Decision Roll and Horizon Roll Processes in Infinite Horizon Discounted Markov Decision Processes
D. J. White · Management Science · 1996
In this paper we look at some aspects of infinite horizon Markov decision processes in which information regarding parameter values is restricted to a finite time horizon, and in which decisions are based upon the finite horizon data but are recomputed as we move forward in time and gain knowledge of later parametric values. A general framework is given. Bounds on the loss of optimality arising from two planning horizon decision processes are given, and the choice of decision process and planning horizon is examined.