This article is cited in 43 papers

Markov decision processes with arbitrary real-valued criteria

E. A. Faĭnberg

Moscow

Abstract: We consider discrete time infinite horizon non-stationary Markov decision models with Borel state and action spaces. A criterion is a real-valued function defined on the space of strategic measures. We obtain general results and then use them to study the following criterions and their combinations: the expected total reward criterion, the expected utility criterion, the expected average criterion, the asymptotic reward criterion.

Received: 19.08.1980

 English version: Theory of Probability and its Applications, 1983, 27:3, 486–503

Bibliographic databases:

• 
• 
•