Diffusion based collaborative decision making in noncooperative social network games

We consider noncooperative repeated games in a social network where players form social groups (friendship cliques) of identical interests. Players in each social group collaborate to optimize their payoff function via a diffusion learning based regret-matching procedure. Each player fuses its regrets with those of other members in its social group and uses the fused information to choose its actions. We show that, if all players follow the proposed algorithm, the global behavior converges to the set of correlated equilibria. The collaboration amongst players within social groups results in faster coordination and enables the players to respond in real time to the evolution of the game.