Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!uunet!ftpbox!mothost!pts-nntp!news
From: dmason@pts.mot.com (Dale Mason)
Subject: Alopex
Message-ID: <D65K7n.1GC@pts.mot.com>
Sender: news@pts.mot.com
Nntp-Posting-Host: sun192
Reply-To: dmason@pts.mot.com
Organization: Motorola Paging Products Group
Date: Tue, 28 Mar 1995 13:30:59 GMT
Lines: 25

I did not read the FAQ, because I thought this would not be there and have tried
looking at some books to no avail, but there is a learning Algorithm that is of
recent exposure called "ALOPEX".  Anyone heard of it?  I am not concerned with its
advantages and disadvantages, but am in desparate need of finding the Algorithm on 
how it works. 

First off, here is what I know :

	All patterns are feed into the Network and their error is taken and 
	summed :
			e[i] = (d[i] - out[i])^2

			E = SUM( e[i] ) for i = 1 to m, where m is 
				the number of patterns


	But that is where I am stuck, I know that down the line, the weights are
	adjusted through a feedback, based on a probability P[j].  Unfortunately
	looking for the total algorithm is becoming a pain in the butt and I would
	appreciate any help from the experts here.

Thanks

  P. Dale Mason

