Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!mailrus!cornell!anoop
From: anoop@svax.cs.cornell.edu (Anoop Kumar Mangalick)
Newsgroups: comp.ai.neural-nets
Subject: Optimal NN architectures
Keywords: architectures convergence
Message-ID: <35762@cornell.UUCP>
Date: 6 Jan 90 03:22:05 GMT
Sender: nobody@cornell.UUCP
Reply-To: anoop@svax.cs.cornell.edu (Anoop Kumar Mangalick)
Organization: Cornell Univ. CS Dept, Ithaca NY
Lines: 8

Does anybody have any pointers to results, theoretical or otherwise,
concerning rate of convergence of backprop NN's for different NN
architectures?  I am looking for any rules, including rules of thumb,
which researchers use in designing NN architectures.

Thanks in advance.

Anoop Mangalick