Deep Reinforcement Learning. AlphaGo and Other Technologies

This book is a detailed guide to the latest tools in deep reinforcement learning and their limitations. We will implement and practically test cross-entropy methods and value iteration (Q-learning), as well as policy gradients. Various reinforcement learning (RL) environments are used for experiments, ranging from classical CartPole and GridWorld to Atari emulators and continuous control environments (based on PyBullet and RoboSchool). Many examples are based on unconventional environments where we will develop a model of the environment from scratch. In this book, you will learn the role of RL methods in the context of deep learning, implement complex deep learning models. - Study the foundation of RL: Markov decision processes. - Consider examples of RL methods implementation: cross-entropy method, DQN, A3C, TRPO, PPO, DDPG, D4PG, and others. - Learn how to work with discrete and continuous action spaces in various environments. - See how to develop a system that learns to play Atari games using reinforcement learning. - Create your own environment based on the OpenAI Gym model for training a trading agent. - Implement the AlphaGo Zero method for playing Connect4. - Explore the application of RL in speech processing: learn how to train a dialogue bot on movie quotes.