Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news4.ner.bbnplanet.net!news3.near.net!paperboy.wellfleet.com!news-feed-1.peachnet.edu!gatech!news.mathworks.com!tank.news.pipex.net!pipex!in2.uu.net!usc!crash!apagani.cts.com!user
From: dean_abbott@partech.com (Dean Abbott)
Subject: Re: A ?NEW? kind of neural net,
Organization: PAR Government Systems Corp.
Date: Thu, 7 Sep 1995 20:27:06 GMT
Message-ID: <dean_abbott-0709951327060001@apagani.cts.com>
References: <42hrde$7tn@astfgl.idb.hist.no>
Sender: news@crash.cts.com (news subsystem)
Nntp-Posting-Host: apagani.cts.com
Lines: 49

In article <42hrde$7tn@astfgl.idb.hist.no>, Frank Schmidt
<Frank.Schmidt@stud.idb.hist.no> wrote:
> 
>  Standard n-nets are generally complex in structure and
> extremely simple on the neuron level. When I first began,
> (in Pascal THEN!:) I twisted it around and tried to figure
> out how to make an intelligent program, so "my" neurons are
> vastly more complex than those in standard n-nets. But "my" nets
> are in return much more simple, optimizing the net with greatly
> lesser overhead...So "my" solution is much more concentrated on
> the neurons & optimizing, rather than the net in general.
> 
> 

Polynomial networks take this approach: use more complex nodal elements
(polynomial or trigonometric basis functions rather than linear) and
compose
functions of the basis elements into a network. They are generated much faster
than highly parallelized neural networks, and usually more accurate, and
implement much more quickly (in software and general purpose hardware).
The intelligence in PNs comes from the higher-order terms and
cross-products that are employed in the nodal elements. While they share
some commonalities with GMDH, they are more general than GMDH network, and
have better pruning and stopping criteria (and more flexible nodal
elements).

Check out:

Barron, A.R., and R.L. Barron, "Statistical Learning Networks: A Unifying
View", 1988 Proceedings fo the 20th Symposium on the Interface.

Ward, D.G., "Generalized Networks for Complex Function Modeling",
Proceedings of the 1994 IEEE International Conference on Systems, Man, and
Cybernetics, San Antonio, TX, Oct. 2-5, 1994.

Elder, J.F., IV, & Brown, D.E. (1995), "Induction and Polynomial 
  Networks", Chapter 3 in 'Advances in Control Networks and Large
  Scale Parallel Distributed Processing Models (Vol. II)', ed. M.D.
  Fraser, Norwood, NJ:  Ablex.  [to appear probably this summer]


Dean.

-- 
Dean Abbott                      |
PAR Government Systems Corp.     |
1010 Prospect St., Suite 200     |
La Jolla, CA 92037               | 
dean_abbott@partech.com          |
