Search-based training for logical rule extraction by Multilayer Perceptron.

Miroslaw Kordos,
Faculty of Automatic Control, Electronics and Computer Science,
The Silesian University of Technology, Gliwice, Poland.

Wlodzislaw Duch,
School of Computer Engineering, Nanyang Technological University, Singapore,
and Department of Informatics, Nicholas Copernicus University,
Grudziadzka 5, 87-100 Torun, Poland.


Abstract. Search-based non-gradient training techniques are used to train an MLP-like neural network with quantized parameters. The network training is quite fast and the final network function is converted to crisp or fuzzy logical rules using a simple analysis of its weights. Various modifications of the method are presented, each generating a specific form of rules. Depending on the desired information one of the methods can be chosen. Feature selection and data discretization are automatically performed.

Joint International Conference on Artificial Neural Networks (ICANN) and International Conference on Neural Information Processing (ICONIP), Istanbul, June 2003, pp. 86-89

Preprint for comments in PDF, 160 KB.

BACK to the publications of W. Duch.
BACK to the on-line publications of the Department of Informatics, NCU.