摘要
神经网络算法是一种非常经典的分类算法,然而神经网络的一个不足之处就是容易陷入过拟合。针对这种不足,正则化神经网路算法与提前终止迭代算法被提了出来。为了进一步研究这两种算法性能的差异,本文通过20个UCI标准数据集上对着这两种方法进行了性能测试。实验显示在分类准确率上正则化神经网路算法要更优秀一些,但是在分类速度上提前终止迭代算法更占优势。
The neural network algorithm is a very classic classification algorithm. However, over-fitting is easy to arisen for neural network algorithm. For this shortfall, regularized neural network algorithm and early termination of the iterative method was proposed. In order to further study the differences of performance between these two algorithms, in this paper, we use 20 UCI standard data sets to test the performance of the two methods. The experiments show that the regularization neural network algorithm exhibits a superiority over the early stopping iteration algorithm at classification accuracy, but the early stopping iteration algorithm is much better at the classification speed.
出处
《科技通报》
北大核心
2013年第10期112-114,共3页
Bulletin of Science and Technology
关键词
神经网络
分类算法
过拟合
正则化
neural network
classification algorithm
over-fitting
regularized
