Gradient-Newton learning algorithm for feedforward neural netwoks

In this paper we propose a new kind of Gradient-Newton learning algorithm for feedforward neural networks based on two different forms of error function. In this algorithm we take the advantage of gradient algorithm for its fast speed in the first stage of learning. After that we use Newton iteration algorithm in its convergence domain based on another error function. It needs only to compute the first order of difference in our algorithm, so it has lower computational complexity ann has fast learning speed Finally, we give two computing examples.