Sign inSign in

AI & ML · Book

Reinforcement Learning: An Introduction

Richard S. Sutton, Andrew G. Barto

Sutton and Barto treat reinforcement learning as an agent maximizing reward in an uncertain environment, centre the book on core online learning algorithms, and set heavier mathematics in shaded boxes.

Richard S. Sutton, Andrew G. BartoCourse text for COMP0089MIT Press2018552 pagesnamed in 7 lectures

Watch the course it teaches · COMP0089Find a copy

Opens a bookshop outside Lectern.

Read alongside

Named in lectures

1:04:28

Markov Processes, Reward Processes & Bellman Equations · David Silver · DeepMind COMP0089

Sutton and Barto, cited for updated notation

36:38

Planning by Dynamic Programming — Policy & Value Iteration · David Silver · DeepMind COMP0089

Sutton and Barto, cited for Jack's Car Rental example

39:54

Model-Free Prediction — Monte Carlo, TD(0) & TD(λ) · David Silver · DeepMind COMP0089

Sutton and Barto, cited for the Driving Home example

4 more moments

Concepts

Good fit if

Turing Laureate Richard Sutton and Andrew Barto's standard textbook assigned in both David Silver's DeepMind course and Sergey Levine's Berkeley CS285.

About the book

Part I stays within the tabular case, covering methods such as UCB, Expected Sarsa and Double Learning. Part II extends these ideas to function approximation, neural networks, the Fourier basis, off-policy learning and policy-gradient methods. Part III relates the field to psychology and neuroscience, works through case studies including AlphaGo, Atari game playing and IBM Watson's wagering, and closes on societal impacts.

Authors

Readers

No reader has written about it yet.

Write the first review

Where next

15 connections