摘要
作者借用线性代数方程组定解讨论中的相关概念 ,类比表述了前向人工神经网络的定解和泛化问题 ,同时说明解决泛化问题仅仅给定网络 VC维是不够的 ,还需要研究样本集规模、向量维度、相关性以及模拟对象的复杂度。文章给出了相应的算例 ,一定程度上讨论了前向神经网络的演进特征。
With the help of some relevant concepts of linear algebra, this paper describes some problems in solving a feed-forward neural network and the evolvement of the neural system and shows that to know the VC dimensions is not enough for the evolvement of the neural system and it is necessary to analyze the scale, dimension, correlativity of the training data set, and the complication of the modeling problem. Two relevant examples are also presented in the paper to illustrate the evolvement characteristics of feed-forward neural networks.
出处
《物探化探计算技术》
CAS
CSCD
2001年第2期144-149,共6页
Computing Techniques For Geophysical and Geochemical Exploration
关键词
前向神经网络
泛化问题
相关性
秩
向量维度
neural networks
evolvement problem
correlativity
rank
dimension of a vector