摘要
本文在前馈网络的基础上,提出了一种新的网络模型.在该网络中,由于将线性优化技术引入其学习训练过程,从而提高了网络的学习速度,同时,该网络仍保持了神经网络中非线性映射的基本特点.由于这种网络的学习算法是基于正交投影算法,故我们称之为正交投影神经网络(OrthogonalProjectionNeuralNetwork──OPNN).最后,给出了两个利用OPNN网络分别进行异或分类和函数逼近的例子.
In this paper, a new network model, orthogonal projection neural network (OPNN), is proposed. OPNN is still highly non-linear in the parameters, but its learning algorithm is based on solving linear optimization by Modifying Gram-Schemidt orthogonal projection algorithm. Therefore, OPNN has the proper ties of non-linear I/O mapping, fast learning algorithm and global optimization. Computer simulations show the efficiency of OPNN.
出处
《计算机学报》
EI
CSCD
北大核心
1996年第9期673-678,共6页
Chinese Journal of Computers
基金
国家博士后科学基金
关键词
神经网络
正交投影
非线性映射
Neural network, orthogonal projection, MGS algorithm