Monte Carlo optimization approach for decentralized estimation networks undercommunication constraints

摘要: We consider designing decentralized estimation schemes over bandwidth limited communication links with a particular interest in the tradeoff between accuracy and cost of communications due to, e.g., energy consumption. take two classes in–network processing strategies into account which yield graph representations through modeling sensor platforms as vertices by edges well tractable Bayesian risk that comprises transmissions penalty for errors. This approach captures broad range possibilities “online” observations constraints imposed enables rigorous design setting form constrained optimization problem. Similar structures exhibited solutions to problem has been studied previously context detection. Under reasonable assumptions, can be carried out message passing fashion. adopt this framework estimation, however, corresponding involve integral operators cannot be evaluated exactly general. develop an approximation using Monte Carlo methods obtain particle approximate computational both strategies and their optimization. The proposed procedures operate scalable efficient fashion and, owing non-parametric nature, produce results any distributions provided samples be produced from marginals. In addition, exhibits graceful degradation asymptotically becomes more costly, parameterized Bayesian risk.