Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!honeysuckle!das-news2.harvard.edu!cam-news-feed3.bbnplanet.com!cam-news-hub1.bbnplanet.com!cpk-news-hub1.bbnplanet.com!news.bbnplanet.com!europa.clark.net!169.207.30.81!newsfeeds.sol.net!newsspool.sol.net!munnari.OZ.AU!news.mel.connect.com.au!news.access.net.au!news.teragen.com.au!cantor.edge.net.au!news.apana.org.au!goanna.cs.rmit.edu.au!news.rmit.EDU.AU!matilda.vut.edu.au!news
From: dunne@yarra.vut.edu.au (Robert Dunne)
Subject: Re: Outlier detection
In-Reply-To: schatter@gilbreth.ecn.purdue.edu's message of 11 Aug 1997
	16:47:02 GMT
Message-ID: <tp7mds43pi.fsf@yarra.vut.edu.au>
Lines: 18
Sender: dunne@yarra.vut.edu.au
Reply-To: dunne@matilda.vut.edu.au
Organization: victoria uni of technology
X-Newsreader: Gnus v5.1
References: <5skr60$c9i@mozo.cc.purdue.edu> <tpafip40ww.fsf@yarra.vut.edu.au>
	<5snfm6$s7a@mozo.cc.purdue.edu>
Date: Tue, 12 Aug 1997 00:58:33 GMT


> In article <5snfm6$s7a@mozo.cc.purdue.edu>
>   schatter@gilbreth.ecn.purdue.edu (Subrata
>   Chatterjee) writes: > Thanks for the info. on your
>   paper. Yes, as Bishop suggests(pg. 208), and also
>   pointed out by Warren sometime back in this
>   newsgroup, one can use the Minkowski error instead
>   of the MSE error to prevent fitting the
>   outliers. However, I am not sure whether the
>   outputs can have a probabilistic interpretation
>   with a median error fit (Minkowski error).  Also,
>   it would be nice to have some techniques to detect
>   the outlier before presenting the dataset to the
>   network.

If we are talking about a classification task -- than I doubt that the
Minkowski error will do much good. This will just limit the "influence
of the residual" whereas it is the "influence of the position" in
feature space that causes the non-robust behavior for a classifier.  


-- 
* Rob Dunne  
* Victoria University of Technology , Footscray Campus      
* Department of Computer and Mathematical Sciences                
* P.O. 14428, MCMC.                   Fax:   +61 3 9688 4050  
* MELBOURNE 8001, AUSTRALIA           Tel:   +61 3 9688 4757   
* <http://matilda.vut.edu.au/~dunne>  Email: dunne@matilda.vut.edu.au

You can measure a programmer's perspective by noting his attitude on
the continuing viability of FORTRAN.
                -- Alan Perlis
