Estimating a regression function (Q921773)

A general regression problem is considered. The unknown regression function is proposed to lie in a possible nonparametric class of functions G. The author considered the methods of least squares, of least absolute deviations and of penalized least squares. These procedures differ in respect to their loss functions, but the author provides a general technique to obtain rates of convergence for the resulting estimators. The link with empirical process theory makes it possible to relate the rates of convergence to the entropy of G.