Work

Analysis and Optimization for Over-the-Air Federated Learning with Energy Harvesting

Year: 2023
Type: dissertation
Abstract: Analysis and Optimization for Over-the-Air Federated Learning with Energy HarvestingFederated learning (FL) has attracted significant attention as a promising training framework that effectively utilizes distributed data and computational resources while addressing communication and privacy concerns.However, the communication bottleneck poses a major challenge in traditional FL systems for edge-intelligent networks, where high-dimensional model parameters need to be iteratively updated between the parameter server (PS) and local clients.This frequent back-and-forth communication during the iterative FL process leads to the issue of communication overload.To mitigate this, over-the-air (OTA) computation has been introduced for the FL system, where the PS only requires the average of local model updates instead of individual updates.The communication overhead can be reduced for OTA FL by making full utilization of the spectrum resources when clients send the gradients to PS.In addition, the privacy leakage from the client to the PS can be avoided for OTA FL since the signals received at the PS side through OTA computation are aggregated signals.Nevertheless, local clients often have limited resources, particularly in terms of battery capacity, which can affect the convergence performance of the OTA FL system.Moreover, replacing batteries in local clients is inconvenient and labor-intensive.In most of the current studies, energy harvesting technologies have not been considered for the OTA FL system.Besides, energy harvested from the environment such as wind, solar power, and human motion can enable local clients to perform sustained training and achieve green computing.Some prior research studies have successfully applied energy harvesting techniques to wireless transmission, task offloading, and resource allocation for mobile edge computing.Different from the previous studies, we propose a novel energy-aware OTA FL system that incorporates the energy harvesting technique to supply power to local clients.Besides, the client selection and receive beamforming design are optimized to realize fast convergence for the OTA FL system under the energy constraint.In this thesis, our objective is to apply energy harvesting techniques to the energyconstrained OTA FL system to achieve durable training, energy-efficient computation, and rapid convergence.Additionally, we aim to thoroughly investigate the impact of energy harvesting settings on the convergence performance of the OTA FL system.To accomplish this, we first optimize client selection and energy management to achieve fast convergence in the OTA FL system with energy harvesting clients,
Authors: Caijuan, Chen, サイケン, チン, Caijuan, Chen
Language: English
Open Access status: green
OpenAlex ID: W7026773156
Cites: 102
Cited by:
Related to: 10