Newsgroups: comp.ai.fuzzy,comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!fas-news.harvard.edu!newspump.wustl.edu!news.ecn.bgu.edu!news.moneng.mei.com!uwm.edu!cs.utexas.edu!swrinde!news.dell.com!tadpole.com!uunet!in1.uu.net!psinntp!megatest!djones
From: djones@Corp.Megatest.COM (Dave Jones)
Subject: Kosko considered fuzzy (Re: [HELP] Fuzzy Neural Networks)
Message-ID: <D5Lx0A.Fz@Corp.Megatest.COM>
Organization: Megatest Corporation
References: <3kbid5$5cv@rowan.coventry.ac.uk>
Date: Fri, 17 Mar 1995 22:55:09 GMT
Lines: 43
Xref: glinda.oz.cs.cmu.edu comp.ai.fuzzy:4236 comp.ai.neural-nets:22802

From article <3kbid5$5cv@rowan.coventry.ac.uk>, by mty018@coventry.ac.uk (J.Dunlop):
> In article <1995Mar16.141534.173@genrad.co.uk>,
> Everard Brown <browne@genrad.co.uk> wrote:
>>Geetings neuroids (and regular gurus)
>>Where can I find more details on FAM (Fuzzy Associative Memory) methods
>>in artificial NNs. I am assuming that this is a 'fuzzy' extension to BAM,
> 
> Koskos book on Fuzzy Logic and Neural Networks should help


I've been trying for weeks to make sense out of Kosko's _Neural Networks
and Fuzzy Systems_. I consider it a major accomplishment if I can
figure out what the domain and ranges of a function are, after scanning
a chapter back and forth for hours. He lends a whole new meaning to
the word "fuzzy". For example, at one point, he says fuzzy sets are points in
a unit hypercube. A while later he says, "Fuzzy associative memories
(FAMS) are transformations. FAMS map fuzzy sets to fuzzy sets." Then in the
very next sentence he says, "They map unit cubes to unit cubes."
Well, which is it? The book is full of stuff like that. At least
in that case it is clear that he is confounding a set and its power set,
so you are given fair warning that you have to figure out which he
really means. In most cases there is no warning, and I never do figure
out what it is he is trying to define.

At one point I was digging along, hopelessly confused, when I saw a
convolution integral that was very familiar to me. It was at the
center of a theorem that I and fellow grad students were asked to
solve independently as a term project in a graduate course on signal
processing theory at the University of Houston in the mid 1970's. It was based
around some work that was done on a Navy grant in the sixties. I can't
recall the mathematician's name right now. But the point is, I could not
have been more familiar with the thing. I had boiled it down to its essence
and even calculated error bounds for the integral's approximations, a
decidedly nontrival calculation. But until I saw that formula, I didn't have
a clue as to what Kosko he was talking about. Even then, his description of
it still made no sense to me. On top of that, he attributed the result to
"Gluck and Parker [1988, 1989]". I know it was known at least twenty-five
years before that.


             Dave


