Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!godot.cc.duq.edu!news.duke.edu!news.mathworks.com!newsfeed.internetmci.com!in1.uu.net!news.interpath.net!sas!newshost.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: [Q]Comparison between ART and Kohonen's Net?
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <Dr2Ky8.H3G@unx.sas.com>
Date: Wed, 8 May 1996 04:39:44 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <4mgpb3$2nr@news.eecs.nwu.edu> <Dr1xu2.L3@unx.sas.com> <4moqd1$2ai@news.eecs.nwu.edu>
Organization: SAS Institute Inc.
Lines: 41


In article <4moqd1$2ai@news.eecs.nwu.edu>, hunghl@news.eecs.nwu.edu (Hai-Lung Hung) writes:
|> Warren Sarle (saswss@hotellng.unx.sas.com) wrote:
|> 
|> : In article <4mgpb3$2nr@news.eecs.nwu.edu>, hunghl@news.eecs.nwu.edu (Hai-Lung Hung) writes:
|> : |>   Is there any paper or book talking about the comparison between ART
|> : |> and Kohonen's Net? In other words, how do we select between these two
|> : |> approaches?
|> 
|> : ART1, ART2, ART3, Fuzzy ART, ...?
|> : Plain vanilla Kohonen nets, Kohonen self-organizing maps, or Kohonen
|> : learning vector quantization?
|> : And what do you want to _do_ with the network?
|> : -- 
|> 
|> Ok. Let's say Fuzzy ART and Kohonen self-organizing maps.
|> I know the advantages of Kohonen's net and ART-type networks respectively.
|> But given any set of input data, which one is the better clustering method? 

Those methods have rather different purposes. A SOM attempts to describe
the distribution of the data in terms of a low-dimensional array of
clusters. Fuzzy ART partitions the data more on the basis of the order
in which the cases are presented than on the distribution of the data
(see ftp://ftp.sas.com/pub/neural/fart.doc). So you need to decide what
the purpose of the analysis is before you can decide which method is
more appropriate.

|> Is the distribution or type of input data important? In other words, what
|> is the consideration of selecting Kohonen's nets or ART-type networks?

A SOM is not likely to be very useful (as far as I can tell) unless the
data are concentrated near some low-dimensional subspace. Fuzzy ART, on
the other hand, is not much affected by the distribution (such as the
presence or absence of natural clusters) aside from the shape of the
convex hull.

-- 

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.
