On Bayesian Analysis of Parsimonious Gaussian Mixture Models

The article delves into the Bayesian analysis of parsimonious Gaussian mixture models, with a particular focus on the Mixtures of Factor Analyzers (MFA) model. It discusses the advantages of MFA over the general Gaussian mixture model (GMM) in terms of parameter reduction and computational efficiency. The authors introduce eight variants of the MFA model, each with different assumptions about the covariance structures, and propose a fully Bayesian approach for inferring the number of clusters and covariance structures using Reversible Jump Markov Chain Monte Carlo (RJMCMC). The article also includes simulations and a real-world application to DNA methylation data, demonstrating the effectiveness of the proposed methods. Additionally, the authors discuss the limitations of their approach and compare it to existing methods, highlighting the strengths and potential improvements of their Bayesian framework.