ZO-AdaMM: Zeroth-Order Adaptive Momentum Method for Black-Box Optimization

arXiv1910.06513MaRDI QIDQ6327210

Author name not available (Why is that?)

Publication date: 14 October 2019

Abstract: The adaptive momentum method (AdaMM), which uses past gradients to update descent directions and learning rates simultaneously, has become one of the most popular first-order optimization methods for solving machine learning problems. However, AdaMM is not suited for solving black-box optimization problems, where explicit gradient forms are difficult or infeasible to obtain. In this paper, we propose a zeroth-order AdaMM (ZO-AdaMM) algorithm, that generalizes AdaMM to the gradient-free regime. We show that the convergence rate of ZO-AdaMM for both convex and nonconvex optimization is roughly a factor of $O(sqrtd)$ worse than that of the first-order AdaMM algorithm, where $d$ is problem size. In particular, we provide a deep understanding on why Mahalanobis distance matters in convergence of ZO-AdaMM and other AdaMM-type methods. As a byproduct, our analysis makes the first step toward understanding adaptive learning rate methods for nonconvex constrained optimization. Furthermore, we demonstrate two applications, designing per-image and universal adversarial attacks from black-box neural networks, respectively. We perform extensive experiments on ImageNet and empirically show that ZO-AdaMM converges much faster to a solution of high accuracy compared with $6$ state-of-the-art ZO optimization methods.

Has companion code repository: https://github.com/KaidiXu/ZO-AdaMM

This page was built for publication: ZO-AdaMM: Zeroth-Order Adaptive Momentum Method for Black-Box Optimization

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6327210)