Maicon Alves, Federal University of Santa Catarina

Abstract

We propose and study the iteration complexity of a parallel version of the forward-backward (proximal gradient) splitting method for minimizing a (possibly) large sum of convex functions with many smooth and nonsmooth terms.  We obtain pointwise (nonergodic) as well as ergodic nonasymptotic convergence rates by embedding the  proposed method within the partial inverse framework of Spingarn.

Slides     Video