Generalization of ERM in Stochastic Convex Optimization: The Dimension Strikes Back ∗

–Neural Information Processing Systems 

In stochastic convex optimization the goal is to minimize a convex function F (x) =.