基于最小二乘法的多层前向神经网络构造学习算法

Least Square Based Constructive Learning Algorithm for Multilayer Neural Networks

提出了一种基于最小二乘法的多层前向神经网络构造学习算法．神经网络的结构在学习的过程不断变化，通过动态生成节点和学习误差反向传递，利用最小二乘法实现神经网络的快速学习．仿真结果表明，该学习算法具有学习速度快，学习精度高。

This paper presents a new constructive learning algorithm for multilayer neural networks.This new algorithm backpropagates learning error to the dynamic created nodes in the hidden layer of the neural network and utilizes least square algorithm to calculate the weights of those dynamic created nodes .Simulations show that this new algorithm can make the neural network learn the training samples more quickly and accurately while possessing a good ability of generalization.