Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!fas-news.harvard.edu!newspump.wustl.edu!news.starnet.net!wupost!howland.reston.ans.net!news.sprintlink.net!crash!mkppp.cts.com!user
From: Dean_Abbott@partech.com (dean abbott)
Subject: Re: NN Vs Stats......
Organization: pgsc
Date: Fri, 20 Jan 1995 01:45:48 GMT
Message-ID: <Dean_Abbott-1901951752120001@mkppp.cts.com>
References: <1995Jan11.145719.1@ulkyvx.louisville.edu> <3fi9ec$jus@maui.cs.ucla.edu> <1732A9117S86.RVANRAAM@bcsc02.gov.bc.ca>
Sender: news@crash.cts.com (news subsystem)
Nntp-Posting-Host: mkppp.cts.com
Lines: 39

In article <1732A9117S86.RVANRAAM@bcsc02.gov.bc.ca>,
RVANRAAM@bcsc02.gov.bc.ca wrote:

> In article <3fi9ec$jus@maui.cs.ucla.edu>
> edwin@maui.cs.ucla.edu (E. Robert Tisdale) writes:
>  
> >
> >ahdeso02@ulkyvx.louisville.edu writes:
> >
> >>I just attended a class and the teacher categorically stated that
> >>all neural nets are statistical model. I personally beg to differ.
> >
> >Your teacher is right.  Training artificial neural networks is simply
> >finding the best estimates of the network connection weights and biases
> >-- the statistics.  There is no such thing as Neural Networks versus
> >Statistics.  It is a false dilemma created by people who understand
> >neither artificial neural networks or statistics very well.
> >
> >Hope this helps, Bob Tisdale.
> A lot of neural networks people use seem to be nothing more
> than a nonlinear regression because you try to find values for
> a number of weights which will minimize some expression
> that represents a complicated function consisting of
> usually exponential functions.
>  
In the very loosely defined field of neural networks, only part of the
research is just relying on nonlinear regression alone (for a fixed model
structure).  In statistics and neural network research many emphasize
model selection as well, that is finding the model architecture (in
backprop MLPs this would be number of neurons in each layer and the number
of layers).  See papers like
Barron, A.R., and R.L. Barron, "Statistical Learning Networks:  A Unifying
View", 20th? Symposium on the Interface, 1988.  (I've forgotten where it
was)

-- 
PAR Government Systems Corp.     |
1010 Prospect St., Suite 200     |
La Jolla, CA 92037               |
