Sham Kakade - Parallelizing and Accelerating Stochastic Gradient Descent for Convex and Non-Convex Optimization