A second-order descent method with active-set prediction for group sparse optimization.

Abstract: In this talk, we propose a second-order algorithm for the solution of finite and infinite dimensional group sparse optimization problems. Group sparse optimization has gained a lot of attention in the last years due to several important classification problems requiring group sparse solutions. The most prominent application example is the group LASSO problem, which consists in minimizing a least-squares fitting term together with the group sparsity l1/l2 norm. The method is built upon the steepest descent directions of the nonsmooth problem, which are further modified by using second-order information. A prediction step is also proposed for faster identification of the strong active set. A general convergence result is proved in Rn, and the active set behavior is analyzed. The work ends with comparative computational experiments to test the performance of the devised algorithm.

Date: Oct 18, 2023 at 17:15:00 h
Venue: Sala de Seminario John Von Neumann, CMM, Beauchef 851, Torre Norte, Piso 7.
Speaker: Sofía López
Affiliation: Escuela Politécnica Nacional, Ecuador.
Coordinator: Emilio Vilches
More info at:
Event website
Abstract:
PDF

Posted on Oct 16, 2023 in Optimization and Equilibrium, Seminars