Primal-dual subgradient methods for minimizing uniformly convex functions