Automated model selection and parameter estimation of log-normal mixtures via BYY harmony learning

Abstract

Bayesian Ying-Yang (BYY) harmony learning system is a newly developed framework for statistical learning. Via the BYY harmony leaning on finite mixtures, model selection can be made automatically during parameter learning. In this paper, this automated model selection learning mechanism is extended to logarithmic normal (log-normal) mixtures. Actually, an adaptive gradient BYY harmony learning algorithm is proposed for log-normal mixtures. It is demonstrated by the experiments that the proposed BYY harmony learning algorithm not only automatically determines the number of actual log-normal distributions in the sample dataset, but also leads to a satisfactory estimation of the parameters in the original log-normal mixture. © Springer-Verlag Berlin Heidelberg 2013.