一种多层前馈神经网络的快速修剪算法

Title:

Fast unit pruning algorithm for multilayer feedforward network design

作者:

乔俊飞; 张颖;

北京工业大学电子信息与控制工程学院;

Author(s):

QIAO Jun-fei; ZHANG Ying（College of Electronic and Control Engineering; Beijing University of Technology; Beijing 100022; China）;

关键词:

最优脑外科算法;  神经网络修剪算法;  自组织设计算法

Keywords:

optimal brain surgeon;  neural network pruning;  self-organization design

分类号:

TP183

DOI:

CNKI:SUN:ZNXT.0.2008-02-022

文献标志码:

A

摘要:

针对目前神经网络在应用中难于确定隐层神经元数的问题,提出了一种神经网络结构的快速修剪算法.该算法在最优脑外科算法(OBS)的基础上,通过直接剔除冗余的隐层神经元实现神经网络结构自组织设计.实验结果表明,快速修剪算法与常规的最优脑外科算法相比,具有更简单的网络结构和更快的学习速度.

Abstract:

For it is difficult to determine the numbers of hidden neurons in the application of neural networks,a fast unit pruning algorithm for the structure of neural network was presented in the paper.The algorithm which based on optimal brain surgeon(OBS)eliminated the unneeded hidden neurons directly,in which way carried out the self-organization design on the structure of neural networks.The results of comparative studies with OBS showed that the fast unit pruning algorithm could reduce both neural network comp...