Performance estimation of stochastic first order methods

We present an extension to the performance estimation methodology, allowing for a computer-assisted analysis of stochastic first-order methods. The approach is applicable in the convex and non-convex settings, and for various suboptimality criteria including the standard minimal gradient norm of the iterates. We describe the approach, new numerical and analytical bounds on the stochastic gradient method, and a construction of a worst-case example establishing tightness of the analyses.