摘要
神经网络训练算法以传统的BP算法为基础,不断衍生新的改进算法,如加动量的BP算法,RPORP等。本文把实际问题抽象分为连续型和离散型数学问题,将现有几种训练算法分别应用在这两类问题中,通过对训练结果准确率及性能的对比,总结不同的算法适合应用的领域。
Neural network training algorithm is based on the traditional BP algorithm, and constantly rise to new improved ones, such as an increased momentum BP algorithm, RPORP, etc., In this paper, several anlysis and experiments have been done for soloving the practical problems such as continuous and discrete mathematical problems. The existing types of training algorithms are used in these two issues, the results accuracy and performance comparison between the different algorithms is summed up for different application areas.
出处
《软件》
2011年第10期29-31,34,共4页
Software
