presented and published in the proceedings of the
"A Simulation Tool for Managing Intelligent Ensembles"
R.S. Renner and B.A. Juliano
California State University - Chico
Dept. of Computer Science
Chico, CA
R.C. Lacher
Florida State University
Dept. of Computer Science
Tallahassee, FL
ABSTRACT
The creation of neural network ensembles with existing simulators can be laborious and time consuming. Statistical analysis of these ensembles is even more difficult, as most simulators provide only a reporting of individual networks. The creation, selection, and analysis of ensemble networks is most often performed in an adhoc fashion, with the statistical analysis performed either manually or by some external utility after the network results have been stored. If network ensembles are to be of any practical use, they must be efficiently generated and provide statistical feedback automatically upon creation. The Neural Network Ensemble Simulator (NNES) presented as part of this research is created for just such a purpose. NNES supports the creation and combination of unlimited students into the ensemble, while maintaining a voluminous array of statistical data about the ensemble and its student members.
Keywords: Neural Network Ensemble Simulator, ensembles, constructive networks, Cascade-Correlation, combination methods, experiment management tools.
Thanks to Lutz Prechelt for access to the PROBEN1 data repository,
and Scott Fahlman and Marc White for access to the CMU repository and CNNS.