TitleA Diffusion-based distributed EM algorithm for density estimation in wireless sensor networks
Publication TypeConference Paper
Year of Publication2013
AuthorsPereira, SSilva, Pagès-Zamora, A, López-Valcarce, R
Conference NameInt. Conf. Acoust., Speech, Signal Process. (ICASSP)
Keywordsdynacs, wsn
Abstract Distributed implementations of the Expectation-Maximization(EM) algorithm reported in the literature have been proposed forapplications to solve specific problems. In general, a primaryrequirement to derive a distributed solution is that thestructure of the centralized version enables the computationinvolving global information in a distributed fashion. Thispaper treats the problem of distributed estimation of Gaussiandensities by means of the EM algorithm in wireless sensornetworks using diffusion strategies, where the informationis gradually diffused across the network for the computationof the global functions. The low-complexity implementationpresented here is based on a two time scale operationfor information averaging and diffusion. The convergence toa fixed point of the centralized solution has been studied andthe appealing results motivates our choice for this model. Numericalexamples provided show that the performance of thedistributed EM is, in practice, equal to that of the centralizedscheme.