Markov Decision Processes with Discounted Cost: The action elimination procedures

Abstract

Several accelerating procedures are developed for solving discounted Markov decisions processes problems. The most of them are focused on discounted reward by finding lower and upper bounds for the optimal value function. In the same framework, interested authors proposed some tests to eliminate non-optimal action in order to reduce de computational complexity.In this paper, we will present a new suboptimal test for the standard value iteration algorithm to minimize the discounted cost. This criterion has many applications areas like Economics and Management, Financial Modeling, Artificial Intelligence