Emeritus Professor Anestis Antoniadis from the University Joseph Fourier, Grenoble will present the Department of Statistics Seminar entitled, "Smoothing and variable selection using P-splines".

In this talk Professor Antoniadis will focus on nonparametric penalized estimation in additive and additive varying coefficient models. A main interest is also on variable selection for such models. Particular attention goes to recent variable selection procedures such as grouped Lasso, grouped SCAD, COSSO, but also to the nonnegative garrote method introduced originally for variable selection in a multiple linear regression model. We show how the latter method combined with P-splines estimation leads to an estimation and variable consistent method in the settings of additive and varying coefficient models. The performances of this and other related selection procedures are investigated in a simulation study and illustrations on real data examples are provided. This talk is based on joint works with Irene Gijbels, Sophie Lambert-Lacroix and Anneleen Verhasselt.