Seminars

back Back to all Seminars

EE Seminar: Perturbed Learning Automata in Coordination Games

Start Date: April 17, 2018
End Date: April 17, 2018

By Dr. Georgios Chasparis (Software Competence Center Hagenberg, GmbH, Austria)
 
Recently, multi-agent formulations have been utilized to tackle distributed optimization problems, due to the increased communication and computational complexity of centralized schemes. In such multi-agent formulations, decisions are usually taken in a repeated fashion, where agents select their next actions based on their own prior experience (i.e., performance measurements). In this seminar, I will present a class of reinforcement-based learning (namely, perturbed learning automata) for convergence to efficient outcomes in (multi-agent) coordination games. Prior work in this class of learning dynamics primarily analyzes asymptotic convergence through stochastic approximations, where convergence can be associated with the limit points of an ordinary-differential equation (ODE). However, analyzing global convergence through an ODE-approximation requires the existence of a potential function, which naturally restricts the analysis to a fine class of games. To overcome these limitations, an alternative framework is proposed for analyzing asymptotic convergence that is based upon an explicit characterization of the invariant probability measure of the induced Markov chain (i.e., stochastic stability). We further describe a methodology for computing the invariant probability measure in positive-utility games, together with an illustration in the context of coordination games. In the second part of this seminar, I will briefly present an experimental study of this class of dynamics in the context of resource allocation for massively parallel applications in many-core computing platforms. Comparison is performed with the standard Linux scheduler.
 
Biography: Georgios Chasparis was born in Athens in 1978. He received the mechanical engineering degree from the National Technical University of Athens, Greece, in 2001, and the M.Sc. and Ph.D. degrees from the University of California Los Angeles, CA, in 2004 and 2008, respectively. From 2008 to 2010, he was a Postdoctoral Fellow in the Department of Electrical and Computer Engineering at the Georgia Institute of Technology, GA, and from 2010 to 2012, he was a Postdoctoral Fellow in the Department of Automatic Control at Lund University, Sweden. Since 2012, he has been with the Department of Data Analysis Systems at the Software Competence Center Hagenberg, GmbH, Austria, where he is currently a Research Team Leader in Prognosis, Control and Optimization. His research interests include evolutionary learning in games, distributed control and optimization, and operations research.
 

More Information:

For more info contact: Prof. Jeff Shamma: email: jeff.shamma@kaust.edu.sa
 
Date: Tuesday 17th Apr 2018
Time:01:30 PM - 03:00 PM
Location: Building 3, Level 5, Room 5220