Path: utzoo!news-server.csri.toronto.edu!cs.utexas.edu!sun-barr!lll-winken!uunet!mcsun!unido!laura!anja!siemon
From: siemon@anja.uucp (H. Peter Siemon)
Newsgroups: comp.ai.neural-nets
Subject: TechRep available
Keywords: Genetic Algorithms, Feedforward Nets
Message-ID: <3083@laura.UUCP>
Date: 7 Mar 91 13:08:46 GMT
Sender: news@laura.UUCP
Organization: University of Dortmund, Germany
Lines: 19

The following tech report is available via surface mail:

	K.-U. Hoeffgen, H. Peter Siemon
	Approximation of Functions with Feedfarward Nets

Contents:

	We show how to systematically construct feedforward nets for 
	any L-2 function from Rn to Rm. The number of units and 
	connections is known in advance and depends on the desired
	accuracy of approximation. Learning is done in one shot.

	We then show ways to improve the net with a genetic algorithm.
	As the learning is one shot the procedure is quick and effective.

An extended abstract of the report has been published in First Int.
Workshop on Parallel Problem solving from Nature, Oct 1990, Preprint.
The tech report is available (free of charge) via surface mail, email
your (surface) address to me.