Consensus Estimation via Belief Propagation

In this paper, a new problem, consensus estimation, is formulated, whose setting is complementary to the well-known CEO problem. In particular, a set of nodes are employed to sense and estimate a common source, and the purpose is to reach the best possible estimate for all nodes, through local processing and information exchange over the network. The belief propagation algorithm is adopted to provide a common information processing and dissemination framework for such a purpose. The discussion is also extended to the application of estimating a Markov random field.