Customer Reviews

Introduction to Neural Networks for Java, 2nd Edition

5 star:		(1)
4 star:		(1)
3 star:		(1)
2 star:		(1)
1 star:		(0)

 

 

I've read several author's works on neural networks, and while valuable, they almost always limit the subject to theory.

This book limits it's scope to practice, which is fine by me.... If you want theory, feel free to look it up on wikipedia or one of the billion books on AI that only cover theory.

This book assumes you've read a little theory and jumps straight into practice: in it, the author walks you from hands-on from creating the basic neural nodes to creating and training simple decision nets, to building applications for predicting stocks moves and playing backgammon.

I've ready about neural nets, but was unsure on how to apply them in practical applications: this book clarified their design and usage. However, I will warn you, it's not an easy read, and requires you to have the code loaded on your computer nearby... this book is about practice, and the author pumps a lot of information out.

Jeff Heaton undertakes the admirable task of writing a book that provides some background in neural network theory overlaid with a layer of practical Java coding how-to. Unfortunately, the book delivers on neither of these intentions particularly well. I have only a very basic background in neural networking (see "Neural Networks: A Comprehensive Foundation - 2nd edition" by Haykin) and was expecting Mr. Heaton to provide some theoretical explanation of why various network architectures are relevant to particular types of problems, how training algorithms differ from each other and why this is important...actually any bit of theory at all to shed some light on what the code was supposed to do. Instead, there's a brief chapter on matrix math, which in my opinion was not especially clear, and an occasional ball and stick diagram and almost no theoretical explanation of where any of it comes from or why it's important. I learned next to nothing about neural networking from this book, and what I may have learned is probably available on wikipedia in the space of a few paragraphs.

I could resign myself to the lack of neural networking explanation if the book instead presented a robust discussion of Java design as applied to neural networking architectures and algorithms. But, alas, this is not to be found either. The Java code is presented with no insight into the author's design decisions and therefore offers little in the way of educational material. Unless you are truly a Java novice, the code in the book will seem obvious and underwhelming. While it's apparent that basic neural networks can be constructed with relatively simple code, the author's failure to provide any explanation of code design or to frame the code within the context of a larger neural networking library perhaps results in the Java coding how-to portion of the book failing to deliver as well.

In short, I read the first 6 chapters of this book and decided not to waste any further time with it. If you want to understand neural networks, you won't find that here. If you want to learn to write Java code to build neural networks, you won't find that here either. You'll find code that the author has already written that you can use, but there won't be much educational value in it. The book truly is more of a user's manual or technical documentation for the author's neural networking Java classes and not much more. Perhaps that is useful if you want something simple you can drop into a project and run with. My suggestion to those who wish to learn and understand how to build neural networks in Java is to learn a little about the networks themselves then hack out some Java code yourself. You'll understand what the code means and be in a much better position to extend that code. And you'll definitely learn something along the way, which, unfortunately, I did not while reading this book.

The usefulness of this book depends on your purpose. If your primary purpose is to acquire a running start solving AI problems using the author's software this is a useful book. If you want to study neural networks or to develop java programming skills, this is probably the wrong book for you. The author gives good introductions to genetic algorithms and simulated annealing as well as their use in training neural networks. There are also valuable insights for the careful reader; e.g., the approach used for the tic tac toe game (chapter 6) has very limited usefulness, because it took "nearly 20 hours to train" the neural network. This raises the question of how long it would take for a more realistic problem. (The author provides the weights, allowing the reader to experiment with the neural network, without the pain of performing the training.) I found the chapters on genetic algorithms and simulated annealing to be the most useful.

The author'a website provides downloads of all his software with liberal licensing.

For self-study, "Elements of Artificial Neural Networks" by Mehrotra et al or an older book, "Fundamentals of Neural Networks" by Fausett, would be an excellent companion to Heaton, if you would like a combination of accessible theory with a working software suite that can be readily modified to solve real problems.

I happened to start reading the first edition of the book and realized quite fast that it was outdated, however after browsing Amazon for a while I decided to go with the same author, and I wasn't left disappointed.

This is a very good book for anyone starting learning Neural Networks. It might not give you everything in detail, but as far as giving a hands on approach to learning NN this is the book to read. If you, like me, happened to get the first edition I would recommend you to upgrade as well. This edition of the book is much more mature.

I would caution anyone that don't know object oriented programming that this book is based that. I bought the C# version of the book as well, and it doesn't seem any different than syntax wise.