posted 2010-12-07T14:22:00Z (apropos 2010-12-03).
chandrasekaran et al. seem to generalize the notion of lasso to things like nuclear norm minimization, in terms of "atomic sets" of dictionary elements like one-sparse vectors or rank-1 matrices. interestingly, they offer an alternative justification for an L1-type penalty.
---> back to algomagic.com