TY - JOUR
T1 - Wireless Communications for Collaborative Federated Learning
AU - Chen, Mingzhe
AU - Poor, H. Vincent
AU - Saad, Walid
AU - Cui, Shuguang
N1 - Funding Information:
The work was supported in part by the Key Area R&D Program of Guangdong Province with grant No. 2018B030338001, by the National Natural Science Foundation of China with grant NSFC-61629101, by Guangdong Research Project No. 2017ZT07X152, and by the U.S. National Science Foundation under Grants CCF-1908308 and CNS-1814477.
Publisher Copyright:
© 1979-2012 IEEE.
PY - 2020/12
Y1 - 2020/12
N2 - To facilitate the deployment of machine learning in resource and privacy-constrained systems such as the Internet of Things, federated learning (FL) has been proposed as a means for enabling edge devices to train a shared learning model while promoting privacy. However, Google's seminal FL algorithm requires all devices to be directly connected with a central controller, which limits its applications. In contrast, this article introduces a novel FL framework, called collaborative FL (CFL), which enables edge devices to implement FL with less reliance on a central controller. The fundamentals of this framework are developed and a number of communication techniques are proposed so as to improve CFL performance. An overview of centralized learning, Google's FL, and CFL is presented. For each type of learning, the basic architecture as well as its advantages, drawbacks, and operating conditions are introduced. Then four CFL performance metrics are presented, and a suite of communication techniques ranging from network formation, device scheduling, mobility management, to coding are introduced to optimize the performance of CFL. For each technique, future research opportunities are discussed. In a nutshell, this article showcases how CFL can be effectively implemented at the edge of large-scale wireless systems.
AB - To facilitate the deployment of machine learning in resource and privacy-constrained systems such as the Internet of Things, federated learning (FL) has been proposed as a means for enabling edge devices to train a shared learning model while promoting privacy. However, Google's seminal FL algorithm requires all devices to be directly connected with a central controller, which limits its applications. In contrast, this article introduces a novel FL framework, called collaborative FL (CFL), which enables edge devices to implement FL with less reliance on a central controller. The fundamentals of this framework are developed and a number of communication techniques are proposed so as to improve CFL performance. An overview of centralized learning, Google's FL, and CFL is presented. For each type of learning, the basic architecture as well as its advantages, drawbacks, and operating conditions are introduced. Then four CFL performance metrics are presented, and a suite of communication techniques ranging from network formation, device scheduling, mobility management, to coding are introduced to optimize the performance of CFL. For each technique, future research opportunities are discussed. In a nutshell, this article showcases how CFL can be effectively implemented at the edge of large-scale wireless systems.
UR - http://www.scopus.com/inward/record.url?scp=85099086378&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85099086378&partnerID=8YFLogxK
U2 - 10.1109/MCOM.001.2000397
DO - 10.1109/MCOM.001.2000397
M3 - Article
AN - SCOPUS:85099086378
SN - 0163-6804
VL - 58
SP - 48
EP - 54
JO - IEEE Communications Magazine
JF - IEEE Communications Magazine
IS - 12
M1 - 9311931
ER -