Inferring the group Lasso structure via bilevel optimization

J. Frecon, S. Salzo, and M. Pontil
ICML Workshop - Modern Trends in Nonconvex Optimization for Machine Learning, 2018


We present a continuous bilevel optimization problem for inferring the Group Lasso structure. It relies on an approximation where the lower level problem is replaced by a smooth dual forward-backward scheme with Bregman distances. Theoretical guarantees regarding its convergence to the exact problem are also provided.


PDF   BibTex   Toolbox