Reinforcement Learning: An Introduction / 2nd Edition by Sutton & Barto

If reinforcement learning has a canonical text, this is it. Sutton and Barto wrote the first edition when the field was small, and the second edition remains the book every RL course assigns and every RL paper assumes you have read.

What the book covers

• The RL problem: agents, environments, rewards and the exploration-exploitation trade-off
• Multi-armed bandits as the simplest case
• Markov decision processes, value functions and the Bellman equations
• Dynamic programming, Monte Carlo methods, and temporal-difference learning
• n-step bootstrapping, eligibility traces, and planning with learned models
• Function approximation for prediction and control, and policy gradient methods
• Connections to psychology and neuroscience, and case studies including TD-Gammon and AlphaGo

Who it’s for

Graduate students, researchers and engineers working in reinforcement learning. Mathematically rigorous but unusually well written for a text of its type — it explains intuition before formalism, consistently.

Edition details

• Authors: Richard S. Sutton, Andrew G. Barto
• Publisher: MIT Press
• Edition: 2nd Edition
• Format: Hardcover
• ISBN: 9780262039246
• Language: English

Ships free within the United States from Shelfcove.