Multilayer Perceptron Trained with Numerical Gradient.

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.


An application of numerical gradient (NG) to training of MLP networks is presented. Several versions of the algorithm and the influence of various parameters on the training process are dis-cussed. Optimization of network parameters based on global search with numerical gradient is presented. Examples of two-dimensional projection of the error surface are shown and the influence of various numerical gradient parameters on the error surface is presented. The speed and accuracy of this method is compared with the search-based MLP training algorithm.

Joint International Conference on Artificial Neural Networks (ICANN) and International Conference on Neural Information Processing (ICONIP), Istanbul, June 2003, pp. 106-109

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.