RE: Degree of Support (DoS) in FuzzyTech Software


Subject: RE: Degree of Support (DoS) in FuzzyTech Software
From: Lim Wee Han (cvelimwh@nus.edu.sg)
Date: Sat Feb 19 2000 - 03:35:07 MET


Dear Prof. Constantin von Altrock,

Thanks for the quick reply in answering my queries. I still not clear about
how does Back Propagation Neural Network (BPNN) applied to fuzzy logic
system. I have the experience in using the BPNN. My questions are as
below:
1) How does the NN weights represented in fuzzy logic? Such as which
weight represent which fuzzy logic parameters.
2) How to control the neurofuzzy module so that I can reproduce the
same result (DoS) when each time I do the training?
3) If the neurofuzzy module cannot be reproduce, which training result
should I used as the final model?
4) In my application, I find the DoS for certain may be very small but
still very significant especially in extreme case. What method can we used
to reduce the number of rules? I only think of removing rules with DoS less
than certain threshold value?
5) How does the interpretation of DoS (rule weight) for different
rules?
6) Can you give list of published paper from Kosko and Zimmermann that
related to fuzzytech neurofuzzy module?
7) I have come across this paper, Zimmermann et al. (1996), "Modeling
the German Stock Index DAX with Neurofuzzy", in Proc. Fourth European
Congress on Intelligent Techniques and Soft Computing (EUFIT96). Does this
paper neurofuzzy similar to fuzzytech?

Thanks.
Regards,
Lim Wee Han

 -----Original Message-----
From: Constantin von Altrock [mailto:constantin@vonaltrock.de]
Sent: Tuesday, February 15, 2000 5:47 AM
To: cvelimwh@nus.edu.sg; Multiple recipients of list
Subject: RE: Degree of Support (DoS) in FuzzyTech Software

Dear Lim Wee Han

The NeuroFuzzy Module of fuzzyTECH implements a modified error
backpropagation algorithm (EBP) similar to neural networks. An EBP applies
the input part of a data vector to be trained to the inputs of the current
system. Then it computes the differences (errors) between the computed
values of the output variables and the output part of the data vector. Using
partial differentiation of the information processing knodes of the fuzzy
logic system (that is membership functions and rule weights (degree of
support - DoS) ), the EBP determines the subcomponent of the fuzzy logic
system that is most likely to reduce the error, if modified. Then, it
modifies the component accordingly.

See the literature of Kosko and Zimmermann for theoretical details and the
literature referenced on www.fuzzytech.com for practical applications.

Constantin.

-----Original Message-----

I am currently a user of FuzzyTech 5.12 Professional software. As you know
that this software allows you to learn the rule base from the numerical
training data using the neural network capability. Each rule will be assign
a value of 0 to 1 which called the degree of support (DoS). When the DoS =
0, the rule is insignificant and DoS=1, the rule is significant. The DoS
also allow to take any value in between 0 to 1 for partial significant rule.

My question is how does the software change the DoS based on the training
data? As it involve the Neural Network, what type of NN implement in this
software?

I would be glad if somebody can let me know about it. It will be better if
you can give the references for details implementation.

############################################################################
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



This archive was generated by hypermail 2b25 : Thu Apr 06 2000 - 15:59:45 MET DST