Efficient Adaptive Federated Optimization

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Lee, Su Hyeong, Sharma, Sidharth, Zaheer, Manzil, Li, Tian
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909868617629696
author Lee, Su Hyeong
Sharma, Sidharth
Zaheer, Manzil
Li, Tian
author_facet Lee, Su Hyeong
Sharma, Sidharth
Zaheer, Manzil
Li, Tian
contents Adaptive optimization is critical in federated learning, where enabling adaptivity on both the server and client sides has proven essential for achieving optimal performance. However, the scalability of such jointly adaptive systems is often hindered by resource limitations in communication and memory. In this paper, we introduce a class of efficient adaptive algorithms, named $FedAda^2$ and its enhanced version $FedAda^2$++, designed specifically for large-scale, cross-device federated environments. $FedAda^2$ optimizes communication efficiency by avoiding the transfer of preconditioners between the server and clients. Additionally, $FedAda^2$++ extends this approach by incorporating memory-efficient adaptive optimizers on the client side, further reducing on-device memory usage. Theoretically, we demonstrate that $FedAda^2$ and $FedAda^2$++ achieve the same convergence rates for general, non-convex objectives as its more resource-intensive counterparts that directly integrate joint adaptivity. Extensive empirical evaluations on image and text datasets demonstrate both the advantages of joint adaptivity and the effectiveness of $FedAda^2$/$FedAda^2$++.
format Preprint
id arxiv_https___arxiv_org_abs_2410_18117
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Efficient Adaptive Federated Optimization
Lee, Su Hyeong
Sharma, Sidharth
Zaheer, Manzil
Li, Tian
Machine Learning
Distributed, Parallel, and Cluster Computing
Optimization and Control
Adaptive optimization is critical in federated learning, where enabling adaptivity on both the server and client sides has proven essential for achieving optimal performance. However, the scalability of such jointly adaptive systems is often hindered by resource limitations in communication and memory. In this paper, we introduce a class of efficient adaptive algorithms, named $FedAda^2$ and its enhanced version $FedAda^2$++, designed specifically for large-scale, cross-device federated environments. $FedAda^2$ optimizes communication efficiency by avoiding the transfer of preconditioners between the server and clients. Additionally, $FedAda^2$++ extends this approach by incorporating memory-efficient adaptive optimizers on the client side, further reducing on-device memory usage. Theoretically, we demonstrate that $FedAda^2$ and $FedAda^2$++ achieve the same convergence rates for general, non-convex objectives as its more resource-intensive counterparts that directly integrate joint adaptivity. Extensive empirical evaluations on image and text datasets demonstrate both the advantages of joint adaptivity and the effectiveness of $FedAda^2$/$FedAda^2$++.
title Efficient Adaptive Federated Optimization
topic Machine Learning
Distributed, Parallel, and Cluster Computing
Optimization and Control
url https://arxiv.org/abs/2410.18117