The ∏ Method for Estimating Multivariate Functions from Noisy Data (Q3980939)

The author considers the problem of nonparametric estimation of smooth regression functions. It is a well-known fact that in the case of one- dimensional regressors many satisfactory methods are available. The situation in two or more dimensions is less satisfactory because the complexity of multivariate surfaces can grow exponentially with dimension, but usually the data do not. To disarm the curse of dimensionality, a method, called \(\Pi\)-method, is proposed, which consists of approximating the regression function by a sum of products of univariate functions. Thus, multivariate estimation is reduced to the estimation of the univariate functions in the products. A convergent algorithm is described, and the control of the degrees of freedom used in the fit is included. Many examples are given to demonstrate the good quality of the \(\Pi\)-method. The author points out that usually only a few products are enough to fit even fairly complicated functions.