Reinforcement Learning: Basics, Applications, Challenges, and Opportunities
Machine learning can be classified into three major branches: supervised, unsupervised, and reinforcement learning. Over the past few years, supervised and unsupervised learning have been applied to various engineering problems and shown promising results. Most recently, reinforcement learning has emerged as a powerful tool addressing dynamical systems, attracting much attention in both academia and industry. This tutorial intends to bring up the basic concepts of reinforcement learning, associated applications, and challenges and future directions. Background of reinforcement learning is given first; basic building blocks of reinforcement learning such as Markov decision process, Bellman equations, value iteration and policy iteration are introduced. Value-based, policy based, and model-based methods are then discussed. We move on to multi-agent reinforcement learning, which can address a variety of complex problems. Several applications are examined, including autonomous systems, supply chain management, smart energy systems, and smart city. We will finish with some discussions on the challenges of real-world reinforcement learning and possible research directions.