Re: Please help in Neural Network Implementaion !!


Subject: Re: Please help in Neural Network Implementaion !!
From: Paul Victor Birke (nonlinear@home.com)
Date: Thu Feb 10 2000 - 00:16:20 MET


Myriam Othmane wrote:
>
> Hello,
> I am not a expert in RBF, but I think RBF is more robust to outliers
> (apologize for my english writting). The MLP will not generalize well with
> this data (he goes in saturation) and the average error of the validation
> set will become very instable. With the RBF and the locally (multivariate
> gaussian distribution) , the effect of a 'outlier' (extreme value) will
> vanish more easily.

Yes I believe this also. This is why choice of believable variance for
each centre is important. To identify and cutoff (or dropout) the
outlier identified as not a part of that local population!

> Indeed, I think the outliers will increase the variance (spread, smoothing
> parameter) but they will have less impact on the convergence of the
> algorithm.

True

> I have just experienced this result from very skew distribution (with a lot
> of extremes values), may be it was a artefact of my data.

Problem is when you have an outlier that is really part of the
distribution. So you kind of know what kind of distribution you might
have before hand. This is why straight Gaussians are a bit misleading.
It would be nice to have Weibull distributions to handle inherent skew.

all the best, Manuel

Paul V. Birke, P. Eng. NN Researcher in Guelph ON CANADA

############################################################################
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 - 15:59:43 MET DST