Bayesian estimation of unknown parameters over networks

We address the problem of sequential parameter estimation over networks using the Bayesian methodology. Each node sequentially acquires independent observations, where all the observations in the network contain signal(s) with unknown parameters. The nodes aim at obtaining accurate estimates of the unknown parameters and to that end, they collaborate with their neighbors. They communicate to the neighbors their latest posterior distributions of the unknown parameters. The nodes fuse the received information by using mixtures with weights proportional to the predictive distributions obtained from the respective node posteriors. Then they update the fused posterior using the next acquired observation, and the process repeats. We demonstrate the performance of the proposed approach with computer simulations and confirm its validity.