Gradient Norm Regularization Second-Order Algorithms for Solving Nonconvex-Strongly Concave Minimax Problems
This work addresses optimization challenges in machine learning for problems like adversarial training, but it is incremental as it matches existing complexity bounds.
The paper tackles nonconvex-strongly concave minimax problems by proposing gradient norm regularized second-order algorithms, achieving an iteration complexity of Õ(ℓ^{1.5}ρ^{0.5}μ^{-1.5}ε^{-1.5}) to find second-order stationary points, matching the best known complexity.
In this paper, we study second-order algorithms for solving nonconvex-strongly concave minimax problems, which have attracted much attention in recent years in many fields, especially in machine learning.We propose a gradient norm regularized trust-region (GRTR) algorithm to solve nonconvex-strongly concave minimax problems, where the objective function of the trust-region subproblem in each iteration uses a regularized version of the Hessian matrix, and the regularization coefficient and the radius of the ball constraint are proportional to the square root of the gradient norm. The iteration complexity of the proposed GRTR algorithm to obtain an $O(ε,\sqrtε)$-second-order stationary point is proved to be upper bounded by $\tilde{O}(\ell^{1.5}ρ^{0.5}μ^{-1.5}ε^{-1.5})$, where $μ$ is the strong concave coefficient, $\ell$ and $ρ$ are the Lipschitz constant of the gradient and Jacobian matrix respectively, which matches the best known iteration complexity of second-order methods for solving nonconvex-strongly concave minimax problems. We further propose a Levenberg-Marquardt algorithm with a gradient norm regularization coefficient and use the negative curvature direction to correct the iteration direction (LMNegCur), which does not need to solve the trust-region subproblem at each iteration. We also prove that the LMNegCur algorithm achieves an $O(ε,\sqrtε)$-second-order stationary point within $\tilde{O}(\ell^{1.5}ρ^{0.5}μ^{-1.5}ε^{-1.5})$ number of iterations.The inexact variants of both algorithms can still obtain $O(ε,\sqrtε)$-second-order stationary points with high probability, but only require $\tilde{O}(\ell^{2.25}ρ^{0.25}μ^{-1.75}ε^{-1.75})$ Hessian-vector products and $\tilde{O}(\ell^{2}ρ^{0.5}μ^{-2}ε^{-1.5})$ gradient ascent steps.