Understanding Neural Networks

by John Iovine

Understanding Neural Networks The Experimenter's Guide is an introductory text to artificial neural networks. The book begins with examining biological neurons in the human brain and defining their real world mathematical and electronic equivalent.
Building upon this foundation the book contains hardware and software projects that illustrate neural networks. Hardware projects include a op-amp neuron that tracks a light source, speech recognition system, and machine vision system.
Software projects include a Perceptron program and Back-Propagation networks.
This book provides a comprehensive introduction to neural networks.

Genres: Mathematics

166 pages, Kindle Edition

First published September 1, 1998

Reviews

[Good Reads · Rating 1 out of 5]

The book contains very basic, egregious mistakes. For example, in one place the author writes "Homonyms are words that sound alike. For instance, the words cat, bat, sat and fat sound alike." No! Homonyms are, in short, words that are pronounced exactly the same but have different meanings. Those words listed as examples may rhyme, and those words may sound confusing to speech-recognition systems, but those words are most definitely not homonyms!

So perhaps this glaring mistake can be ignored because this is a book about neural networks and not one about grammar? Perhaps... but if the author can't be bothered to properly research the definitions of basic terminology, would you trust him to properly research more advanced concepts?

If you said "no," you are probably right. As just one example, later in the book the author goes on to claim "Hopfield networks are capable of solving mathematical problems deemed impossible for a computer to solve, like the travelling salesman problem." No! First of all, the travelling salesman problem is an example of an NP-hard problem, which means that it takes a long, long time for computers to solve problems like these, not that it is impossible to do so. Second of all, anything possible with Hopfield networks is of course possible on computers, since Hopfield networks run on computers. The author probably meant "rule-based systems" instead of "computers," but still, sloppy use of language. Finally, that sentence is outright incorrect, because if Hopfield networks are indeed able to offer optimal solutions to the travelling salesman problem in less than exponential time, then P = NP would effectively be proved, and thus far that famous problem is still unsolved. Instead, Hopfield networks are merely capable of offering solutions that are pretty darn good, but not the best. Big difference. Look at how careless the author simply had to be in order to make such ludicrous mistakes in a statement like that. Do you really think you can trust this guy on anything he says?

Not only is the book riddled with mistakes, both big and small, but it hardly comes close to the claim that it is a "comprehensive introduction" to neural networks. The book has only talked about feed-forward networks, and showed examples only of artificial neurons modeled with step functions. Nothing at all about slightly more realistic neuron models like leaky fire-and-integrate, which can be gleaned from just a cursory reading of Wikipedia. Nor was anything mentioned about recurrent neural networks. If I didn't already know they existed before reading this book, I would have no idea to even explore them if I wanted to do further reading on my own.

In short, avoid this book. If you know anything about neural networks, this book won't really teach you anything more than spending a few minutes browsing through some Wikipedia articles on the topic. If you don't know much about neural networks or computer science in general, stay away because this book will corrupt your understanding of it.

[GONZA · Rating 3 out of 5]

I really like this argument, but as usual I jump the algorithms part, so what I understood from this book is the pattern of recognition of voices and images and how are they trying to use them to understand the mind and to build a useful computer that, for example, can distinguish between cancer cells better than a human eye.

Mi piace questo argomento, ma come al solito salto tutta la parte degli algoritmi, quindi quello che ho capito si riassume nel modello di riconoscimento di voci ed immagini che stanno cercando di perfezionare, così da una parte cercando di comprendere meglio il funzionamento del cervello, dall'altro possono assemblare programmi che riescano a distinguere cellule cancerose meglio di un occhio umano.

THANKS TO NETGALLEY AND IMAGES SI INC.

[Michael · Rating 3 out of 5]

quick, light introduction with code and hardware implementations, but with some out -of-date chips.