前向神经网络:一个新的非参数回归方法

Feed-forward neural networks:A new approach to nonparametric regression

从统计建模的观点,前向神经网络可以看作是一个新的非参数回归方法.通过模拟例子和实际例子对前向神经网络和局部多项式光滑方法的有限样本行为进行了对比,结果表明前向神经网络稍微优于局部多项式光滑方法.此外,对前向神经网络的优点和存在的问题进行了深入讨论.

From the view of statistical modeling, the feed-forward neural networks can be regarded as a new ap proach to the nonparametric regression. The finite sample performance of the feed-forward neural networks is compared with that of the local polynomial smoothers by using simulated example and real example, the results demonstrate that the feed-forward neural networks performs a little better than the local polynomial smoothers. In addition, the advantages of the feed-forward neural networks and its difficulties in implementation are also dis cussed.