Optimality of Event-Based Policies for Decentralized Estimation over Shared Networks