1: \begin{abstract}

2:   We extend the Approximate-Proximal Point (\aprox) family of model-based

3:   methods for solving stochastic convex optimization problems, including

4:   stochastic subgradient, proximal point, and bundle methods, to the

5:   minibatch and accelerated setting. To do so, we propose specific

6:   model-based

7:   algorithms and an acceleration scheme for which we provide non-asymptotic

8:   convergence guarantees, which are order-optimal in all

9:   problem-dependent constants and provide linear speedup in

10:   minibatch size, while maintaining the desirable robustness traits

11:   (e.g.\ to stepsize) of the \aprox family.

12:   Additionally, we show improved convergence rates and

13:   matching  lower bounds identifying new fundamental constants

14:   for ``interpolation'' problems, whose 

15:   importance in statistical machine learning is growing; this, for

16:   example, gives a parallelization strategy for alternating

17:   projections. We corroborate our theoretical results with

18:   empirical testing to demonstrate the gains accurate

19:   modeling, acceleration, and minibatching provide.

20: \end{abstract}

21: