摘要
建立多层人工神经网络的最优化求解方法,以多层神经网络总体平均误差为目标函数,以权值和阈值作为设计变量,采用共轭梯度法对网络的权值和阈值进行优化计算,完成神经网络训练的方法。将优化原理应用到多层神经网络权值和阈值的计算问题,实现网络权值和阈值的快速计算,为分析神经网络的合理结构提供了必要条件。
This paper puts forward a kited of optimization arithmetic of multi-layered neural network. A kind of neural network training method is completed through defining network average error as optimum objective function, weights and thresholds as design variables, computing network weights and thresholds quickly by using conjugate gradient algorithm in order to offer necessary conditions for analyzing the feasible structure of neural network.
出处
《信息技术》
2008年第3期100-101,108,共3页
Information Technology
关键词
神经网络
共轭梯度法
权值
阈值
neural network
conjugate gradient algorithm
weights
thresholds
