ABSTRACT
This chapter explains dynamic games under uncertainty and a variable number of interacting players. It extends the COmbined fully DIstributed PAyoff-reinforcement and Strategy-Reinforcement Learning (CODIPAS-RL) algorithms to general-sum dynamic games with a variable number of interacting players and random updates. The chapter briefly overviews recent literature on fully distributed learning related to the learning framework. It develops fully distributed schemes for dynamic robust game with random set of interacting players. The chapter shows the main results on almost sure convergence for strategy-learning, payoff learning, combined fully distributed payoff and strategy reinforcement learning. The chapter presents the CODIPAS-RL for a particular class of stochastic games with incomplete information, and a variable number of players. There are different models of stochastic games. Stochastic games in the sense of Shapley, stochastic differential games, and stochastic difference games. A specific class of stochastic games is the class of two-player zero-sum stochastic games.
