Statement of Interest: Model Reduction Techniques for Computing Approximately Optimal Solutions for Markov Decision Processes

Robert Givan and Thomas Dean

We have recently developed a method for solving implicit (or factored) Markov decision processes (MDPs) with very large state spaces. This method reduces a large implicit MDP to a family of related smaller explicit MDP’s which we call an "approximate MDP". An approximate MDP defines a set of (exact) MDPs by specifying upper and lower bounds the transition probabilities and rewards. We have also developed algorithms that operate on approximate MDPs to estimate value functions and find approximately optimal policies.


This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.