TY - GEN
T1 - Optimal Design of Hybrid Federated and Centralized Learning in the Mobile Edge Computing Systems
AU - Hong, Wei
AU - Luo, Xueting
AU - Zhao, Zhongyuan
AU - Peng, Mugen
AU - Quek, Tony Q.S.
N1 - Funding Information:
VII. ACKNOWLEDGMENT This work was supported in part by Beijing Natural Science Foundation (Grant L182039), and National Natural Science Foundation of China (Grant 61971061).
Publisher Copyright:
© 2021 IEEE.
PY - 2021/6
Y1 - 2021/6
N2 - It is a dilemma to balance the tradeoff between the computation efficiency and communication cost of deploying deep learning models in the mobile edge computing (MEC) systems, due to the isolation of collected data and computation capability. To solve this problem, a hybrid federated and centralized learning scheme is first proposed in this paper, where the learning model can be jointly generated based on the centralized learning model and the federated learning model. It can make full use of both the collected data of user terminals and power full computation capability of edge computing servers. Second, to guarantee the model accuracy with communication, computation, and data constraints, an optimization algorithm is designed to keep a sophisticated tradeoff of model accuracy and training cost. Finally, the experiment results base on the image data set are provided, which show that our proposed algorithm can significantly improve the model accuracy with low costs.
AB - It is a dilemma to balance the tradeoff between the computation efficiency and communication cost of deploying deep learning models in the mobile edge computing (MEC) systems, due to the isolation of collected data and computation capability. To solve this problem, a hybrid federated and centralized learning scheme is first proposed in this paper, where the learning model can be jointly generated based on the centralized learning model and the federated learning model. It can make full use of both the collected data of user terminals and power full computation capability of edge computing servers. Second, to guarantee the model accuracy with communication, computation, and data constraints, an optimization algorithm is designed to keep a sophisticated tradeoff of model accuracy and training cost. Finally, the experiment results base on the image data set are provided, which show that our proposed algorithm can significantly improve the model accuracy with low costs.
UR - https://www.scopus.com/pages/publications/85112818116
UR - https://www.scopus.com/pages/publications/85112818116#tab=citedBy
U2 - 10.1109/ICCWorkshops50388.2021.9473489
DO - 10.1109/ICCWorkshops50388.2021.9473489
M3 - Conference contribution
AN - SCOPUS:85112818116
T3 - 2021 IEEE International Conference on Communications Workshops, ICC Workshops 2021 - Proceedings
BT - 2021 IEEE International Conference on Communications Workshops, ICC Workshops 2021 - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2021 IEEE International Conference on Communications Workshops, ICC Workshops 2021
Y2 - 14 June 2021 through 23 June 2021
ER -