Learning Correlated Equilibria in Noncooperative Games with Cluster Structure

Abstract

We consider learning correlated equilibria in noncooperative repeated games where players form clusters. In each cluster, players observe the action profile of cluster members and receive local payoffs, associated to performing localized tasks within clusters. Players also acquire global payoffs due to global interaction with players outside cluster, however, are oblivious to actions of those players. A novel adaptive learning algorithm is presented which generates trajectories of empirical frequency of joint plays that converge almost surely to the set of correlated ε-equilibria. Thus, sophisticated rational global behavior is achieved by individual player’s simple local behavior.