Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!satisfied.apocalypse.org!news.mathworks.com!uunet!ulowell.uml.edu!umassd.edu!umassd!s_vogt
From: s_vogt@umassd.cis.umassd.edu (Stefan Vogt)
Subject: Q: GA & ANN design
Message-ID: <D770EA.GBL@umassd.edu>
Sender: usenet@umassd.edu (USENET News System)
Organization: University of Massachusetts Dartmouth
X-Newsreader: TIN [version 1.2 PL2]
Date: Mon, 17 Apr 1995 18:51:43 GMT
Lines: 37


Hi,

I've read two papers about artificial neural network design with genetic
algorithms (see below).

My questions related to that are:

 * To find the fitness (qualtity) of an coded ANN (member) I have to
   train a network. Neither training for n epoches nor training until
   a desired error is met, gives a comparable measurment of the
   network fitness.

 * How much time does the GA need? How many ANN do I have train? to
   get good results for a specific problem. Is this practical for
   real-world applications with bigger nets than in the papers?

Thanx for your help!

After all I think the GA approach for ANN gives a nice design method,
if no a priori knowledge is available.

Bye, Steff
 
                                           -----------------------------
                                           "Good news is just life's way
                                            of keeping you off balance."


!#@$%#^$%^*&^%*$%^&(+~$#^@&)*%&%^%#@^*)^#@!$&*(?<|+),&%^$@$(_*_-\|?,<#@^%

Stefan Vogt                            http://www.cis.umassd.edu/~s_vogt/
Cedar Dell 5357                        August-Knabe-Weg 4
UMass Dartmouth                        59494 Soest
North Dartmouth, MA 02747-2300         Germany
+ (508)-990-9190                       +49 (2921)-62244
s_vogt@cis.umassd.edu                  s_vogt@ira.uka.de
