Slow Learners are Fast

John Langford, Alexander Smola, Martin Zinkevich

Online learning algorithms have impressive convergence properties when it comes to risk minimization and convex games on very large problems. However, they are inherently sequentia…