Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!nntp.club.cc.cmu.edu!news.duq.edu!newsfeed.pitt.edu!portc02.blue.aol.com!cpk-news-hub1.bbnplanet.com!news.bbnplanet.com!cam-news-hub1.bbnplanet.com!uunet!in2.uu.net!uucp3.uu.net!sangam!konark!saathi.ncst.ernet.in!sasi
From: Sasikumar M <sasi@saathi.ncst.ernet.in>
Subject: Re: BackProp any ideas
In-Reply-To: <57av36$b71@camel2.mindspring.com>
X-Nntp-Posting-Host: saathi.ncst.ernet.in
Content-Type: TEXT/PLAIN; charset=US-ASCII
Message-ID: <Pine.ULT.3.94.961128141144.21209A-100000@saathi.ncst.ernet.in>
Sender: news@konark.ncst.ernet.in (News Administration)
Organization: National Centre for Software Technology, Bombay, India
References: <57av36$b71@camel2.mindspring.com>
Mime-Version: 1.0
Date: Thu, 28 Nov 1996 08:45:14 GMT
Lines: 24


On Mon, 25 Nov 1996, Deric Cheng wrote:

> Anybody know where I can get some documentation on back prop.  I'm
> stuck with this correction term that my book told me.
> 
> Whj(t+1) = Whj(t) + nDhj*Xi
> 

There is some mess-up. You are updating the weight between two
neurons, so w must be identified by three items (from-neuron,
to-neuron, which-layer). For example, Rich and Knight, for
a 1-hidden layer net, uses W1ij and W2ij. 1 and 2 indicates the
layers.

The formula there are:

  w1ij(t+1) = w1ij(t) + n. D1j. xi
  w2ij(t+1) = w2ij(t) + n. D2j. hi

I hope this clarifies.

   - Sasikumar

