A New Algorithm to Improve the Generalization Ability of Feed-forward Neural Networks

A hybrid learning approach is presented in which genetic algorithms are used to optimize both the network architecture and the regularization coefficient.Comparison is made among this approach,the back-propagation(BP) algorithm with momentum term and the BP algorithm with fixed regularization coefficient.Numerical results demonstrate that the proposed approach is of highly computational accuracy,quickly convergent speed,and high(generalization) capability.