Diibuted componentwise EM algorithm or mixture models in sensor networks

This work considers mixture model estimation in sensor networks in a distributed manner. In the statistical literature, the maximum likelihood (ML) estimate of mixture distributions can be computed via a straightforward application of the expectation and maximization (EM) algorithm. In sensor networks without centralized processing units, data are collected and processed locally. Modifications of standard EM-type algorithms are necessary to accommodate the characteristics of sensor networks. Existing works on the distributed EM algorithm focus mainly on estimation performance and implementation aspects. Here, we address the convergence issue by proposing a distributed EM-like algorithm that updates mixture parameters sequentially. Simulation results show that the proposed approach leads to significant gain in convergence speed and considerable saving in computational time.