对偶线性规划神经网络及算法步长的选取  

Study of Dual Linear Programming Neural Networks and Selection of Its Learning Rate

作　　者：裴炳南[1] 保铮[1] 

机构地区：[1]西安电子科技大学雷达信号处理国家重点实验室,陕西西安710071

出　　处：《电子学报》2002年第1期110-113,共4页Acta Electronica Sinica

基　　金：国家自然科学基金 (No .69831 0 4 0 )

摘　　要：文章在对偶线性规划框架内研究神经网络的分析性质 ,用特征值方法界定了Hess阵和神经网络学习步长的取值范围 ,给出了用李普希兹常数表示的算法步长公式 .数值仿真结果表明 ,理论分析是正确的 。The research into the properties of neural networks is made in the framework of the dual linear programming theory.The variation range of a Hessian matrix as well as that of learning rate of the networks is confined by means of the eigenvalue.A Lipschith constant based formula for the algorithm′s convergence is given.Simulation is given to illustrate that the theory is correct and the formula given is efficacious.

关 键 词：对偶线性规划 神经网络 学习率 算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]