Block-Coordinate Frank-Wolfe Optimization for Structural SVMs
This provides an efficient solver for structural SVMs, which are used in machine learning for structured prediction tasks, but it is incremental as it builds on existing Frank-Wolfe methods.
The paper tackled the problem of optimizing structural SVMs by proposing a randomized block-coordinate Frank-Wolfe algorithm, which achieves a similar convergence rate as the full algorithm with lower iteration cost and outperforms competing solvers in experiments.
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.