矩阵的广义逆的正交反向传播算法  

Orthogonal Back-Propagation Algorithm of Generalized Inverse of a Matrix

作　　者：张邦礼[1] 叶仲泉[1] 曹长修[1] 刘一武[1] 李银国[1] 

机构地区：[1]重庆大学自动化系

出　　处：《重庆大学学报（自然科学版）》1996年第4期103-106,共4页Journal of Chongqing University

基　　金：高等学校博士点基金

摘　　要：利用多层前向神经网络研究了矩阵广义道的计算，但算法采用正交反向传播算法（ＯＢＰＡ）．利用ＯＢＰＡ算法，经过有限次迭代即可得到矩阵广义道的精确解。Calculating generalized inverse of a matrix is a basic algorithm for inverstigating linear equations and optimization problems. In this paper we investigate the algorithm of generalized inversion of a matrix by using Orthogonalized Back-Propagation Algorithm (OBPA). There is no problem for convergence in OBPA. It is shown how OBPA avoids some of the difficulties posed by the Back-Propagation(BP)algorithm.

关 键 词：矩阵 广义逆 神经网络 OBPA算法 

分 类 号：O241.6[理学—计算数学]