DeepRead
Cover of Neuro-dynamic programming

About the book

This book covers neuro-dynamic programming, a set of methods for solving large-scale sequential decision problems, mainly Markov decision processes, when exact dynamic programming is infeasible. It spans the mathematical foundations of dynamic programming, simulation-based learning algorithms, and the use of approximation architectures such as neural networks to represent cost functions and policies. The central claim is that exact dynamic programming fails for large state spaces because of the curse of dimensionality, so approximate methods must replace the exact value function. The authors argue that combining dynamic programming with simulation and function approximation can produce practical methods, and they supply convergence analysis for several of them, including temporal-difference learning, Q-learning, and approximate policy iteration. They also present case studies to show that these methods work on hard problems.

Written by Claude from what it knows of this book.

What Hacker News says 75% recommend · 0% critical

0122010201520202010 peak · 1
This bookAverage book in Bayesian and probabilistic modeling

In the commenters’ own words

What Open Library readers say

No Open Library reader has rated it yet.

Open Library’s page for the book

What critics say, via Wikipedia

No Wikipedia article about this book yet, so no critics to quote.

Reddit communities — coming later

Details

First published
1996
ISBN
9781886529106
Editions known
2 at Open Library
Open Library
OL2021798W

Dynamic programmingGeneralMathematical optimizationMathematicsNeural networks (Computer science)

Explore in the mosaic

Read next: its neighbours on the mosaic