Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!nntp.club.cc.cmu.edu!godot.cc.duq.edu!newsgate.duke.edu!news-feed-1.peachnet.edu!usenet.eel.ufl.edu!bofh.dot!spool.mu.edu!howland.reston.ans.net!swrinde!elroy.jpl.nasa.gov!news.msfc.nasa.gov!newsfeed.internetmci.com!in2.uu.net!news.interpath.net!sas!newshost.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: Q: Shortcut connections
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <Ds8I9r.2op@unx.sas.com>
Date: Thu, 30 May 1996 20:01:03 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <4odkmd$g5l@mirv.unsw.edu.au> <4oe004$dvd@cantaloupe.srv.cs.cmu.edu> <31AB1F34.41C67EA6@wbmt.tudelft.nl>
Organization: SAS Institute Inc.
Lines: 17


In article <31AB1F34.41C67EA6@wbmt.tudelft.nl>, Wim Pasterkamp <W.R.Pasterkamp@wbmt.tudelft.nl> writes:
|> 
|> Would adding linear nodes to the hidden layer (no shortcuts) have
|> the same effect? This may be easier to implement.

Adding linear nodes to the hidden layer would have the same effect as
shortcut connections if the number of linear hidden nodes were equal to
max(number of inputs, number of outputs), but it would give you
redundant weights, requiring more memory and training time than
shortcut connections.

-- 

Warren S. Sarle       SAS Institute Inc.   The opinions expressed here
saswss@unx.sas.com    SAS Campus Drive     are mine and not necessarily
(919) 677-8000        Cary, NC 27513, USA  those of SAS Institute.
