Multi-Agent Dynamical Systems with Asymmetric Information and with Elements of Learning
Speaker
Tamer Başar
Swanlund Endowed Chair Emeritus and Center for Advanced Study (CAS) Professor Emeritus of ECE, and Research Professor with CSL and ITI
Coordinated Science Laboratory
University of Illinois Urbana-Champaign
Abstract
Decision making in dynamic uncertain environments with mulGple agents arises in many disciplines and applicaGon domains, including control, communicaGons, distributed opGmizaGon, social networks, and economics. Here a natural framework, and a comprehensive one, for modeling, opGmizaGon, and analysis is the one provided by stochas'c dynamic games(SDGs), which accommodates different soluGon concepts depending on how the interacGons among the agents are modeled, parGcularly whether they are in a cooperaGve mode (with the same objecGve funcGons, as in teams) or in a noncooperaGve mode (with different objecGve funcGons) or a mix of the two, such as teams of agents interacGng noncooperaGvely across different teams (and of course cooperaGvely within each team). What also affects (strategic) interacGons among the agents is the asymmetric nature of the informaGon different agents acquire (and do not share or only parGally share (selecGvely) with others, even within teams). What makes such problems even more challenging in a dynamic environment with networked agents is the dependence of the informaGon available to one agent at some point in Gme on the policies or decisions of other agents who have already acted at earlier instants of Gme. Such decision problems, iniGally studied in a team framework, are known as those with nonclassical information where opGmal policies of team agents must be designed to balance a tradeoff between contribuGon to opGmality of the team objecGve funcGon and signaling through their acGons useful informaGon to other agents in their neighborhood who would be acGng aQer them. Existence of such a tradeoff between signaling and opGmizaGon creates even more challenging issues in SDGs with mis-aligned objecGves among at least a subset of agents, which however can be addressed effecGvely for a specially structured subclass of such games, namely mean-field games. This talk will provide an overview of the landscape above, first for a general class of stochasGc dynamic teams and games, and then for a subclass modeling consensus and dissensus in social networks. The talk will also cover reinforcement learning embedded into policy development when agents do not have precise informaGon on the underlying models as well as issues of robustness of equilibria with respect to the decision horizon.
About Speaker
Tamer Başar received BSEE from Robert College, and MS, MPhil, and PhD from Yale University. Since 1981, he has been with the University of Illinois Urbana-Champaign (UIUC), where he is currently Swanlund Endowed Chair Emeritus and Center for Advanced Study (CAS) Professor Emeritus of ECE, and Research Professor with CSL and ITI. At UIUC, he has served as Director of CAS (2014-2020), Interim Dean of Engineering (2018), and Interim Director of the Beckman Institute (2008-2010). He is a member of the US NAE, Foreign Member of Academia Europaea, Fellow of AAA&S, and Fellow of IEEE, IFAC,
SIAM, AAIA, AIIA, and ACA. He has served as President of IEEE CSS, AACC, and ISDG, receiving also the highest awards of these societies as well as of IFAC. He has also received the IEEE CS Technical Field Award, several honorary doctorates and professorships, and Wilbur Cross Medal from Yale University. He has authored or coauthored over 1000 publications in systems, control, communications, networks, dynamic games, social networks and incentive designs, multi-agent systems and learning, cyber-physical systems, and data-driven distributed optimization, with also current research activities in these areas.