• Contribute
• What's New
• Submit News

8.5

Author: José Principe, Neil Euliano, W. Curt Lefebvre Excerpt: This one-of-a-kind interactive book and CD-ROM will help you develop a better understanding of the behaviour of adaptive systems. Developed as part of a project aimed at innovating the teaching of adaptive systems in science and engineering, it unifies the concepts of neural networks and adaptive filters into a common framework. It begins by explaining the fundamentals of adaptive liner regression and builds on these concepts to explore pattern classification, function approximation, feature extraction, and time-series modelling/prediction. .: Buy at Amazon.com :. .: Buy at Amazon.co.uk :.

Anybody who has read my reviews on NeuroDimension software (NeuroSolutions, Genetic Server) will know how I feel about ND documentation - it is some of the best in the industry. Therefore, I was very much looking forward to reading a book written by some of the ND team. It comes as no surprise, that Neural and Adaptive Systems is an incredibly comprehensive book.

It is important to note that the book is much more of a textbook both in style and in format. The book is divided up into 11 chapters covering neural networking and adaptive systems right from the mathematics behind linear regression, then moving on to the perceptron and other networks. The book covers a variety of architectures: multilayer perceptrons, radial basis functions, Kohonen, In-star Out-star, Hopfield, Recurrent, and time-lagged feedforward networks to name a few. The thing I liked about the book, though, is that it is not divided by architecture. Instead, the book makes a logical progession through the fundamentals, introducing new architectures as necessary. This made the book a lot more fluid, while managing to introduce a lot of information in each chapter.

The layout of the chapters were well done too, with a lot of diagrams helping to visualize different architectures and the various mathematical principles discussed. There were a lot of mathematical digressions (or clarifications), but these were clearly partitioned from the rest of the text making it easy to skip over them. Obviously, being an ND book it was heavily tied into NeuroSolutions (an Evaluation copy of NS is included on the accompanying CD). Each chapter virtually has a NeuroSolutions example for each section (Chapter 3 has 24 examples!). The CD contains all the necessary breadboards required to see the examples at work immediately. At the end of each chapter there is a brief summary, list of the NS Examples used, exercises to do, concept map of the chapter and references used!

The book claims to be "interactive" and NeuroDimensions has gone to great trouble trying to make it as interactive as possible. The entire book comes in standard Help format on the CD, hyperlinked to the NeuroSolutions examples. It is simply a case of finding the example you want, and pressing a button! One problem I did have with the "interactive book" was the format. The help file didn't look anything close to the professional standard the book did. The text fonts and colours were "Win 3.1-style" and some of the figures were badly reproduced. While this didn't affect the usefulness of the book, it affected the readability of it. Chances are though I would read the hard copy of the book and use the interactive help file as a mediator between the text and NeuroSolutions - so the bad formatting doesn't affect much.

The examples themselves are really neat. Most of them demonstrate something simple, but a lot of them show off the true power of NeuroSolutions. One example I really liked was the use of MLPs to predict the S&P 500. Check the screenshot from NeuroSolutions below - the red is the desired output and black is the network output.

Finally, who can read this? Anyone without a decent mathematical background should perhaps look elsewhere for a book that covers such systems from a more superficial level. Neural and Adaptive Systems really digs deep into neural networks by showing how they work and the theory and mathematics behind them. A decent knowledge of calculus and statistics is definitely required. I covered most of the mathematical theory in my last year of high school, but if you can't integrate, differentiate or don't know what a Taylor series is - perhaps look elsewhere. This said, ND in their true style has given the book a huge appendix (3 actually) at the back of the book explaining most of the mathematical concepts talked about.

In conclusion, Neural and Adaptive Systems truly lives up to all ND standards - well written, presented, and formatted. Combine that with the powerful and intuitive NeuroSolutions and a myriad of examples, this book couldn't supply more to the reader. While not for the casual reader, nor the faint-hearted, every last detail is explained giving the reader a great sense of achievement upon completion.