Introduction to Reinforcement Learning

In this chapter, we're going to introduce the fundamental concepts of Reinforcement Learning (RL), which is a set of approaches that allows an agent to learn how to behave in an unknown environment thanks to rewards that are provided after each possible action. RL has been studied for decades, but it has matured into a powerful approach in the last few years, with advances making it possible to employ deep learning models together with standard (and often simple) algorithms in order to solve extremely complex problems (such as learning how to play an Atari game perfectly).

In particular, we will discuss:

• The concept of the Markov Decision Process (MDP)
• The concepts of environment, agent, policy, and reward
• The policy iteration algorithm
• The value iteration algorithm
• The TD(0) algorithm

We can now introduce the main concepts that characterize a reinforcement learning scenario, focusing on the features of each...