The Institut fur Neuroinformatik, the independent research institute based in Bochum, Germany, has a Java program that is said to enable engineers to create and train neural networks using either the popular back-propagation-of-errors learning method or by growing a unique neural architecture specifically adapted to an application. A fully programmed neural-network demonstration can be downloaded from its web site: http://www.neuroinformatik.ruhrunibochum.de/ini/VDM/research/gsn/DmoGNG/GNG.html. Electronic Engineering Times reports most neural-network learning algorithms force the engineer to choose a network topology ahead of time which means the developer must choose between rival approaches to learning before understanding the advantages and disadvantages of each. The Institute focuses on automatically growing a network topology perfect for a particular application.
This article is from the CBROnline archive: some formatting and images may not be present.
CBR Online legacy content.