CAREER:The Symbiosis of Graphical Models and Games

Regents Of The University Of Michigan - Ann Arbor, Ann Arbor MI

Investigators

Abstract

Many natural, social and engineered systems exhibit or facilitate complex behavior. Such behavior often results from the deliberate actions of, and interactions between, a large number of individuals. The need to study behavior in complex systems of network-structured interactions in large populations promotes interest in computational game-theoretic models. Graphical games build on classical models in game theory, as well as compact, structured representations in probabilistic graphical models. The result is a practical and computationally amenable model to handle networked large-population systems. The creation of technology for scientists and policy makers to study and work with large real-world complex systems of (strategic) interactions requires further advances in graphical games and models. The project seeks to fill knowledge gaps by advancing computational aspects of game theory, graphical models and machine learning, and laying the foundation for a systematic two-way knowledge transfer between computational game theory and graphical models. The research program strengthens the connection between graphical models and game theory by casting probabilistic inference problems as equilibrium computation, creating algorithms to learn games from behavioral data, and characterizing equilibrium structure and computation. The educational program includes the infusion of research results into general education, at all levels, via development of new courses and integration into existing ones; and a concerted effort to bridge the Departments of Economics and Computer Science at Stony Brook. Collaborations through the Center for Game Theory in Economics and the International Summer Festival on Game Theory, held annually at Stony Brook, serve as conduits for outreach and dissemination.