摘要
神经网络具有固有的优点和广泛的应用,但因其结构设计至今没有确定可寻的方法,使其发展受到很大的影响.为此,提出了基于正交试验的神经网络设计方法.经实证分析,该方法可从任意网络开始,利用正交试验原理改进网络结构,提高网络性能,直至最终设计出较优(或最优)的神经网络;具有原理简单,计算简便,易于应用等特点.
The development of neural network was impeded because there is not a certain method to design its structure although they have some strong points and applications. A design method of neural network is proposed based on experiment of Orthogonality in this paper. The experiment results show the structure and the performance of neural network can be improved with the principle of orthogonal experiment. This method have such features as simple in principle, easy to calculate and suitable application.
出处
《数学的实践与认识》
CSCD
北大核心
2008年第13期116-122,共7页
Mathematics in Practice and Theory
关键词
正交试验
神经网络设计
例子
orthogonal experiment
design of neural network
example
