Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!rochester!cornellcs!newsstand.cit.cornell.edu!newstand.syr.edu!news.maxwell.syr.edu!news.mathworks.com!newsgate.duke.edu!interpath!news.interpath.net!news.interpath.net!sas!newshost.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: Kohonen: scaling all inputs the same?
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <E4J9zG.ECw@unx.sas.com>
Date: Fri, 24 Jan 1997 22:18:52 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <19970122093057.BKAMP@pi0192.kub.nl> <32E51A99.26B93CEA@phys.canterbury.ac.nz>
Organization: SAS Institute Inc.
Lines: 30


In article <32E51A99.26B93CEA@phys.canterbury.ac.nz>, "Ryurick M. Hristev" <physrmh@phys.canterbury.ac.nz> writes:
|> Kamp B. wrote:
|>  
|> > As I understand, a Kohonen network weight structure towards a specific
|> > output neuron represents some sort of "average" pattern of the cluster
|> > that is represented by that output neuron. Therefore, data has to be
|> > scaled to avoid a wide variance around this cluster mid-value.
|>  
|> > However, should all inputs be scaled to the same interval (for example
|> > [0,1])?
|>  
|> No, that's done just to simplify the search for the winner.

There are much more important considerations involved than simplifying
the search for the winner. The most commonly used methods of
unsupervised learning, including various kinds of vector quantization,
Kohonen networks, Hebbian learning, etc., depend on Euclidean distances
or scalar-product similarity measures. The considerations are therefore
the same as for standardizing inputs in RBF networks--see "Should I
standardize the input variables (column vectors)?" in the the Neural
Network FAQ, part 2 of 7: Learning, at
ftp://ftp.sas.com/pub/neural/FAQ2.html#A_std_in
-- 

Warren S. Sarle       SAS Institute Inc.   The opinions expressed here
saswss@unx.sas.com    SAS Campus Drive     are mine and not necessarily
(919) 677-8000        Cary, NC 27513, USA  those of SAS Institute.
 *** Do not send me unsolicited commercial or political email! ***

