一种投票权值调整的神经网络集成增量学习方法

An Ensemble of Neural Networks Based Incremental Learning Algorithm with Weightsupdated Voting

  • 摘要: 针对神经网络集成增量学习中集成输出投票权值的设定问题,给出了一种投票权值调整的神经网络集成增量学习方法。该方法定义了神经网络集成中子神经网络训练集的类核函数,通过计算待识样本与类核函数之间的核函数距离得到集成输出中子神经网络的投票权值。这种投票权值设定方法可以根据子神经网络分类器对待识样本的分类性能自适应地调整集成输出的投票权值,是一种更加合理的集成输出投票权值设定方法。仿真实验表明,这种投票权值调整的神经网络集成增量学习方法比投票权值固定的方法增量学习性能更优。

     

    Abstract: In this paper,an ensemble of neural networks based incremental learning algorithm with weightsupdated voting is described.The algorithm defines the kernel function of the component neural network training data sets.The voting weights are updated based on the distance between the test pattern and the kernel function.This method can adaptively update the voting weights with the classification performance of the component neural network to the test pattern and it is more rational than the stable voting weights method in the ensemble of neural networks based incremental learning algorithm.Experimental results show that the algorithm with weightsupdated voting is more promising than that with stable weights voting method.

     

/

返回文章
返回