Understanding 99% of Artificial Neural Networks: Introduction & Tricks [Paperback]

Marcelo Bosque (Author)

Book Description

Publication Date: March 13, 2002

There is a deep desire in men, in order to reproduce intelligence and place it in a machine. Neural Networks are an attempt to reproduce the synaptic connections of our brain in a computer. Duplicating the way we use our neurons to think in a machine, it is expected to have a device that could be able to do "intelligent" tasks, the ones reserved just to humans some time ago. Neural Network are a reality now, not a fantasy, and they have been made in order to recognize patterns (a face ,a photograph or a song, are patterns) and forecast trends. I have seen many books about this subject in my life. All of them are hard to read, and tedious to learn, so I decided to make my own one. For beginner readers, I have tried to use a simple language, in order to be understood by anyone who wants to know about nets. An easy to read, practical and concise work. If you are interested in the brain functions and how can we simulate it in a computer, you'll get here a different way to penetrate into their secrets.For advanced readers who want to make their own nets, I have included a methodology for building neural networks and complete sample computer source-code with tricks that will save you a lot of time while designing it.

Editorial Reviews

About the Author

Marcelo Bosque is graduated from University of Buenos Aires, where he teaches. During 1998 he wrote a serial of articles about Neural Networks. The positive effect and good response of the readers encouraged him to perform a deeper approach to this subject. This book is the result of his research.

Product Details

• Paperback: 146 pages
• Publisher: IUniverse (March 13, 2002)
• Language: English
• ISBN-10: 0595219969
• ISBN-13: 978-0595219964
• Product Dimensions: 9.2 x 6.1 x 0.4 inches
• Shipping Weight: 8.2 ounces (View shipping rates and policies)
• Average Customer Review: 2.6 out of 5 stars  See all reviews (28 customer reviews)
• Amazon Best Sellers Rank: #1,980,134 in Books (See Top 100 in Books)

Customer Reviews

Most Helpful Customer Reviews

29 of 29 people found the following review helpful:

2.0 out of 5 stars An editor's nightmare; a reader's disappointment, October 30, 2004

By 

Dr. Robert C.A. Goff, MCSD, MCSE, MCDBA "Coun... (Blacksburg, VA USA) - See all my reviews

This review is from: Understanding 99% of Artificial Neural Networks: Introduction & Tricks (Paperback)

There are two hurdles to overcome with this "book." The first is the challenging English. I can safely say that I could decipher only about 85% of the English, from a linguistic standpoint. Critical statements about important issues regarding neural nets are so scrambled in word order and choice of vocabulary that they are simply not understandable to this multi-lingual reviewer. While this could have been remedied by employing a competent bi-lingual translator to edit the work, the author obviously chose not to engage one. The second hurdle rests in the concepts of neural nets. These are not explained in a useful way.

My hope was that the Visual Basic code might be useful. The author's choice of VB Script (ASP 3), rather than a more strongly typed language leaves the reader in a position of having to guess the data types that are being used (seldom declared at all). Even VB 6 would have been a better choice and made the code easier to understand. Variable names are either cryptic or (I'm guessing here) Portuguese . A conscientious use of descriptive variable names would have produced self-documenting code. My criticism here is that the code itself is obtuse.

The text is actually only 56 pages long, and that littered with large, nearly meaningless diagrams that clarify little, and sophomoric philosophical comments and sweeping generalities. In such a brief work, the loss of focus is exasperating.

The table of contents contains 82 entries for a book of 120 pages (including back matter). The index is useless. For example, there are separate entries for "artificial", "Artificial" and "ARTIFICIAL", each listing identical page references for more than 1/3 of all the pages in the book. In the bibliography, all but one of the entries is prior to 1993, and many are incomplete references, such as an author, year and title, but no publisher or named periodical.

Why should you care about all these shortcomings? I am already familiar with neural net theory, and yet I could not understand much of the text. I am an experienced software developer, and yet I have difficulty following the sample code in the appendix.

SUMMARY: The entirety of this material should have been cleaned up linguistically, then submitted to a magazine as an article. It hardly comprises sufficient content for a book. As it stands, it fails to fulfill the promise of its title. A reader new to neural nets is unlikely to be satisfied with their expenditure of time and money here. Sorry.

P.S. As you look through the other reviews of this book, it's not too difficult to identify the numerous "planted" ones. They subvert the usefulness of the Average Rating. You would be wise to read all the reviews.

32 of 33 people found the following review helpful:

2.0 out of 5 stars The Title of the Book is an Unfulfilled Goal., July 13, 2003

By 

Me "Me" (United States) - See all my reviews

This review is from: Understanding 99% of Artificial Neural Networks: Introduction & Tricks (Paperback)

Most striking about the book is the poor grammar. True, we engineers are not known to be a linguistically elegant lot, but the grammatical errors in this text do serve as obstacles to any clear explanations of the topic. It should have been proofread by a native English speaker. Regarding the technology explained, the book misses the mark because it doesn't simplify the areas in which people truly have difficulty. Specifically, it doesn't clearly describe how error is "backpropagated" in the standard backpropagation neural net. That topic is the core of the subject and the book misses it. The author merely baby talks the easy parts of the subject and gives misleading pseudocode. There are many books that are much better at presenting simple, intuitive explanations of the theory. Lastly, the book assumes that backprop and Kohonen nets are "99% of artificial neural networks". That is an outdated view. Today, there are *many* types of networks (e.g., recurrent backpropagation nets, reinforcement learning nets, competitive learning nets, counterpropagation nets, neural gas nets, growing neural gas nets, etc.) None of these is addressed at all. Frankly, I would have been minimally satisfied with a clear explanation of the backprop and the Kohonen algorithms, but even that was lacking. I give it the second star only because the goal was a worthy one.

10 of 10 people found the following review helpful:

2.0 out of 5 stars Call this a book?, January 4, 2004

By 

D. Shapiro (Incline Village, Nevada United States) - See all my reviews
(REAL NAME)   

Amazon Verified Purchase

This review is from: Understanding 99% of Artificial Neural Networks: Introduction & Tricks (Paperback)

I was disappointed with this book. First off, an English editor was sorely needed for this book. Although you can get though it, if you're going to write a book, come on, get someone who speaks the language to review it. Content wise, its more like a technical article you would find in a magazine. Not a whole lot of content. 50% of the book (which is only just over 100 pages) is vb source code. I've read other beginning books on the subject and this one is far too light even for someone new to the topic of neural networks.