In the past few years, computers have been able to learn to play Atari games, Go, and first-person shooters at a superhuman level. Underlying all these accomplishments is deep reinforcement learning (RL). Unlike traditional supervised learning methods, in which networks are trained using hand-labeled data, the reinforcement learning paradigm utilizes a reward signal provided by the environment itself to train the network.
Arthur Juliani offers a deep dive into reinforcement learning, from the basics using lookup tables and GridWorld all the way to solving complex 3D tasks with deep neural networks. Along the way, Arthur introduces a variety of RL algorithms, including Q-Learning, Policy Gradient, and Actor-Critic, and shows how to extend them using deep neural networks to solve problems with much more complex and varied state and action spaces.
Arthur Juliani is a machine learning engineer at Unity Technologies. A researcher working at the intersection of cognitive neuroscience and deep learning, Arthur is currently working toward a PhD at the University of Oregon.
©2017, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • email@example.com