Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!gatech!news.mathworks.com!uunet!in2.uu.net!news.interpath.net!sas!newshost.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: Fundamental question of the field (do you have the answer?)
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <DupIEI.45v@unx.sas.com>
Date: Wed, 17 Jul 1996 21:30:18 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References:  <4sgl88$dar@clark.zippo.com>
Organization: SAS Institute Inc.
Lines: 27


In article <4sgl88$dar@clark.zippo.com>, braz@ime.usp.br (Rodrigo de Salvo Braz) writes:
|> ...
|> Given a learning set picked out of a bigger population set, I can find
|> lots of functions which fit ...
|> 
|> So one could think using a neural net in a learning set helps nothing,
|> for I could (theoretically) get any of those (very different) possible
|> functions.
|> ...
|> Of course, one could say the neural nets get the simpler function, the
|> one which generalizes best over the non-learned examples. It is what I
|> answered my teacher, but it sounds quite vague, doesn't it?

It sounds optimistic.

|> So, the question is: is there any studies defining which function a
|> neural net learns? Probably it is model-dependent, but that would be
|> very good anyway.

See 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.
