Designing games for distributed optimization with a time varying communication graph

Abstract

The central goal in multi-agent systems is to engineer a decision making architecture where agents make independent decisions in response to local information while ensuring that the emergent global behavior is desirable with respect to a given system level objective. Our previous work identified a systematic methodology for such a task using the framework of state based games. One core advantage of the approach is that it provides a two step process that can be decoupled by utilizing specific classes of games. Exploiting this decomposition could lead to a rich class of distributed learning algorithm. However, a drawback of our previous approach is the dependence on a time-invariant and connected communication graph. These conditions are not practical for a wide variety of multi-agent systems. In this paper we propose a new game theoretical approach for addressing distributed optimization problems that permits relaxations in the structure of the communication graph.