Probabilistic inference over sensor networks for clusters: Extension to multiple states

The sensor network cluster location refers to the problem of dividing a set of sensors into different clusters according to pairwise affinities and selecting a number of sensors to act as the headers to serve other neighboring sensors. Each non-header sensor will be served by the header sensor with the highest affinity. In this manuscript, we take the uncertainty of the affinities into consideration and extend the model to the case of multiple states. A sensor may have different affinities to its neighboring sensors at different states. The detection of such optimal sensor headers is an NP-hard problem and approximate solutions have to be sought if tractability is to be ensured. To find an efficient computational approach, we propose a method based on the recent advances in graphical models and probabilistic inference. In our experimental studies, we have verified the effectiveness of the solution for large-scale sensor networks.