For those of you who are new to neural networks and have been trying to find a suitable introductory book on the subject without having a good mathematical and/or statistic background should be clear that it can sometimes be a nightmare to find such a boo
k.
When writing a book, readability and comprehension are two aspects the author needs to take into consideration. Fausett does this very well indeed in her book Fundamentals of Neural Networks. As a broad introductory text books, this is without doubt one of the best book you could currently read on the subject of neural networks. It does not include any source code of any kind (this is normally badly written and compiler specific). The algorithms for many different kinds of simple neural nets are pre sented in a clear step by step manner, in plain English. The book is full of clear explanation and information about neural networks research. She does an excellent job of conveying the information to the reader in terms that a person who has very little knowledge about neural networks can understand. Each chapter has descriptive diagram, well presented examples and written text. Diagrams are an integral part of this book, they convey important information in an effective way and the author uses them we ll.
Some books overwhelm you with unnecessary and difficult to understand lines of intimidating mathematical equations and formulae, which is a major deterrent for beginners. In this book the mathematics are introduced in a relatively gentle manner. There ar e no unnecessary complications or diversions from the main theme. The examples that are used to demonstrate the various algorithms are detailed and simple. If you are new to neural nets and you don't want to be swamped by a huge amount of intimidating loo king mathematics, a programming language that you don't know, etc., this is the book for you.
Except for the first two chapters, which should be read in sequence, all the other chapters are independent and can be read separately. Each chapter contains its own list of references and exercises. Almost in every chapter the main exercises are to do wi th character recognition. So, by solving the same problem using different neural networks you can understand the advantages and disadvantages of different neural networks. Furthermore because each character can be represented by a binary vector of zeros and ones the task of representing the data is made very easy and simple. The book is composed of 7 chapters. The first chapter gives a good historical background to the subject and explains in some detail a variety of neural net architectures and the reas on for their need. Chapter 2 illustrates the advantages of some different nets and data representations over the others by solving the type of problems for which some nets fail to find an answer. The importance of representing data in binary and bipolar i s made clear. The examples are simple enough to understand with few difficulties, furthermore, all this is done using simple neural nets. Chapter 3 is about pattern association. As with the other chapters, the examples used are very descriptive. Chapter 4 is about competitive nets, Kohonen self organising maps. The author explains these in some detail. Chapter 5 is dedicated to Adaptive Resonance Theory (ART) Chapter 6 is all about backpropagation (It doesn't cover the variety of backpropagation, but wha t it does cover is very precise and clear). Chapter 7 gives some insight into a variety of different net architectures and mentions some of the advanced issues.
Other good features of the book relate to the consistancy in notation. The author is consistant in use of notation. All the notations used in the book are introduced and explained at the beginning of the book. Last but not the least, whenever the author u ses a notation which hasn't been introduced, she explains it well before using it. Although the book is an introductory one and its aim is to introduce the beginners to the subject, it is also a very useful book for the researchers. It has a lot of citati ons to a great number of papers and work done by well known authors of the field. Anything that she doesn't cover well (justifiably, since neural networks is a big subject on its own) she makes sure to point the user to the right materials in the literatu re.
As is the case with every book, there are always some shortcomings which could be improved. However, there are very few bad things one can say about this book, one of which is backpropagation. Backpropagation is a big subject in neural networks since it i s the main method of training multilayer nets. I think the author could have given more space to backpropagation and explained different variations of it.
In summary, I think this is the best starting point for the outsider and/or beginner, a truly excellent text. A clear aim of this book is to draw attention to the basic facts. It goes step by step and content area by content area to ensure proper implemen taion. As a good primer I would highly recommend it to those people who are new to the field of neural networks and have to start from scratch or those who have some knowledge about the subject but, are far from being clear.
