一种提高神经网络集成系统泛化能力的方法

A Method of Improving Neural Network Ensemble System's Generalization Error

为了充分利用神经网络的原始训练数据,提高神经网络集成系统的泛化能力,提出了一种有效的神经网络集成方法.通过在训练样本上加入一定量的噪声,增大训练样本集,使得不同的个体网络在不同的训练样本上训练,在提高个体网络精度的同时,增加了集成中个体网络的差异度.实验结果表明,该方法能有效的提高神经网络集成系统的泛化能力与计算精度.

In order to fully using original training data and improving generalization error, a method for neural network ensemble was proposed. By adding noises into the input data to augment the original training data set. Individual neural networks can be trained on different training samples. It improves the accuracy of the individual networks while increasing the diversity of the individual networks. Experimental results show that this neural network ensemble method is efficient for improving generalization error.