Enhancing the generalization ability of neural networks by using Gram-Schmidt orthogonalization algorithm
Wentong Wan, Kotaro Hirasawa, Jinglu Hu, Junichi Murata · 2002
The generalization ability of neural networks is an important criterion when determining whether one algorithm is powerful or not. Many new algorithms have been devised to enhance the generalization ability of neural networks. In this paper an algorithm using the Gram-Schmidt orthogonalization algorithm on the outputs of nodes in the hidden layers is proposed with the aim to reduce the interference among the nodes in the hidden layers, which is much more efficient than the regularizers methods. Simulation results confirm the above assertion.