An Analysis of Learning Algorithm for Layered Neural Network with Coupled Gradient Descent Method
A new learning method of layered neural network is proposed in this paper. This method uses two networks that have same structures and different initial conditions. In these two networks, neurons of one network updates their weights not only by using their own information and but information of the other network, and vice versa. This method, collaboration and competition between networks, can be efficient for avoiding local minima. We explore efficiency of this method quantitatively through applying it to XOR and 4bit parity check problems.
Teijiro Isokawa Nobuyuki Matsui
Dept. of Computer Engineering, Himeji Institute of Technology,2167 Shosha, Himeji, Hyogo 671-2201, Japan
国际会议
8th International Conference on Neural Information Processing(ICONIP 2001)(第八届国际神经信息处理大会)
上海
英文
789-792
2001-11-14（万方平台首次上网日期，不代表论文的发表时间）