1: \begin{abstract}

2:   Sparse modeling is a powerful framework for data analysis and

3:   processing. Traditionally, encoding in this framework is performed by

4:   solving an $\ell_1$-regularized linear regression problem, commonly

5:   referred to as \emph{Lasso} or \emph{Basis Pursuit}. In this work we

6:   combine the sparsity-inducing property of the Lasso at the individual

7:   feature level, with the block-sparsity property of the \emph{Group Lasso},

8:   where sparse groups of features are jointly encoded, obtaining a sparsity

9:   pattern hierarchically structured. This results in the \emph{Hierarchical

10:     Lasso (HiLasso)}, which shows important practical advantages.  We then

11:   extend this approach to the collaborative case, where a set of

12:   simultaneously coded signals share the same sparsity pattern at the higher

13:   (group) level, but not necessarily at the lower (inside the group) level,

14:   obtaining the collaborative HiLasso model \emph{(C-HiLasso)}. Such signals

15:   then share the same active groups, or classes, but not necessarily the

16:   same active set. This model is very well suited for applications such as

17:   source identification and separation. An efficient optimization procedure,

18:   which guarantees convergence to the global optimum, is developed for these

19:   new models. The underlying presentation of the framework and optimization

20:   approach is complemented by experimental examples and theoretical results

21:   regarding recovery guarantees.

22: \end{abstract}

23: