A comparison of mixture models for density estimation
P. Moerland · 1999
. Gaussian mixture models (GMMs) are a popular tool for density estimation. However, these models are limited by the fact that they either impose strong constraints on the covariance matrices of the component densities or no constraints at all. This paper presents an experimental comparison of GMMs and the recently introduced mixtures of linear latent variable models. It is shown that the latter models are a more exible alternative for GMMs and often lead to improved results. 1 Introduction Density estimation is an important issue in machine learning with applications to data visualization, modelling of class-conditional densities, and initialization of radial basis function networks. In this paper, we focus on semi-parametric density estimation based on mixture distributions. A wellknown approach is the use of Gaussian mixture models (GMMs, for example [1]). The use of a GMM with full covariance matrices leads to a huge number of parameters for a high-dimensional input space and pr...