GENERALIZATION TECHNIQUES IN DEEP REINFORCEMENT LEARNING

Abstract

Reinforcement learning (RL) has achieved massive success in recent years, enabling computers to solve problems that are traditionally thought to be "impossible". An important factor in the success of RL is due to the advances in deep reinforcement learning (DRL). However, despite its success, DRL suffers from several significant issues, among them is the issue of generalization. Many techniques to improve generalization have been developed, particularly in the context of supervised learning. In this thesis, we seek to investigate some of these generalization techniques in the context of DRL to understand when they are useful, examine the details and changes necessary to make them work well, and identify the remaining issues. Through a set of theoretical analyses and empirical evaluations, we show how these techniques can improve the generalization performance of deep reinforcement learning agents and shed light on some of their limitations.