Path: utzoo!attcan!uunet!samsung!uakari.primate.wisc.edu!uflorida!winnie!zach.fit.edu!dfausett
From: dfausett@zach.fit.edu ( Donald W. Fausett)
Newsgroups: comp.ai.neural-nets
Subject: Backprop Training
Keywords: neural networks, backpropagation, training
Message-ID: <1331@winnie.fit.edu>
Date: 13 Aug 90 15:43:46 GMT
Sender: usenet@winnie.fit.edu
Reply-To: dfausett@zach.fit.edu ( Donald W. Fausett)
Organization: Florida Institute of Technology, ACS, Melbourne, FL
Lines: 10


	The reason that bipolar (-1,+1) data is better for training than
binary (0,1) data is that no learning occurs on a connection when its input
signal is zero.  It is easy to see the reason for this.  During the
backpropagation phase, the delta error term for a unit is multiplied by
the input signal to that unit in order to compute the update for the
weight on that connection. If the input signal is zero, then the weight
update is zero => the value of the weight does not change => no learning
occurs.  When using backpropagation, it is always better to convert binary
input patterns to bipolar form before training the network.