Talk of Prof. John S. Baras

June 16, 2015

-- Title: Distributed Learning for Collaborative Inference and Decision Making in Multi-Agent Systems

Time: June 16, 2015
Download as iCal:

Prof. John S. Baras
Department of Electrical and Computer Engineering
Institute for Systems Research University of Maryland College Park, USA

 

Tuesday 2015-06-16 16:00
IST-Seminar-Room V2.268 - Pfaffenwaldring 9 - Campus Stuttgart-Vaihingen

 

Abstract

We consider collaborative decision making and control in multi-agent systems. Learning is an important ingredient in such systems. The emphasis is to derive as simple as possible distributed algorithms that work provably very well, while having minimal knowledge of the system and its parameters; thus the need for distributed learning. Agents’ understanding of others’ behaviors is shaped through observing their actions over a long time. In order to maximize their pay-off, they need to learn the others’ behavior and coordinate with them. We consider a behavior learning algorithm for a class of behavior functions and study its effects on the emergence of coordination in the network. The conditions under which the learning algorithm converges are studied. Next we consider multi-agent systems, with each agent picking actions from a finite set and receiving a payoff depending on the actions of all agents. The exact form of the payoffs is unknown and only their values can be measured by the respective agents. We develop a decentralized algorithm that leads to the agents picking welfare optimizing actions utilizing the interactions in the payoffs from the agents’ actions, and if needed very simple bit-valued information exchanges between the agents over a directed communication graph. Conditions that guarantee convergence to welfare minimizing actions w.p. 1 are derived. We next consider the continuous time and continuous state space version of the problem based on ideas from extremum seeking control. We show convergence of the proposed algorithm to an arbitrarily small neighborhood of a local minimizer of the welfare function. Our results show how indirect communications(signaling between the agents via their interactions through the system) and direct communications (direct messages sent between theagents) can complement each other and lead to simple distributed control algorithms with remarkably good performance. Several applications are briefly discussed. We close by describing current and future research directions.

Biographical Information

Diploma in Electrical and Mechanical Engineering from the National Technical University of Athens, Greece, 1970; M.S., Ph.D. in Applied Mathematics from Harvard University 1971, 1973. Since 1973, faculty member in the Electrical and Computer Engineering Department, and in the Applied Mathematics, Statistics and Scientific Computation Program, at the University of Maryland College Park. Since 2000, faculty member in the Fischell Department of Bioengineering. Since 2014, faculty member in the Mechanical Engineering Department. Founding Director of the Institute for Systems Research (ISR), 1985 to 1991. Since 1991, Founding Director of the Maryland Center for Hybrid Networks (HYNET). Since 2013, Guest Professor at the Royal Institute of Technology (KTH), Sweden. IEEE Life Fellow, SIAM Fellow, AAAS Fellow, and a Foreign Member of the Royal Swedish Academy of Engineering Sciences (IVA). Received the 1980 George Axelby Prize from the IEEE Control Systems Society, the 2006 Leonard Abraham Prize from the IEEE Communications Society, the 2014 Tage Erlander Guest Professorship from the Swedish Research Council, and a three year (2014-2017) Senior Hans Fischer Fellowship from the Institute for Advanced Study of the Technical University of Munich, Germany. Professor Baras' research interests include automatic control communication and computing systems and networks, and model-based systems engineering.

To the top of the page