A family of second-order methods for convex \(\ell _1\)-regularized optimization (Q312690)

The paper deals with the minimization of a convex objective function augmented with an \(L_1\) regularization term, and to this end it proposes a new Block Active-Set (BAS) method which involves multiple changes in the active manifold estimate at every estimation step. The basic BAS method is firstly described, and the authors propose a new block active-set algorithm to rectify some failures caused by the definition of the matrix involved in the quadratic form of the function to optimize. Modified versions of the orthant method are proposed, and a numerical example illustrates the efficiency of the method compared with other ones (FISTA for example).