1: \begin{abstract}

2:     We consider a generic convex optimization problem associated with 

3:     regularized empirical risk minimization of linear predictors. 

4:     The problem structure allows us to reformulate it as a convex-concave 

5:     saddle point problem. 

6:     We propose a stochastic primal-dual coordinate (SPDC) method,

7:     which alternates between maximizing over a randomly chosen dual variable

8:     and minimizing over the primal variable.

9:     An extrapolation step on the primal variable is performed to obtain 

10:     accelerated convergence rate.

11:     We also develop a mini-batch version of the SPDC method which facilitates

12:     parallel computing, and an extension with weighted sampling probabilities 

13:     on the dual variables, 

14:     which has a better complexity than uniform sampling on unnormalized data.

15:     Both theoretically and empirically, we show that the SPDC method has

16:     comparable or better performance than several state-of-the-art

17:     optimization methods.

18: \end{abstract}

19: