An annealing approach to BYY harmony learning on Gaussian mixture with automated model selection

摘要

Under the Bayesian Ying-Yang (BYY) harmony learning theory and system, we present an annealing approach to making parameter learning on Gaussian mixture with automated model selection. The annealing EM algorithm and its generalization have been derived and discussed. Moreover, several experiments have demonstrated that they are efficient to determine the number of clusters or Gaussians automatically during learning the parameters.