Block-Coordinate Frank-Wolfe Optimization for Structural SVMs - Archive ouverte HAL Access content directly
Conference Papers Year : 2013

Block-Coordinate Frank-Wolfe Optimization for Structural SVMs


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.

Dates and versions

hal-00720158 , version 1 (23-07-2012)



Simon Lacoste-Julien, Martin Jaggi, Mark Schmidt, Patrick Pletscher. Block-Coordinate Frank-Wolfe Optimization for Structural SVMs. ICML 2013 International Conference on Machine Learning, Jun 2013, Atlanta, United States. pp.53-61. ⟨hal-00720158⟩
451 View
0 Download



Gmail Facebook Twitter LinkedIn More