Multi-Task Learning for Subspace Segmentation : Supplementary File

which we advocate in (Wang et al., 2015) as a convex function that leads to convenient iterations that closely resemble those from (Qi et al., 2008). However, the exact form of this function is likely not that important as long as it favors sharing of basis functions within the constraint set. Note that Lemma 2 is based upon the alternative selection ρ(W ) = β∥W ∥2 which accomplishes more or less the same thing but is slightly easier to analyze.