TimelyFL: Heterogeneity-aware Asynchronous Federated Learning with Adaptive Partial Training
CVPR Workshop, 2023
This work proposes TimelyFL, a heterogeneity-aware asynchronous federated learning framework that adaptively adjusts each client’s local training workload based on real-time resource capacity to reduce staleness and boost participation, improving participation rate by 21.13%, convergence efficiency by 1.28x to 2.89x, and test accuracy by 6.25% over the state-of-the-art FedBuff.
BibTeX
@inproceedings{zhang2023timelyfl,
title={Timelyfl: Heterogeneity-aware asynchronous federated learning with adaptive partial training},
author={Zhang, Tuo and Gao, Lei and Lee, Sunwoo and Zhang, Mi and Avestimehr, Salman},
booktitle={2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)},
pages={5064--5073},
year={2023},
organization={IEEE}
}