Cubic Regularized Newton Method with Variance Reduction for Finite-sum Non-convex Problems
Abstract
We study finite-sum non-convex optimization and analyze a variance-reduced cubic Newton method based on EMA-smoothed SARAH estimators for both gradient and Hessian information. The method combines a coarse stochastic backbone with a terminal homotopy refinement: once the iterates enter a certified small-step regime, the algorithm decreases the regularization level geometrically, shortens the stage length, and increases the mini-batch size at the reciprocal rate while restarting exact finite-sum snapshots at each stage. We work under average squared gradient smoothness and average mean-cubed Hessian smoothness, thereby avoiding the trajectory-wise Hessian boundedness assumption that is often used in related analyses. Under these assumptions and a standard inexact cubic-subproblem certificate, we establish that the method returns an -second-order stationary point with total finite-sum oracle complexity . The analysis separates into a coarse progress phase, which yields the -scaled stochastic backbone, and a terminal local bootstrap, which supplies the pointwise accuracy needed to turn the model step certificate into a true second-order certificate.
Cite
@article{arxiv.2510.08714,
title = {Cubic Regularized Newton Method with Variance Reduction for Finite-sum Non-convex Problems},
author = {Dmitry Pasechnyuk-Vilensky and Dmitry Kamzolov and Martin Takáč},
journal= {arXiv preprint arXiv:2510.08714},
year = {2026}
}
Comments
14 pages