Neurofuzzy classifiers & fuzzyGA (Java applets and some questions)

Vadim Akhmedov (vdm98@chat.ru)
Tue, 2 Nov 1999 00:54:26 +0100 (MET)

Dear friends,

I am currently working on last chapter of my PhD thesis which
concerned with classification problem.
I've obtained good results in rules acquisition by genetic algorithms
and membership function tuning by gradient descent method.
Two java applets corresponding to this problems are available on the WWW
site http://www.chat.ru/~vdm98
I have used following approach for MBFs tuning. Each initial rule is
presented
as neuron having one input x (in 1D case), own parameters (A,B,C "parameters
of gaussian", CLS "presented class") and one output y which equals "1" if
class of
pattern x is CLS end "-1" otherwise. Therefore "correct" patterns have pulls
MBFs and "incorrect" have repels it.
But I also have some troubles and I would be grateful if anyone could
help
me to understand following:

1. During the membership tuning by gradient descent the result is depend
on initial MBFs parameters. If we have primary space partitioning (e.g. by
genetic algorithm) our inital MBFs are belongs to specific regions.
Using gradient descent for each MBF to maximize classified patterns number
some
of MBFs can join and finally we have a few equal rules and more
misclassified
patterns. How can we prevent it and what error measure we should use.

2. As far as I know, neurofuzzy classifiers with 3 layers are usually used.
Third layer is either OR neuron or weighted average as in RBF networks.
First approah is clear for me, the second is not.

3. Does anybody have seen any information concerning neurofuzzy classifiers
using GAs as the preprocessor for "initial" rules extraction ? If not then
I will probably be able to introduce that approach as novelty in my PhD
chapter.

4. Tuning the antecedent of rule for classification problem is makes the
tuning
of consequent not necessary, we can obtain it afterwards. Am I right ?

5. Is it possible to use such method for rules learning as Conjugate
gradients ?

It is very intresting for me any papers available through internet concernin
g
neurofuzzy & fuzzyGA approaches for classification problems.

Sincerely yours,

Vadim A Akhmedov

############################################################################
This message was posted through the fuzzy mailing list.
(1) To subscribe to this mailing list, send a message body of
"SUB FUZZY-MAIL myFirstName mySurname" to listproc@dbai.tuwien.ac.at
(2) To unsubscribe from this mailing list, send a message body of
"UNSUB FUZZY-MAIL" or "UNSUB FUZZY-MAIL yoursubscription@email.address.com"
to listproc@dbai.tuwien.ac.at
(3) To reach the human who maintains the list, send mail to
fuzzy-owner@dbai.tuwien.ac.at
(4) WWW access and other information on Fuzzy Sets and Logic see
http://www.dbai.tuwien.ac.at/ftp/mlowner/fuzzy-mail.info
(5) WWW archive: http://www.dbai.tuwien.ac.at/marchives/fuzzy-mail/index.html