Distributed clustering using distributed mixture of probabilistic PCA

This paper considers the clustering algorithm based on mixture of probabilistic principle component analyzers in distributed environments. The EM procedure of it is first transformed to a summing variant. Following the classic distributed EM framework for mixture of exponential family distributions and utilizing the summing variant, we propose the distributed EM for mixture of probabilistic principle component analyzers. The proposed algorithm avoids transferring all data from distributed nodes to a central node. Experiment verifies the validity and feasibility of the proposed method. For some datasets, the proposed method can even enhance the log-likelihood as well as the clustering performance.