Research of Federated Learning Communication Data Transmission Algorithm Based on User Grouping Policy

Since being proposed, Machine Learning (ML) has gained a great leap. However, the shortcoming of user privacy restricts its further application in communication. Therefore, Federated Learning (FL) has been proposed to overcome this shortcoming. FL allows users to compute gradient information and upl...

Full description

Saved in:
Bibliographic Details
Published in2022 8th International Conference on Control Science and Systems Engineering (ICCSSE) pp. 170 - 174
Main Authors Cai, Yujun, Li, Shufeng, Xia, Zhiping, Tan, Yubo
Format Conference Proceeding
LanguageEnglish
Published IEEE 14.07.2022
Subjects
communication overhead
Control systems
Fading channels
Federated learning
neural network
Privacy
Simulation
System performance
Training
user grouping policy
Online AccessGet full text
DOI10.1109/ICCSSE55346.2022.10079164

Cover

More Information
Summary:Since being proposed, Machine Learning (ML) has gained a great leap. However, the shortcoming of user privacy restricts its further application in communication. Therefore, Federated Learning (FL) has been proposed to overcome this shortcoming. FL allows users to compute gradient information and upload them to the server for global training. Nevertheless, it still has the problem of high communication overhead. In this work, we propose three user grouping policies, that is, parity, random, and orderly, to study the proper method to reduce communication overhead during the training process of federated learning. Simulation results prove that the parity and the random policy outperform the orderly policy and the no grouping one in test accuracy and channel loss.
DOI:10.1109/ICCSSE55346.2022.10079164