To Talk or to Work: Energy Eficient Federated Learning over Mobile Devices via the Weight Qantization and 5G Transmission Co-Design Rui Chen rchen19@uh.edu University of Houston Houston, Texas, USA Liang Li liliang_1127@outlook.com Xidian University Xi’an, China Kaiping Xue kpxue@ustc.edu.cn University of Science and Technology of China China Chi Zhang chizhang@ustc.edu.cn University of Science and Technology of China China Lingjia Liu ljliu@vt.edu Virginia Tech Blacksburg, Virginia, USA Miao Pan mpan2@uh.edu University of Houston Houston, Texas, USA ABSTRACT Federated learning (FL) is a new paradigm for large-scale learning tasks across mobile devices. However, practical FL deployment over resource constrained mobile devices confronts multiple challenges. For example, it is not clear how to establish an efective wireless network architecture to support FL over mobile devices. Besides, as modern machine learning models are more and more complex, the local on-device training/intermediate model update in FL is becoming too power hungry/radio resource intensive for mobile devices to aford. To address those challenges, in this paper, we try to bridge another recent surging technology, 5G, with FL, and de- velop a wireless transmission and weight quantization co-design for energy efcient FL over heterogeneous 5G mobile devices. Briefy, the 5G featured high data rate helps to relieve the severe commu- nication concern, and the multi-access edge computing (MEC) in 5G provides a perfect network architecture to support FL. Under MEC architecture, we develop fexible weight quantization schemes to facilitate the on-device local training over heterogeneous 5G mobile devices. Observed the fact that the energy consumption of local computing is comparable to that of the model updates via 5G transmissions, we formulate the energy efcient FL problem into a mixed-integer programming problem to elaborately determine the quantization strategies and allocate the wireless bandwidth for heterogeneous 5G mobile devices. The goal is to minimize the over- all FL energy consumption (computing + 5G transmissions) over 5G mobile devices while guaranteeing learning performance and training latency. Generalized Benders’ Decomposition is applied to develop feasible solutions and extensive simulations are conducted to verify the efectiveness of the proposed scheme. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. Mobihoc ’21, June 03ś05, 2021, Shanghai, China © 2020 Association for Computing Machinery. ACM ISBN 978-1-4503-XXXX-X/18/06. . . $15.00 https://doi.org/10.1145/nnnnnnn.nnnnnnn CCS CONCEPTS · Computing methodologies → Distributed artifcial intelli- gence; Neural networks; · Theory of computation → Mixed discrete-continuous optimization. KEYWORDS 5G networks, federated learning, weight quantization, optimization ACM Reference Format: Rui Chen, Liang Li, Kaiping Xue, Chi Zhang, Lingjia Liu, and Miao Pan. 2020. To Talk or to Work: Energy Efcient Federated Learning over Mobile Devices via the Weight Quantization and 5G Transmission Co-Design. In Mobihoc ’21: ACM International Symposium on Theory, Algorithmic Foundations, and Protocol Design for Mobile Networks and Mobile Computing , June 03ś05, 2021, Shanghai, China. ACM, New York, NY, USA, 10 pages. https://doi.org/10. 1145/nnnnnnn.nnnnnnn 1 INTRODUCTION Due to the incredible surge of mobile data and the growing com- puting capabilities of mobile devices, it becomes a trend to apply deep learning (DL) on these devices to support fast responding and customized intelligent applications [7]. Recently, federated learning (FL) is expected as a promising DL solution to provide an efcient, fexible, and privacy-preserving learning framework on a large scale of mobile devices. Under the FL framework [20], each mobile device executes model training locally and then transmits the model updates instead of raw data to an FL server. The server would aggregate the intermediate results and broadcast the updated model to the participating devices. Its potential has prompted wide applications in various domains such as keyboard predictions [12], physical hazards detection in smart home [35], health event de- tection [3], etc. However, it faces signifcant challenges to deploy FL over mobile devices in practice. First, although mobile devices are gradually equipped with artifcial intelligence (AI) computing capabilities, the limited resources (e.g., battery and storage capacity) restrain them from training deep and complicated learning models. Second, it is not clear how to establish an efective wireless network architecture to support FL over mobile devices. Last but not least, the power-hungry local computing and wireless communications arXiv:2012.11070v1 [cs.NI] 21 Dec 2020