Fast aggregation of student mixture models

Studies on Mixtures of Student (t-)distributions have demonstrated their ability to conduct clustering tasks with valuable robustness to outliers, compared to their Gaussian mixture counterparts. Concurrently, distributed clustering has motivated much interest in methods for building a partition by consensus of multiple partitions. This paper addresses the latter need by aggregating mixtures of Student distributions. It involves minimizing iteratively an approximate KL divergence between mixtures, which themselves approximate each Student component as a finite Gaussian mixture.