Re: The relationship btw Fuzzy & Neural Net?

Panos Nikoulis (nikoulis@mtgps018.NoSubdomain.NoDomain)
Wed, 27 Mar 1996 19:55:26 +0100


In article <4j86e0$mp1@ixnews3.ix.netcom.com>, jdadson@ix.netcom.com(Jive Dadson ) writes:
|> In <4j7pp0$1p8@newsbf02.news.aol.com> amendment9@aol.com (Amendment9)
|> writes:
|> >
|> >Try getting a hold of "Neural Networks and Fuzzy Systems" by Bart
|> >Kosko to start.
|>
|> I have a master's degree in mathematics; I have studied the subjects of
|> "fuzzy" and "neural" for a year now, read and understood several
|> Doctoral dissertations on the subjects, written a backpropogation
|> algorithm using the Davidon-Fletcher-Powell quasi-Newton method, and I
|> still can't make heads or tails of that book. I picked it up
|> again just last night and the only parts I can understand at
|> all are the parts I know about from elsewhere. The paragraphs
|> to me often seem to be disconnected concatenations of
|> subject/verb/direct-object declarations full of apparent and outright
|> contradictions but with no sense of direction or motivation. I can very
|> seldom figure out what the symbols refer to, which ones are scalars,
|> which ones vectors, which ones functions, etc. Perhaps Kosko and I just
|> function on a different wavelength, but I cannot share your
|> recommendation.
|>
|> Jive
|>

I couldn't agree more. Although I have great respect for Bart Kosko, his
writing suffers from an apparent lack of direction and unnecessarily heavy
math which makes his books painful to read. The second volume (I think
it's called "Neural Networks With Engineering Applications") is somewhat
better, but I still wouldn't recommend it for first-time reading.

For a great book on neural nets (introductory or not), try "Introduction
to the Theory of Neural Computation", by Herzog et.al., published by the
Santa Fe Institute.

--Panos