Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!news.alpha.net!uwm.edu!reuter.cse.ogi.edu!henson!n9389616
From: n9389616@henson.cc.wwu.edu (Kosuke Imamura)
Subject: heterogenious neuron
Message-ID: <1995Mar26.052425.2262@henson.cc.wwu.edu>
Summary: feedforward NN with heterogenious neuron
Keywords: heterogeniou neuron
Organization: Western Washington University
Date: Sun, 26 Mar 1995 05:24:25 GMT
Lines: 13

 I am interested in studying feed forward NN with neuron having its
own transfer function, instead of using the same transfer function
for every neuron.  According to "Design Rules of Multilayer Percetrons
", SPIE Vol.1710, learning speed is dependent on how far the weighted sum
of inputs is from threshold value.

 If this is the case, can't we have neurons with its own transfer function
which can be adjusted after initial training round?

 Does any one know papers or experiment?  Thanks.

 kosuke@dinner.cs.wwu.edu

