摘要
在Powell重启动共轭梯度法基础上,利用共轭迭代过程产生的二阶导数信息,构造出当前点的牛顿方向,从而得出一类快速共轭梯度法。用于神经网络逼近非线性函数的学习结果表明,该算法的收敛速度均高于使用相同构造公式的共轭梯度算法。
Based on the restart conjugate gradient method by Powell,the current point's Newton direction has been con-structed according to the information of second derivative in the course of conjugate-gradient calculation,which produces Conjugate-Gradient-and-Newton Hybrid(CGNH)method.The learning result applied in neural network to approach to non-linear function shows that the rate of convergence of CGNH method is better than that of conjugate gradient method using the same constructing equation.
出处
《计算机工程与应用》
CSCD
北大核心
2004年第35期84-86,172,共4页
Computer Engineering and Applications
关键词
共轭梯度与牛顿混杂算法
收敛速度
神经网络
Conjugate-Gradient-and-Newton Hybrid method,the rate of convergence,neural network