Customer Reviews

An Introduction to Neural Networks

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

 

 

This is one of the best books I have ever read. It introduces neural networks, with a strong emphasis on biological plausibility. For example, the book compares the visual systems of simple animals with neural network feature extraction. Anderson moves effectively among evolutionary biology, cognitive science, artificial intelligence, and behavioral psychology. His insights are important, clear, and often funny as well. The book gently introduces source code for implementing the various neural networks that he describes.

There's nothing really wrong with this book - it's just not useful for someone wanting to actually program a neural network system.

I read all sorts of stuff about the nervous systems in horseshoe crabs, but I don't find myself able to do anything with neural networks. Therefore, I'm scouring the Internet to find some source code examples or a tutorial of some kind.

If you want to know miscellaneous information about neural networks, go ahead and buy the book. But if you actually want to construct neural networks, buy something else.

I highly recommend this book to anyone looking to learn about neural networks. It starts with a good overview of the biological basis that will be appropriate for the engineering student. The author is witty and appropriately cynical.

As the book states, this is an INTRODUCTION, it is not a reference or practical guide to construction. It is rather informative, specifically in the biological sense, and the author does a good job introducing necessary information before using it, such as a review/introduction to vector and matrix mathematics; however, some external reading my be necessary to understand if you do not already understand some of these basics.

Note: I have only read the first 1/3 of this book so far as my first book on Neural Networks.

In my opinion, the author does not write very clearly as he often provides examples or explanations that require a fair amount of assumptions and/or inferences to understand them clearly. On the other hand, he is to the point with no off-topic text. There are also a fair number of errors (typos) in some mathematical formulas and computer code, usually the usage of i or j where the other should have been used or a missing line of code that is clearly described in the text, but forgotten in implementation (the appendices may be correct, but you must download them from [...] ). If the math doesn't make sense to what is written, keep reading and a later formula is usually correct. He also often skips several steps when deriving formulas without explanation beyond, "if [formula] then it is obvious that [new-formula]" so you may have to stop to think about the math involved.

The author is obviously not an advanced computer programmer. The code fragments are in Pascal, which can be easily translated to C/C++, but I would recommend against using this author's code for any reason other than the learning experience in association with reading the book for several reasons: First, the code is not object oriented, and thus will become more complicated than necessary, and second, because he speaks of how important optimization of the code is due to the large number of computations required, and then he immediately provides a 3 line function/procedure that is to be heavily used but could have been 30% more efficient by re-ordering the math (he did suggest the alternative math, and then went ahead and used the less efficient method). Finally, this code was written over 12 years ago in a language that is rarely used. Surely there are more comprehensive and more efficient libraries of code that would be more understandable in your native (primary) programming language.

Let me finish by saying that I am in fact glad to own this book and recommend it to anyone (College level or above) who does not already, but wants to understand the roots of Neural Networks, the links to biology, and get an introduction to many of the most common types of Neural Networks. Be advised, the required reading level is rather high, but the mathematics (at least in the first third of the book) do not go beyond a little calculus (derivatives, integrals, and some partial derivatives), basic Linear Algebra (basic vector and matrix operations, and eigenvectors/eigenvalues), and a basic understanding of statistics.