Article ID: | iaor1992274 |
Country: | Switzerland |
Volume: | 32 |
Start Page Number: | 23 |
End Page Number: | 33 |
Publication Date: | Aug 1991 |
Journal: | Annals of Operations Research |
Authors: | Girlich Hans-Joachim, Sokolichin A.A. |
This paper considers Markovian decision processes in discrete time with transition probabilities depending on an unknown parameter which may change step by step. In the case of the convergence of such a parameter sequence, a policy maximizing the average expected reward over an infinite future is looked for. Under continuity conditions, the uniform optimality is a policy based on ‘estimation and control’ for some multichain models is shown.