Hierarchical reinforcement learning for metrical task systems

摘要

The use of reinforcement learning to implement metrical task systems is limited to smaller scale problems due to the curse of dimensionality inherent in the method. This paper aims to present an algorithm based on decomposition techniques which allows us to apply this approach to realistic control problems. It analyzes aspects associated with the quality of the solution and its limitations, as well as discuss about the relevant theoretical topics of the approach presented.