$\ell^1$ Penalization
joint work with K. Bertin and V. Rivoirard
Trying to answer a question asked by A. Tsybakov, I have considered, with K. Bertin and V. Rivoirard, a density estimation algorithm based on a variation of the Lasso, the Dantzig estimator. In this setting, we have shown how to fully exploit the density framework to calibrate accurately and automatically the Dantzig constraints from the data both theoretically and numerically.