithm can not only select the correct number of actual Gaussians in a dataset, but also obtain good parameter estimations for the parameters in the true mixture.