An Adaptive Knot Selection Method for Regression Splines via Penalized Minimum Contrast Estimation

Abstract

In this talk, a knot selection method for regression splines will be presented. This method yields a penalized least square spline estimator that is adaptive and allows for non-equally spaced knots. If the regression function belongs to a Sobolev space, then the penalized least square spline estimator can converge to the regression function at a nearly optimal rate.