The Research of Artificial Neural Network Based on Nonparametric Regression Theory

The artificial neural network (ANN) is one of important branches of Artificial Intelligence. In recent years, it has widely used in many fields so that a lot of attention of researchers is provoked. In fact, from the statistical perspective, ANN can be shown that it is one of nonparametric regression models. In this paper, some statistical methods for ANN are discussed, such as the estimation of weight coefficients and their standard error, prediction estimation and its standard error, over-fitting and bias-variance tradeoff and model selection, etc. At last, we present a simple summary.