ABSTRACT
This chapter focuses on distributed strategic learning for global optima in specific classes of games. It provides two examples of wireless networking problems and illustrates how game dynamics can be used to solve these problems. The first example focuses on frequency selection in wireless environment. The second example examines user-centric network selection. The chapter summarizes interactive trial-and-error learning based on Markov chain adjustment and evaluates the frequency of visits of Markov chain to Pareto optimal solutions. The pure global optima are rest points of the replicator dynamics. The global optimization problem consists of maximizing the probability of successful transmissions of the entire system. The global optimum value can be obtained as an equilibrium payoff, that is, the so-called Price of Stability is one. The chapter presents a convergence to an efficient outcome: global optimum using Bush-Mosteller based COmbined fully DIstributed PAyoff-reinforcement and Strategy-Reinforcement Learning for different action sets.
