Generalization of ℓ1ℓ1 constraints for high dimensional regression problems

We focus on the high dimensional linear regression Y∼N(Xβ∗,σ2In)Y∼N(Xβ∗,σ2In), where β∗∈Rpβ∗∈Rp is the parameter of interest. In this setting, several estimators such as the LASSO (Tibshirani, 1996) and the Dantzig Selector (Candes and Tao, 2007) are known to satisfy interesting properties whenever the vector β∗β∗ is sparse. Interestingly, both the LASSO and the Dantzig Selector can be seen as orthogonal projections of 0 into DC(s)={β∈Rp,‖X′(Y−Xβ)‖∞≤s}DC(s)={β∈Rp,‖X′(Y−Xβ)‖∞≤s}, using an ℓ1ℓ1 distance for the Dantzig Selector and ℓ2ℓ2 for the LASSO. For a well chosen s>0s>0, this set is actually a confidence region for β∗β∗. In this paper, we investigate the properties of estimators defined as projections on DC(s)DC(s) using general distances. We prove that the obtained estimators satisfy oracle properties close to the one of the LASSO and the Dantzig Selector. On top of that, it turns out that these estimators can be tuned to exploit a different sparsity or/and slightly different estimation objectives.