NN & Fuzzy Expert System (was: Re: A simple question)

Thierry M. Marbach (tmarbach@neuristics.com)
Sun, 21 Jun 1998 15:35:12 +0200 (MET DST)

You got me confused here. Let me state my thoughts.

Generally speaking (and excluding hybrid models), a Neural Net models how a
process works (internally), from historical information. A Fuzzy Expert
System models how an expert controls the system (externally); the
information is generally human experience, although it is sometimes
completed by historical data in order to fine tune the parameters of the
model.

So the question of performance really depends on your goals. A NN lacks any
explanatory power, and relies on the quality and diversity of the
historical data. A Fuzzy Expert System depends on how complete the
knowledge of the operator is, and how well we extract that knowledge.

It seems that very complicated systems are better modeled through expertise
because the training of the NN can become both intensive and difficult, and
the data requirements may be hard to meet. But I don't see one technique to
be more general than the other. Even in cases of pure decision making (vs.
control), you could collect data about the outcome of a decision and train
a NN on it. Problem is, you probably would not be able to collect enough
information, if any.

Am I missing something, or misinterpreting what you wrote?

Thierry

BTW, I like Fuzzy Expert Systems better as a general rule because it is
transparent...

WSiler <wsiler@aol.com> wrote in article
<1998053016062700.MAA29267@ladder03.news.aol.com>...
> >>2. what fuzzy logic controlers can do that a Neural Net can't?"
> >>
>
> >>Assuming a sufficiently broad selection of each, theoretically
> >>nothing.
> >
> If the question is restricted to relatively simple fuzzy controllers,
> the above reply may be correct. But if we enlarge the question to include
fuzzy
> expert systems generally, of which fuzzy controllers are a subset, then
what
> the fuzzy expert system can do is to model the thinking processes of an
expert
> in some field; neural nets, being more-or-less model-free, can't do this.
The
> neural net MIGHT be able to achieve about the same outputs for the same
inputs
> as a simple fuzzy expert system, but this is not at all the same thing as
> modelling the expert's thinking process. In short, a NN is a hand-tuned
tool
> for function approximation; a fuzzy expert system is a general AI
technique.
>
> William Siler

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