Goto

Collaborating Authors

 original bcd algorithm


Reviews: Fast Sparse Group Lasso

Neural Information Processing Systems

Summary: This paper presents a fast block coordinate descent algorithm for the sparse-group lasso problem. Two strategies are proposed to improve the computational efficiency. The first strategy is quickly identifying the groups of inactive features. The idea is to use an easy-to-compute upper bound when checking if the inactive-group condition holds. The second strategy is to select a set of candidate groups and update the feature vectors inside those groups first before iterating over all groups.