AbstractWe introduce a proximal version of the stochastic dual coordinate ascent method and show how to accelerate the method using an
inner-outer iteration procedure. We analyze the runtime of the framework and obtain rates that improve state-of-the-art results for various key
→