Constructing Sparse Generative Topographic Mapping using Variational Bayes

Abstract

Generative Topographic Mapping (GTM) is a nonlinear latent variable model introduced by Bishop et al. as a data visualization technique. The GTM uses the EM algorithm to obtain the maximum likelihood estimate. However, maximum likelihood estimation is prone to overfitting to training data. In this paper, we focus on a variational Bayesian approach to avoid overfitting, and we propose the GTM algorithm using the variational Bayesian approach.