Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!howland.reston.ans.net!news.sprintlink.net!adam.cc.sunysb.edu!psinntp!psinntp!nntp.hk.super.net!news.ust.hk!shunkong
From: shunkong@cs.ust.hk (*-- Michael --*)
Subject: Re: No. of hidden neurons ...
Message-ID: <1995Feb24.055008.24777@uxmail.ust.hk>
Sender: usenet@uxmail.ust.hk (usenet account)
Nntp-Posting-Host: cssu135.cs.ust.hk
Organization: The Hong Kong University of Science and Technology
References: <rrg.1125.000A3357@aber.ac.uk> <D44LrC.IM2@hkuxb.hku.hk> <3ih1ad$q1p@maui.cs.ucla.edu> <abMTvA9IBh107h@perception.co.nz>
Date: Fri, 24 Feb 1995 05:50:08 GMT
Lines: 37

In article <abMTvA9IBh107h@perception.co.nz>,
Simon Dawson <simond@perception.co.nz> wrote:
>
>I seem to remember reading a paper a while back that stated quite 
>emphatically that p patterns could/would be memorised by (p-1) hidden
					     ^^^^^^^^^ *Oops!*

>nodes, and as such, this was the limiting upper factor.. straight
>backprop here.. I could dig it out if anyone is wildly interested..
>
>S
>

	I think you're saying "memorised"! Neural net should never memorize
something, but to "learn" and "generalize"... if you choose the hidden
nodes to "memorize" the patterns, each node will become a "grandmother
cell" and cannot adapt to changes in patters. Any comments?


Michael.

-- 

*************************************************
Shun-Kong, Michael WAI           
                                     ^  ^
email   : shunkong@cs.ust.hk         . .
tel. no.: (852)-2358-8833           ( @  ) 
office  : room 4205                   u

Department of Computer Science                 
Hong Kong University of Sci & Tech               
Clear Water Bay, Hong Kong.                    
*************************************************
	                                    


