ABSTRACT
In this chapter, the authors study games with noisy payoffs. They focus on specific classes of stochastic games with incomplete information in which the state transitions are action-independent. Based on stochastic approximation techniques, the authors show that the payoff estimations can be approximated by differential equations. A team problem with uncertainty is a robust game with finite strategy spaces where all players have the same uncertain pay-off function. The authors develop a fully distributed learning procedure to learn the expected payoffs. They examine the connection between payoff reinforcement learning(Payoff-RL) and differential equations. Most of the learning schemes developed are based on numerical measurement of the payoffs. The Payoff-RL is an attractive method of learning because of the simplicity of the computational demands per iteration, and also because of this proof of convergence to a global payoff function under stationary strategies.
