Optimization and Estimation in Sensor Networks and other Distributed Systems

Abstract: Large-scale systems with components spread across space presents significant challenges in analysis and design of optimization and estimation algorithms. These difficulties stem primarily from the difficulty of reasoning about dynamic update rules applied in an asynchronous fashion, and with very limited local access to data and signals. While "classical" computing research has answered many of the main questions posed by such systems, very little exists that easily integrates with the body of theory developed by systems, signals, and dynamics engineers; as a consequence, interfaces between these domains are often very rough, if they exist at all. We will discuss a family of techniques aimed at remedying this gap in modeling and formalism. In short, we will present a geometric framework for converting formal descriptions of certain computations (as optimization problems) into automatically generated distributed algorithms with provable dynamic properties. This formalism allows one to cast many computations relevant to sensor networks and estimation in a "centralized" fashion, while obtaining automatically generated, provably convergent distributed algorithms to implement the computation in the network.Bio: Dr. Spanos received his PhD in Control and Dynamical Systems from the California Institute of Technology, where he also recently completed a Post-Doctoral position. His doctoral research focused on dynamic coordination problems in distributed systems, including filtering and optimization tasks in sensor networks. His current research attempts to integrate knowledge and techniques from dynamics, systems theory, networks, and artificial intelligence. In this vein, he has been active in a number of "non-traditional" collaborations, including predictive systems for epidemiological models of breast cancer (with the Harvard School of Public Health), and optimization of wind power farms in Southern Europe (with Siemens).

Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248

Audiences: Everyone Is Invited

Contact: Shane Goodoff