26449 articles – 20243 Notices  [english version]
Fiche détaillée Communications avec actes
ICML 2013 International Conference on Machine Learning, Atlanta : États-Unis (2013)
Block-Coordinate Frank-Wolfe Optimization for Structural SVMs
Simon Lacoste-Julien1, 2, Martin Jaggi3, Mark Schmidt1, 2, Patrick Pletscher4

We propose a randomized block-coordinate variant of the classic Frank-Wolfe algorithm for convex optimization with block-separable constraints. Despite its lower iteration cost, we show that it achieves a similar convergence rate in duality gap as the full Frank-Wolfe algorithm. We also show that, when applied to the dual structural support vector machine (SVM) objective, this yields an online algorithm that has the same low iteration complexity as primal stochastic subgradient methods. However, unlike stochastic subgradient methods, the block-coordinate Frank-Wolfe algorithm allows us to compute the optimal step-size and yields a computable duality gap guarantee. Our experiments indicate that this simple algorithm outperforms competing structural SVM solvers.
1 :  INRIA Paris - Rocquencourt - SIERRA
2 :  LIENS - Laboratoire d'informatique de l'école normale supérieure
3 :  CMAP - Centre de Mathématiques Appliquées - Ecole Polytechnique
4 :  Machine Learning Laboratory
SIERRA