Distributed model consensus for models of locally biased measurements in wireless sensor networks

In a wireless sensor network, the sensors collect measurements from their local environments and build a model from those measurements in order to draw conclusions. Distributed model consensus allows sensors to make inferences about the global state of the deployment environment, by sharing models among the sensors, rather than raw data. In this paper, we analyze a regression model consensus framework based on graphical models. We compare its performance to a baseline alternative based on gossip averaging. Convergence and accuracy issues arise in the belief propagation used in the graphical model method, when the underlying communication topology contains cycles. Through simulation, we evaluate the performance on random geometric graph network topologies containing cycles.