Parsimonious variational-Bayes mixture aggregation with a Poisson prior

This paper addresses merging of Gaussian mixture models, which answers growing needs in e.g. distributed pattern recognition. We propose a probabilistic model over the parameter set, that extends the weighted bipartite matching problem to our mixture aggregation task. We then derive a variational-Bayes associated estimation algorithm, that ensure low cost and parsimony, as confirmed by experimental results.