1: \begin{abstract}

2: In this paper,  we develop a symmetric accelerated stochastic

3: Alternating Direction Method of Multipliers (SAS-ADMM) for solving separable

4: convex optimization problems with linear constraints.

5: The objective  function  is the sum of a possibly  nonsmooth convex function and an

6: average function of many smooth convex functions.

7: Our proposed algorithm combines both ideas of ADMM and the techniques of

8: accelerated stochastic gradient methods using variance reduction to solve

9: the smooth subproblem.

10: One main feature of SAS-ADMM {is} that its dual variable is symmetrically updated

11: after each update of the separated primal variable,

12: which would allow a more flexible and larger convergence region

13: of the dual variable compared with that of standard deterministic or stochastic ADMM.

14: % To simplify computation of the  possibly nonsmooth subproblem, we add a  proper proximal term to transform   this subproblem into a proximal mapping.

15: This new stochastic optimization algorithm is shown to converge

16: in expectation with  $\C{O}(1/T)$ convergence rate, where $T$ is the number of outer iterations.

17: In addition, 3-block extensions of the algorithm and its variant of an accelerated stochastic

18: augmented Lagrangian method are also discussed.

19: Our preliminary numerical experiments indicate the proposed algorithm is very effective

20: for solving separable optimization problems from big-data applications

21: 

22: \end{abstract}

23: