Community Research and Development Information Service - CORDIS


In the research area of artificial intelligence (AI) a branch that is becoming more and more important is reinforcement learning (RL). RL can be defined as learning how to map situations into actions interacting with the environment, so as to maximize a reward. Written by two pioneers in this field, this book aims at supplying the basic RL ideas and algorithms. Even if the main point of view is the AI and engineering perspective, the Sutton-Barto book was designed to be accessible to readers of different disciplines. It turns out that the level of mathematical knowledge required to understand the material is not too deep and requires familiarity only with elementary notions of probability.
The book has been divided into three parts. The Problem (three chapters) is the introductory part devoted to the problem description. Elementary Solution Methods (three chapters) describes the most important elementary solution methods in authors� opinion: dynamic programming (DP), simple Monte Carlo (MC) methods, and temporal-difference (TD) learning. A Unified View (five chapters) concerns a generalization of the previous methods, gives a unified view of RL, and provides some examples of real RL applications.
Each chapter has many examples and exercises. Sections and exercises marked with a star (*) can be skipped during a first reading. At the end of every chapter there is a very interesting section dedicated to bibliographical and historical remarks.

Additional information

Authors: SUTTON R S, University of Alberta, Department of Computing Science, Alberta (CA);BARTO A G, University of Alberta, Department of Computing Science, Alberta (CA)
Bibliographic Reference: Cambridge, MA: MIT Press, 1998, 322 pp., hardcover, Reviewed by O. Barana, EUR: 50
Availability: Available online at:
ISBN: ISBN: 0-262-19398-1
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top