Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!ogicse!qiclab!nosun!tessi!ronl
From: ronl@tessi.UUCP (Ron Lunde)
Newsgroups: comp.ai.neural-nets
Subject: Re: Information capacity of neural networks
Summary: One reference (book)
Keywords: information capacity, references
Message-ID: <683@godzilla.tessi.UUCP>
Date: 27 Apr 91 20:27:33 GMT
References: <1991Apr10.194045.8447@wiener.ino.qc.ca>
Distribution: comp
Organization: Test Systems Strategies, Beaverton, Oregon
Lines: 20

One source for the information capacity of Hopfield networks,
and variations, is:

Introduction to the Theory of Neural Computation,
by John Hertz, Anders Krogh, Richard G. Palmer
Published by Addison-Wesley, 1991

I'm not sure their book will satisfy everyone, since they are
statistical physicists and tend to have very "physical" views
of neural networks, but they have mathematical proofs of things
like capacity limits, which might be more useful as reference 
material than the popular press.

-----------------------------------------------------------------------------
Ron E. Lunde    ronl@tessi  sun.com!nosun!tessi!ronl
Test Systems Strategies, Inc.            
8205 S.W. Creekside Place        | I'm not sure if my opinions reflect
Beaverton, OR 97005 USA          | those of my employer, or not...
Phone: (503) 643-9281   FAX: (503) 646-4954
-----------------------------------------------------------------------------