Normalized Gaussian Network Based on Variational Bayes Inference and Hierarchical Model Selection

Abstract

This paper presents a model selection method for normalized Gaussian network (NGnet). We introduce a hierarchical prior distribution of the model parameters and the NGnet is trained based on the variational Bayes (VB) inference. The free energy calculated in the VB inference is used as a criterion for the model selection. In order to efficiently search for the optimal model structure, we develop a hierarchical model selection method. The performance of our method is evaluated by using function approximation and nonlinear dynamical system identification problems. Our method achieved better performance than existing methods.