ICML 2022spotlight62 citations
Federated Minimax Optimization: Improved Convergence Analyses and Algorithms
Pranay Sharma, Rohan Panda, Gauri Joshi, Pramod Varshney
Abstract
In this paper, we consider nonconvex minimax optimization, which is gaining prominence in many modern machine learning applications, such as GANs. Large-scale edge-based collection of training data in these applications calls for communication-efficient distributed optimization algorithms, such as those used in federated learning, to process the data. In this paper, we analyze local stochastic gradient descent ascent (SGDA), the local-update version of the SGDA algorithm. SGDA is the core algorithm used in minimax optimization, but it is not well-understood in a distributed setting. We prove that Local SGDA has
BibTeX
@InProceedings{pmlr-v162-sharma22c,
title = {Federated Minimax Optimization: Improved Convergence Analyses and Algorithms},
author = {Sharma, Pranay and Panda, Rohan and Joshi, Gauri and Varshney, Pramod},
booktitle = {Proceedings of the 39th International Conference on Machine Learning},
pages = {19683--19730},
year = {2022},
editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan},
volume = {162},
series = {Proceedings of Machine Learning Research},
month = {17--23 Jul},
publisher = {PMLR},
pdf = {https://proceedings.mlr.press/v162/sharma22c/sharma22c.pdf},
url = {https://proceedings.mlr.press/v162/sharma22c.html},
abstract = {In this paper, we consider nonconvex minimax optimization, which is gaining prominence in many modern machine learning applications, such as GANs. Large-scale edge-based collection of training data in these applications calls for communication-efficient distributed optimization algorithms, such as those used in federated learning, to process the data. In this paper, we analyze local stochastic gradient descent ascent (SGDA), the local-update version of the SGDA algorithm. SGDA is the core algorithm used in minimax optimization, but it is not well-understood in a distributed setting. We prove that Local SGDA has