Distributed Asynchronous Dual-Free Stochastic Dual Coordinate Ascent

In this paper, we propose a new Distributed Asynchronous Dual-Free Coordinate Ascent method (dis-dfSDCA), and prove that it has linear convergence rate in convex case. Stochastic Dual Coordinate Ascent (SDCA) is a popular method in solving regularized convex loss minimization problems. Dual-Free Stochastic Dual Coordinate Ascent (dfSDCA) method is a variation of SDCA, and can be applied to a more general problem when its dual problem is meaningless. We extend dfSDCA method to distributed version, and provide theoretical analysis. We perform large scale experiments on distributed system and experimental results validate our findings.