A nonparametric data mapping technique for active initialization of the multilayer perceptron
A new nonparametric feature mapping technique for pattern classification is proposed and compared experimentally with a principal component and Sammon's mapping methods. We use the mapped training-.set vectors for an active weights initialization of the multilauer perception classier in a (wo-variate mapped .space. Simulations have shown a usefulness of the proposed weights initialization method for designing the pereeptrons when we need to obtain highly nonlinear decision boundaries.
Key words.Multilayer perception training initialization data transformation feature mapping principal components Sammon method
Unable to display preview. Download preview PDF.
- Fukunaga, K. (1990). Introduction to Statistical Pattern Recognition. NY: Academic Press.Google Scholar
- Karouia M., T.Denoeux and R.Langelle. (1995). Influence of Weight Initialization on Multi-layer Perceptron Performance. Proc. ICANN'95, October 9–13, 1995, Paris.Vol 1, 33–38Google Scholar
- Palubinskas G. (1996).On Weights Initialization of Back-propagation Networks. Neural Network World, 6(1), 89–100.Google Scholar
- Raudys S. and M. Skurichina (1992). The role of the Number of Training Samples on Weight Initialization of Artificial Neural Net Classifier. Proc. of 1-st Russian & IEEE Conf on Neural Networks, Rostov-na-Donu, Russia, 1992, IEEE Publication.Google Scholar
- Sammon, J.W. (1970). An Optimal Discriminant Plane.-IEEE Trans Comp. C-19, 826–829.Google Scholar