ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2110.04814
14
14

Finding Second-Order Stationary Points in Nonconvex-Strongly-Concave Minimax Optimization

10 October 2021
Luo Luo
Yujun Li
Cheng Chen
ArXivPDFHTML
Abstract

We study the smooth minimax optimization problem min⁡xmax⁡yf(x,y)\min_{\bf x}\max_{\bf y} f({\bf x},{\bf y})minx​maxy​f(x,y), where fff is ℓ\ellℓ-smooth, strongly-concave in y{\bf y}y but possibly nonconvex in x{\bf x}x. Most of existing works focus on finding the first-order stationary points of the function f(x,y)f({\bf x},{\bf y})f(x,y) or its primal function P(x)≜max⁡yf(x,y)P({\bf x})\triangleq \max_{\bf y} f({\bf x},{\bf y})P(x)≜maxy​f(x,y), but few of them focus on achieving second-order stationary points. In this paper, we propose a novel approach for minimax optimization, called Minimax Cubic Newton (MCN), which could find an (ε,κ1.5ρε )\big(\varepsilon,\kappa^{1.5}\sqrt{\rho\varepsilon}\,\big)(ε,κ1.5ρε​)-second-order stationary point of P(x)P({\bf x})P(x) with calling O(κ1.5ρε−1.5){\mathcal O}\big(\kappa^{1.5}\sqrt{\rho}\varepsilon^{-1.5}\big)O(κ1.5ρ​ε−1.5) times of second-order oracles and O~(κ2ρε−1.5)\tilde{\mathcal O}\big(\kappa^{2}\sqrt{\rho}\varepsilon^{-1.5}\big)O~(κ2ρ​ε−1.5) times of first-order oracles, where κ\kappaκ is the condition number and ρ\rhoρ is the Lipschitz continuous constant for the Hessian of f(x,y)f({\bf x},{\bf y})f(x,y). In addition, we propose an inexact variant of MCN for high-dimensional problems to avoid calling expensive second-order oracles. Instead, our method solves the cubic sub-problem inexactly via gradient descent and matrix Chebyshev expansion. This strategy still obtains the desired approximate second-order stationary point with high probability but only requires O~(κ1.5ℓε−2)\tilde{\mathcal O}\big(\kappa^{1.5}\ell\varepsilon^{-2}\big)O~(κ1.5ℓε−2) Hessian-vector oracle calls and O~(κ2ρε−1.5)\tilde{\mathcal O}\big(\kappa^{2}\sqrt{\rho}\varepsilon^{-1.5}\big)O~(κ2ρ​ε−1.5) first-order oracle calls. To the best of our knowledge, this is the first work that considers the non-asymptotic convergence behavior of finding second-order stationary points for minimax problems without the convex-concave assumptions.

View on arXiv
Comments on this paper