Distributed estimation in wireless sensor networks via variational message passing

In this paper, a variational message passing framework is proposed for Markov random fields. Analogous to the traditional belief propagation algorithm, variational message passing is performed by only exchanging messages between adjacent nodes in a graph and updating local estimations, but with more energy and computation saving achieved. Explicit forms for distributions in the exponential family are derived and applied to a distributed estimation problem in wireless sensor networks. Furthermore, structured variational methods are explored to improve the estimation accuracy, whose performance is elaborated in a Gaussian Markov random field, by both theoretical analysis and simulation results. To our best knowledge, this is the first work to explicitly apply the structured variational approach in wireless sensor networks.