Zeroth-order algorithms for nonconvex–strongly-concave minimax problems with improved complexities
Zeroth-order algorithms for nonconvex–strongly-concave minimax problems with improved complexities
复制标题
用于非凸强凹极小极大问题的零阶算法,并提高了复杂性
DOI:
10.1007/s10898-022-01160-0
复制
发表时间:
2022
影响因子:
1.8
通讯作者:
Razaviyayn, Meisam
中科院分区:
文献类型:
--
作者:
Wang, Zhongruo;Balasubramanian, Krishnakumar;Ma, Shiqian;Razaviyayn, Meisam
In this paper, we study zeroth-order algorithms for minimax optimization problems that are nonconvex in one variable and strongly-concave in the other variable. Such minimax optimization problems have attracted significant attention lately due to their applications in modern machine learning tasks. We first consider a deterministic version of the problem. We design and analyze the Zeroth-Order Gradient Descent Ascent (ZO-GDA) algorithm, and provide improved results compared to existing works, in terms of oracle complexity. We also propose the Zeroth-Order Gradient Descent Multi-Step Ascent (ZO-GDMSA) algorithm that significantly improves the oracle complexity ofZO-GDA. We then consider stochastic versions ofZO-GDAandZO-GDMSA, to handle stochastic nonconvex minimax problems. For this case, we provide oracle complexity results under two assumptions on the stochastic gradient: (i) the uniformly bounded variance assumption, which is common in traditional stochastic optimization, and (ii) the Strong Growth Condition (SGC), which has been known to be satisfied by modern over-parameterized machine learning models. We establish that under the SGC assumption, the complexities of the stochastic algorithms match that of deterministic algorithms. Numerical experiments are presented to support our theoretical results.
登录
查看更多内容
DOI:
--
发表时间:
2017
期刊:
Information Technology Convergence and Services
影响因子:
--
作者:
G. Piliouras;L. Schulman
通讯作者:
L. Schulman
DOI:
--
发表时间:
2019
期刊:
International Conference on Learning Representations
影响因子:
--
作者:
Sina Baharlouei;Maher Nouiehed;Meisam Razaviyayn
通讯作者:
Meisam Razaviyayn
影响因子:
1.8
作者:
D. Bertsimas;O. Nohadani
通讯作者:
D. Bertsimas;O. Nohadani
影响因子:
5.4
作者:
Songtao Lu;Ioannis C. Tsaknakis;Mingyi Hong;Yongxin Chen
通讯作者:
Songtao Lu;Ioannis C. Tsaknakis;Mingyi Hong;Yongxin Chen
影响因子:
3
作者:
K. Balasubramanian;Saeed Ghadimi
通讯作者:
K. Balasubramanian;Saeed Ghadimi