Federated learning (FL) is a promising distributed learning paradigm that enables multiple devices to collaboratively train a global model without sharing local private data. Considering the device heterogeneity in computation and communication capabilities, asynchronous aggregation is introduced in FL to improve the learning efficiency. However, asynchronous FL suffers from communication bottlenecks due to limited wireless resources. To address this issue, we propose AQAFL, an adaptive quantization method for asynchronous FL, while also integrating device scheduling and wireless resource allocation. We first present a convergence analysis of asynchronous FL with quantized transmission, and then formulate an optimization problem to minimize the convergence upper bound under long- term latency and bandwidth constraints by jointly optimizing the quantization bit width, device scheduling, and wireless band- width allocation. Following that, we analytically characterize the optimal bandwidth allocation and prove that it can be obtained via bisection search, and derive the closed-form expression for the quantization bit width. Subsequently, an efficient device scheduling strategy is proposed to determine the number of participating devices in each round. Experiments show that compared with other methods, AQAFL reduces the total com- munication overhead by up to 69% and the overhead required to achieve the same target test accuracy by 32.6%–88.6%, while improving test accuracy by 4.1%–6.5%.
Adaptive Quantization and Device Scheduling for Asynchronous Federated Learning over Resource-constrained Wireless Networks / Li, X., Gao, Y., Liang, F., Chiasserini, C.F.. - In: IEEE TRANSACTIONS ON MOBILE COMPUTING. - ISSN 1536-1233. - (2026).
Adaptive Quantization and Device Scheduling for Asynchronous Federated Learning over Resource-constrained Wireless Networks
Carla Fabiana Chiasserini
2026
Abstract
Federated learning (FL) is a promising distributed learning paradigm that enables multiple devices to collaboratively train a global model without sharing local private data. Considering the device heterogeneity in computation and communication capabilities, asynchronous aggregation is introduced in FL to improve the learning efficiency. However, asynchronous FL suffers from communication bottlenecks due to limited wireless resources. To address this issue, we propose AQAFL, an adaptive quantization method for asynchronous FL, while also integrating device scheduling and wireless resource allocation. We first present a convergence analysis of asynchronous FL with quantized transmission, and then formulate an optimization problem to minimize the convergence upper bound under long- term latency and bandwidth constraints by jointly optimizing the quantization bit width, device scheduling, and wireless band- width allocation. Following that, we analytically characterize the optimal bandwidth allocation and prove that it can be obtained via bisection search, and derive the closed-form expression for the quantization bit width. Subsequently, an efficient device scheduling strategy is proposed to determine the number of participating devices in each round. Experiments show that compared with other methods, AQAFL reduces the total com- munication overhead by up to 69% and the overhead required to achieve the same target test accuracy by 32.6%–88.6%, while improving test accuracy by 4.1%–6.5%.| File | Dimensione | Formato | |
|---|---|---|---|
|
TMC-2026-02-0463.R1-main.pdf
accesso aperto
Tipologia:
2. Post-print / Author's Accepted Manuscript
Licenza:
Pubblico - Tutti i diritti riservati
Dimensione
4.96 MB
Formato
Adobe PDF
|
4.96 MB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/11583/3017948
