Learning of Three-Layered Neural Networks by the Contraction Mapping Principle

-Application to Genome Data Analysis-

Koichi Niijima
Takeshi Shinohara
Yoshihiro Mizoguchi
Shinichi Shimozono

Kyushu Institute of Technology


We describe a method for constructing neural networks using the constraction mapping principle. The learning speed of network weights by this method is higher than by the back propagation method. In numerical simulation, we design a two-layered neural network for classifying whether given data are signal peptides. We also discuss three-layered neural networks.