摘要
以多项式函数作为神经元的激活函数,结合矩阵伪逆的思想预先确定网络权值,并利用区间折半搜寻法自动优化隐层神经元数。通过对Hermit函数的仿真,充分显示了综合优化神经网络算法对函数具有较好的逼近。
This paper considerates polynomial function as neurons activation function and combines with the thought of pseudo inverse matrix, network weights are predetermined, and by using interval binary search method ,the hidden layer number of neurons are optimized by itself. Through the simulation of the Hermit function , it can be fully seen that the comprehensive of optimizing neural network algorithm has good approximation.
出处
《微型机与应用》
2012年第5期69-70,74,共3页
Microcomputer & Its Applications
基金
中央高校基本科研(CDJXS11100050)
关键词
函数逼近
多项式神经网络
权值预确定
区间折半搜寻法
function approximation
polynomial neural network
weights pre-set
interval binary search method
