Urbashi Mitra, Prof.
Date and Time: Jan 30, 2014 ( 3:00 PM)
Location: Orchard room (3280) at the Wisconsin Institute for Discovery Building
Modern wireless technology enables the vision of future large scale systems such as the SmartGrid, a network of ubiquitous and heterogeneous devices wirelessly connected to the Internet, and wireless health monitoring and health modifying sensor networks over communities and not just individuals. All of these applications necessitate methods that simultaneously consider scale, communication, sensing and control. In this talk, key elements of realizing this vision are examined. We shall focus on novel active control methods for networks described by partially observable Markov decision processes. Such models are very general and can encompass sensing networks, as well as communication networks. Following an innovations approach, a Kalman-like filter is derived to estimate the underlying system state. As a case-study, numerical results are provided for physical activity detection in a heterogeneous wireless body area network. We further examine distributed estimation in large scale networks with time-correlated behavior and explore how modern statistical methods such as compressed sensing can be applied to both the distributed estimation problem as well as the network control problem.