Re: Fuzzy Pattern Recognition

Rainer Holve (holve@forwiss.uni-erlangen.de)
Sun, 7 Jun 1998 01:54:03 +0200 (MET DST)

WSiler wrote:
>
> You wrote:
>
> > Actually, the number of possible rules grows exponentially with >the number
> of inputs, i.e. n inputs with m fuzzy sets -> m^n rules.
>
> I had posted a reference to the Combs method, by William Combs of Boeing in
> Seattle, which enables the number of rules to increase linearly with the number
> of variables rather than exponentially. Unfortunately, the Combs method has
> received the same treatment among fuzzy persons that fuzzy math has received
> among non-fuzzy persons; i.e., nearly total inattention.

I guess you relate to

IEEE Trans. Fuzzy Sys 6:1, pp 1-11, 1998
Combinatorial Rule Explosion Eliminated by a Fuzzy Rule Configuration
W.E. Combs and J.E. Andrews

Maybe this is the right forum to discuss this approach...

> While the Combs method seems quite counter-intuitive, I tried it out, and it
> works!

Perhaps you could describe with what kind of problem you tried it.

I have a problem with this paper, because it somehow suggests, that
everything that can be
done with the traditional rulebase style can also be done with the
proposed URM though
with far fewer rules. Well, the traditional fuzzy systems can be viewed
as universal
function approximators and I doubt that this holds for the URM as well.

Maybe I just didn't get the point of that approach but could someone
point out a
URM for the 2-d XOR-problem i.e. (in traditional rules)

if X is LOW and Y is LOW then Z is LOW
if X is LOW and Y is HIGH then Z is HIGH
if X is HIGH and Y is LOW then Z is HIGH
if X is HIGH and Y is HIGH then Z is LOW

I really don't see how this could be modeled with a disjunction of
any number of one-dimensional rules.

THere is this strange constraint 2 in the paper which says

"Everey input and output universal set relation must be monotonic."
It's explained in the Appendix as follows:
If Input P has the fuzzy sets P1<= P2<=P3<= ... <=Pn (Constraint 1) and
the
output R has the fuzzy sets R1<=R2<=R3<= ... <= Rn (also constraint 1)
then if
"If R2 is greater than R1 and P1 relates to to R1 and P2 relates to R2,
then P3 must relate to an element of R >= R2, etc."

This sounds to me as if the method is only able to model globally
monotonic functions,
i.e. wherever you are in the problem space, if you look parallel to of
one of the
input axes, the function your URM models is always increasing
(or decreasing if you turn around :)

If this is the case, I will happily agree that the URM is the best
suited way to
model such functions (but only those). One might ask, how many real
world
problems there are that fit into this form of monotony but I think that
the proposed approach
is clearly unsuited for e.g. pattern recognition where you want to learm
some fuzzy rules from data
to capture an unknown function you know nothing about.

Regards
Rainer

-- 
Rainer Holve.......................................FORWISS            
Am Weichselgarten 7..............D-91058 Erlangen, Germany 
Email: ...................................holve@forwiss.de
URL:........................http://www.forwiss.de/~rrholve
Telefon: +49-9131-691-257............Fax: +49-9131-691-185

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