Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!cam-news-feed3.bbnplanet.com!cpk-news-feed2.bbnplanet.com!news.bbnplanet.com!cpk-news-hub1.bbnplanet.com!www.nntp.primenet.com!nntp.primenet.com!news.sprintlink.net!news-peer.sprintlink.net!cs.utexas.edu!newshost.convex.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: hidden layers
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <E0rrE7.DIB@unx.sas.com>
Date: Tue, 12 Nov 1996 18:01:19 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <55qmhe$ar1@prometheus.acsu.buffalo.edu> <MPG.ced6e89af307c40989681@news.tuwien.ac.at>
Organization: SAS Institute Inc.
Lines: 33


In article <MPG.ced6e89af307c40989681@news.tuwien.ac.at>, dsvozil@echm10.tuwien.ac.at (Daniel) writes:
|> ...
|> Result of the application of the Kolmogorov theorem for ANN is, that if 
|> you want to approximate with the transformation function T any type of 
|> function, you need only three layers in ANN. So the answer for this 
|> question is, that for function-approximation problem (!!!) you need only 
|> 3-layer ANN.

That's a rather misleading oversimplification. See "How many hidden 
layers should I use?" in the Neural Network FAQ, part 3 of 7: 
Generalization, at ftp://ftp.sas.com/pub/neural/FAQ3.html.

|> 2) # of neurons
|> There exists no mathematical prove leading to knowledge how many hidden 
|> neurons you need for your network. It depends on the problem and on your 
|> experience, how many hidden neurons you will use. When I want to decide 
|> about ' of hidden neurons, I usualy train network with 1, 2, 3, .... 
|> hidden neurons. The output error for training set is ALWAYS decreasing, 
|> but for the given number of iterations the output error is at first 
|> decreasing but then begins increase (overtrainig of the network). So use 
|> the number of hidden neurons, where the output error for testing set was 
|> minimal.

This is also an oversimplification. See "How many hidden 
units should I use?" in the Neural Network FAQ, part 3 of 7: 
Generalization, at ftp://ftp.sas.com/pub/neural/FAQ3.html.

-- 

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.
