2000 character limit reached
Proximal Stochastic Dual Coordinate Ascent (1211.2717v1)
Published 12 Nov 2012 in stat.ML, cs.LG, and math.OC
Abstract: We introduce a proximal version of dual coordinate ascent method. We demonstrate how the derived algorithmic framework can be used for numerous regularized loss minimization problems, including $\ell_1$ regularization and structured output SVM. The convergence rates we obtain match, and sometimes improve, state-of-the-art results.
Collections
Sign up for free to add this paper to a collection.
