ICCOPT 2013 Talk, Room 2.2, Tuesday, July 30, 16:30-18:00

 Speaker: Federico Pierucci, INRIA, LEAR team and LJK, Laboratoire Jean Kuntzmann, France
 Title: Conditional gradient algorithm for machine learning with non-smooth loss and decomposable regularization
 Co-authors: Zaid Harchaoui, Juditsky Anatoli, Malick Jérôme

 Abstract:
Scientific Program

We consider the problem of optimizing machine learning objectives with a decomposable regularization penalty and a non-smooth loss function. For several important learning problems, state-of-the-art optimization approaches such as proximal gradient algorithms are difficult to apply and do not scale up to large datasets. We propose a new conditional-type algorithm, with theoretical guarantees, for such problems. Promising experimental results are presented on real-world datasets.


 Talk in: Organized Session Tue.C.22 Coordinate descent and incremental gradient methods for nonsmooth optimization
 Cluster: Convex and nonsmooth optimization


 Go to: Tue.C
 Go to: unframed Scientific Program

 Go to: ICCOPT 2013 Main Webpage