What is Reinforcement Learning?
Reinforcement learning (RL) is an area of machine learning concerned with how agents ought to take actions in an environment in order to maximize the notion of cumulative reward. Reinforcement learning is one of three basic machine learning paradigms, alongside supervised learning and unsupervised learning.
A Brief History
RL, broadly speaking, has had three threads that have come together over time to give us the current incarnation of the term. These threads include early work in the psychology of animal learning, optimal control problems in dynamic programming from the 1950s, and the formulation of temporal difference algorithms.
The preface and first chapter of Richard Sutton’s book on Reinforcement Learning provides a nice recap of these ideas as well as a good overview of modern RL methods.
Mirroring its diverse history, research in reinfocement learning continues to shed light on aspects of cognitive neuroscience, deep learning, and how one field can inform the other. For example, questions about how a single system can accomodate fast and slow learning continues to be an active area of research bringing to bear what we know about human learning and state-of-the-art RL algorithms.