Presented and published in the proceedings of the
International Conference on Artificial Intelligence (ICAI'00)
Las Vegas, NV, June 2000.

 

Systems of Ensemble Networks Demonstrate Superiority over Individual Cascor Nets

 

R.S. Renner
Department of Computer Science
California State University, Chico
Chico, CA 95929-0410 USA

 

ABSTRACT
Ensembles of predictors and classifiers have been a focus of study in neural networks since the early 1990s. The majority of these studies have focused on linear or otherwise weak classifiers. This study focuses on the combination of constructive classifiers, namely, Cascade-Correlation (Cascor) networks, as an alternative to the creation and implementation of individual traditional Cascor networks. In doing so, one can expect to achieve improved generalization, robustness, stability, and often an actual decrease in overall training time. The positive results presented in this study show a potential for up to 29% generalization improvement and 48% stability of the Cascor networks.

Keywords: ensembles, Neural Network Ensemble Simulator, NNES, Cascade-Correlation, CNNS, neural networks


* Present Address: Dept. of Computer Science
California State University - Chico
Chico, CA 95929-0410
E-mail: renner@ecst.csuchico.edu