THE EXPECTED AVERAGE CRITERION FOR NONSTATIONARY MDP
非平稳MDP的期望平均准则

In this paper, we consider the expected average criterion for nonstationary MDP. By probability and martingale method, we prove the existence of #epsilon#(>=0)-optimal Markov policies under weakly ergodic conditions. As a typical example of this paper, we solve the open problem posed by Feinberg and Park again.