Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uflorida!gatech!myke
From: myke@gatech.edu (Mike Rynolds)
Newsgroups: comp.ai.neural-nets
Subject: Re: Size limits of BP (Was Re: ART and non-stationary environments)
Keywords: BP
Message-ID: <18615@gatech.edu>
Date: 3 May 89 18:07:47 GMT
References: <18589@gatech.edu> <15313@eecae.UUCP>
Reply-To: myke@gatech.UUCP (Mike Rynolds)
Organization: School of Information and Computer Science, Georgia Tech, Atlanta
Lines: 17

In article <15313@eecae.UUCP> frey@eecae.UUCP (Zachary Frey) writes:
>I am not familiar with ART, but I am familiar with back-propagation from
>the Rummelhart & McClelland PDP volumes, and I don't remember ever
>seeing anything about a size limit to networks implemented with back-
>propagation.  Could you elaborate?
>
Try increasing the number of internal nodes without changing the input/output
you train it on. If you were to simulate more complex input/output, an increased
number of internal nodes would be necessary to learn the greater complexity.
But even without greater complexity you will notice a rapid decrease in
learning rate as a function of the number of internal nodes, and at a certain
point, it stops learning all together.
-- 
Myke Reynolds
School of Information & Computer Science, Georgia Tech, Atlanta GA 30332
uucp:	...!{decvax,hplabs,ncar,purdue,rutgers}!gatech!myke
Internet:	myke@gatech.edu