Parsimonious reduction of Gaussian mixture models with a variational-Bayes approach, in "Pattern Recognition

This paper 1 proposes a technique for simplifying a given Gaussian mixture model, i.e. reformulating the density in a more parcimonious manner, if possible (less Gaussian components in the mixture). Numerous appli-cations requiring aggregation of models from various sources, or index structures over sets of mixture models for fast access, may benefit from the technique. Varia-tional Bayesian estimation of mixtures is known to be a powerful technique on punctual data. We derive herein a new version of the Variational-Bayes EM algorithm that operates on Gaussian components of a given mix-ture and suppresses redundancy, if any, while preserv-ing structure of the underlying generative process. A main feature of the present scheme is that it merely re-sorts to the parameters of the original mixture, ensur-ing low computational cost. Experimental results are reported on real data. 1