BP网络的学习能力与推广能力之间满足的不确定关系式

Uncertainty Relation Between Learning Ability and Generalization Ability of BP Neural Network

为了研究BP网络训练过程中出现的过拟合现象,引进表征问题复杂性的复相关系数R,通过类比信息传递过程中的一般测不准关系式,揭示了BP网络过拟合现象出现时的网络学习能力与推广能力之间满足的不确定关系式;通过模拟多种不同类型复杂程度函数的过拟合数值试验,确定出关系式中的过拟合参数的取值范围为1×10-5～5×10-4;给出应用BP网络对给定样本集的训练过程中,为提高推广能力,判断出现过拟合现象的方法。

In order to study the overfitting of BP neural network in the training processes, multicorrelative coefficient R is introdced to manifest the complexity of function. A general uncertainty relation between the learning ability and the generalization ability suited to overfitting of BP neural network was revealed in the modeling of BP neural network by analogy of general uncertainty relation in information transmitting processes. Tests of numerical simulation for multitype of complicated functions were carried out to determine the value distribution (1×10-5～5×10-4)of overfitting parameter in the uncertainty relation. Based on the uncertainty relation, the overfitting in the training processes of gived sample sets using BP neural network can be justed.