Reinforcement Learning: A Survey

TL;DR

This paper surveys reinforcement learning from a computer-science perspective, describing how an agent learns behavior through trial-and-error interactions with a dynamic environment and noting its connections to psychology while clarifying terminology. It centers on core issues such as the $exploration-exploitation$ trade-off, foundations via $Markov decision theory$, learning from $delayed reinforcement$, and the construction of empirical models to accelerate learning, as well as the use of $generalization$ and $hierarchy$ and strategies for coping with $hidden state$. The work synthesizes historical foundations with current developments and concludes with a survey of implemented systems and an assessment of the practical utility of contemporary methods. Overall, it aims to be accessible to researchers and to inform the real-world applicability of reinforcement-learning approaches.

Abstract

This paper surveys the field of reinforcement learning from a computer-science perspective. It is written to be accessible to researchers familiar with machine learning. Both the historical basis of the field and a broad selection of current work are summarized. Reinforcement learning is the problem faced by an agent that learns behavior through trial-and-error interactions with a dynamic environment. The work described here has a resemblance to work in psychology, but differs considerably in the details and in the use of the word ``reinforcement.'' The paper discusses central issues of reinforcement learning, including trading off exploration and exploitation, establishing the foundations of the field via Markov decision theory, learning from delayed reinforcement, constructing empirical models to accelerate learning, making use of generalization and hierarchy, and coping with hidden state. It concludes with a survey of some implemented systems and an assessment of the practical utility of current methods for reinforcement learning.