SSRGD: Simple Stochastic Recursive Gradient
     Descent for Escaping Saddle Points

                       Zhize Li
            Tsinghua University, China and KAUST, Saudi Arabia
                 zhizeli.thu@gmail.com

                       Abstract
     We analyze stochastic gradient algorithms for optimizing nonconvex problems. In
     particular, our goal is to find local minima (second-order stationary points) instead
     of just finding first-order stationary points which may be some bad ...