This thesis aims at proposing a new method of solving the nonparametric and non-additive regression problem in presence of ultra-high dimensional data. In this context, there are two relevant aspects: variable selection ...