We study a Bayesian framework for density modeling with mixture of exponential family distributions. Our contributions: •A variational Bayesian solution for finite mixture models • Show that finite mixture models (with a Bayesian setting) can determine the mixture number automatically • Justify this result with connections to Dirichlet Process mixture models •A fast variational Bayesian solutio...

In this paper, we present a novel competitive EM (CEM) algorithm for "nite mixture models to overcome the two main drawbacks of the EM algorithm: often getting trapped at local maxima and sometimes converging to the boundary of the parameter space. The proposed algorithm is capable of automatically choosing the clustering number and selecting the “split” or “merge” operations e7ciently based on...

A new method is proposed to quantify significance in finite mixture models. The basis for this new methodology is an approach that calculates the p-value for testing a simpler model against a more complicated one in a way that is able to obviate the failure of regularity conditions for likelihood ratio tests. The developed testing procedure allows for pairwise comparison of any two mixture mode...

Mixture models, especially mixtures of Gaussian, have been widely used due to their great flexibility and power. Non-Gaussian clusters can be approximated by several Gaussian components, however, it can not always acquire appropriate results. By cancelling the nonnegative constraint to mixture coefficients and introducing a new concept of “negative components”, we extend the traditional mixture...

ÐThis paper proposes an unsupervised algorithm for learning a finite mixture model from multivariate data. The adjective ªunsupervisedº is justified by two properties of the algorithm: 1) it is capable of selecting the number of components and 2) unlike the standard expectation-maximization (EM) algorithm, it does not require careful initialization. The proposed method also avoids another drawb...