James Elson

Autonomous Option Discovery in Hierarchical Reinforcement Learning

Project Summary

Reinforcement learning is an area of machine learning that trains an agent to choose from a selection of low-level actions in appropriate environments. Hierarchical reinforcement learning is an extension of this, such that sequential groups of low-level actions form a higher-level action in a similar way to individual muscle fibres in your arm contracting to extend your elbow. The process of option discovery determines which sequences of low-level actions are abstracted to form a high-level action. My research will focus on efficient and dynamic methods of option discovery, that allows agents to perform well in a multitude of domains through learned, high-level actions.

Research Interests

Reinforcement Learning.

Background

BSc Computer Science, University of Bath.

My Dissertation involved the comparison of reinforcement learning algorithms in the domain of Super Mario.

Supervisors

Dr Özgür Şimşek