SFedXL: Semi-synchronous federated learning with cross-sharpness and layer-freezing

Zhao, Mingxiong; Zhao, Shihao; Feng, Chenyuan; Yang, Howard H.; Niyato, Dusit; Quek Tony Q. S.
Submitted to IEEE Internet of Things Journal, January 2025

Federated learning (FL) emerges as a potential solution for enabling multiple terminal devices to collaboratively accomplish computational tasks within a Unmanned Aerial Vehicle (UAV) swarm. However, traditional FL approaches, predicated on synchronous data aggregation, are not feasible for a UAV swarm owing to the inherently variable and dynamic nature of their communication networks compared with terrestrial systems.
Furthermore, the data procured by UAVs is often highly heterogeneous, attributable to disparities in deployment environments and device attributes. Considering the distinct flight paths and unique operational conditions encountered by different UAVs,
a considerable amount of data remains unlabeled. To tackle the challenges associated with asynchronous operations and the prevalence of unlabeled data, we introduce a novel framework termed Semi-synchronous FL with Cross-Sharpness and Layer-
Freezing (SFedXL), tailored for a UAV swarm. In particular, we devise a cross-sharpness model training strategy aimed at optimizing the utilization of both labeled and unlabeled datasets. Additionally, we propose an innovative semi-synchronous model
aggregation protocol, complemented by client-specific layerfreezing and client cluster scheduling, designed to expedite the training process. Our simulation results indicate that the proposed algorithm surpasses current FL methods in terms of object recognition accuracy and communication efficiency, albeit with a trade-off of increased local computation latency.

Type:
Journal
Date:
2025-02-04
Department:
Communication systems
Eurecom Ref:
8054
Copyright:
© 2025 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.
See also:

PERMALINK : https://www.eurecom.fr/publication/8054