Flexible Decomposition Algorithms for Weakly Coupled Markov Decision Problems

TL;DR

This paper introduces two novel decomposition algorithms for large Markov decision problems, enabling more efficient solutions by dividing problems into smaller, manageable parts with or without inter-part communication.

Contribution

The paper proposes partial and complete decoupling methods for solving large MDPs, including policy caching and inter-part communication strategies, with theoretical guarantees.

Findings

Both algorithms can find optimal or near-optimal policies.

Framework supports efficient knowledge transfer across similar problems.

Algorithms improve scalability of solving large MDPs.

Abstract

This paper presents two new approaches to decomposing and solving large Markov decision problems (MDPs), a partial decoupling method and a complete decoupling method. In these approaches, a large, stochastic decision problem is divided into smaller pieces. The first approach builds a cache of policies for each part of the problem independently, and then combines the pieces in a separate, light-weight step. A second approach also divides the problem into smaller pieces, but information is communicated between the different problem pieces, allowing intelligent decisions to be made about which piece requires the most attention. Both approaches can be used to find optimal policies or approximately optimal policies with provable bounds. These algorithms also provide a framework for the efficient transfer of knowledge across problems that share similar structure.