Decision-making in Networked Multi-agent Systems with UncertaintyDate: 2017-04-27 Add to Google Calendar
Location: Holmes Hall 389
Speaker: Dr. Ceyhun Eksin, postdoctoral fellow at Georgia Institute of Technology
Networked multi-agent systems include multiple autonomous decision-makers whose individual actions give rise to collective phenomena. Examples of such decision-makers are robots in a team, smart meters int he electricity grid, or appliances in Internet of Things. The central challenge in these systems is to design decision-making rules that achieve desired system-wide behavior given the limitations of agent sensing and communication. In this talk, I first address this challenge by introducing the framework of Bayesian network games to model repeated local interactions and rational decision-making in settings of incomplete information. Under this framework, I study asymptotic convergence properties of rational behavior in coordination games, and present issues in tractable computation of rational behavior. I leverage this framework to introduce tractable decentralized algorithms with asymptotic convergence guarantees to rational behavior in potential games. I demonstrate the practical applications of the algorithm by implementing it on a team of ground robots solving a target assignment problem. Second, I consider agent decision-making coupled with environmental feedback in the context of epidemics. In particular, I present a stochastic network game model that captures the rational self-interests of individuals during the spread of susceptible-infected-susceptible (SIS) disease. In this model, I show that a little bit of empathy by sick individuals is critical in controlling the outbreaks. I conclude with future research directions in modeling and design of decentralized decision-making in complex network systems.
Ceyhun Eksin is a Postdoctoral Fellow at the Georgia Institute of Technology affiliated with both the School of Electrical & Computer Engineering and the School of Biological Sciences. He received his PhD in electrical and systems engineering from the University of Pennsylvania, and his MA degree in Statistics from Wharton School, both in 2015. He also received a MS degree in industrial engineering from Bogazici University, Instabul, Turkey in 2008 and his BS degree in control engineering from Instabul Technical University, Instabul, Turkey in 2005. His research interests are in the areas of distributed optimization, game theory and control theory. His current research focuses on game theoretic modeling and optimization of multi-agent systems in biological, communication and energy networks.