15
38

SSRGD: Simple Stochastic Recursive Gradient Descent for Escaping Saddle Points

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 unstable saddle points. We show that a simple perturbed version of stochastic recursive gradient descent algorithm (called SSRGD) can find an (ϵ,δ)(\epsilon,\delta)-second-order stationary point with O~(n/ϵ2+n/δ4+n/δ3)\widetilde{O}(\sqrt{n}/\epsilon^2 + \sqrt{n}/\delta^4 + n/\delta^3) stochastic gradient complexity for nonconvex finite-sum problems. As a by-product, SSRGD finds an ϵ\epsilon-first-order stationary point with O(n+n/ϵ2)O(n+\sqrt{n}/\epsilon^2) stochastic gradients. These results are almost optimal since Fang et al. [2018] provided a lower bound Ω(n/ϵ2)\Omega(\sqrt{n}/\epsilon^2) for finding even just an ϵ\epsilon-first-order stationary point. We emphasize that SSRGD algorithm for finding second-order stationary points is as simple as for finding first-order stationary points just by adding a uniform perturbation sometimes, while all other algorithms for finding second-order stationary points with similar gradient complexity need to combine with a negative-curvature search subroutine (e.g., Neon2 [Allen-Zhu and Li, 2018]). Moreover, the simple SSRGD algorithm gets a simpler analysis. Besides, we also extend our results from nonconvex finite-sum problems to nonconvex online (expectation) problems, and prove the corresponding convergence results.

View on arXiv
Comments on this paper