Federated Learning With Stage-Wise Trajectory Matching for AAV Networks on Non-IID Data
IEEE Transactions on Cognitive Communications and Networking · 2026 · 12 · 6698-6709
DOI: 10.1109/tccn.2026.3670272 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Cognitive Communications and Networking, 2026. Verfasst von Xiufang Shi und 5 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 31 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Major Humanities and Social Sciences Research Projects in Zhejiang Universities, Fundamental Research Funds for the Provincial Universities of Zhejiang.
Abstract (Originalsprache)
Federated learning enables multiple autonomous aerial vehicles (AAVs) to collaboratively train a global model while preserving data privacy. However, data heterogeneity across AAV swarms hinders model convergence and generalization. While existing trajectory matching methods address this problem by synthesizing distilled data from early-stage training trajectory, this trajectory are non-convergent and fail to capture the complete global distribution. To address this issue, we propose a novel federated learning framework based on stage-wise trajectory matching (FedSTM). This framework performs trajectory matching during both the early and late stages of training to synthesize distilled data that comprehensively represent the global distribution, and subsequently utilizes these distilled data to fine-tune the global model, effectively enhancing its generalization capability and stability. Additionally, we introduce a decorrelation regularization term during local training to mitigate dimensional collapse in the representation space, thereby improving the quality of the global model’s trajectory. Extensive experimental results on multiple datasets demonstrate that the proposed method surpasses baseline approaches in model performance and convergence. Code is available at https://github.com/zhangweikk/FedSTM
Förderung
- National Natural Science Foundation of China
- Major Humanities and Social Sciences Research Projects in Zhejiang Universities
- Fundamental Research Funds for the Provincial Universities of Zhejiang
- Key Research and Development Program of Zhejiang Province
- Zhejiang Provincial Major Science and Technology Project
Zitieren
APA
Shi, X., Zhang, W., Li, Y., Wu, M., Li, R., & He, S. (2026). Federated Learning With Stage-Wise Trajectory Matching for AAV Networks on Non-IID Data. IEEE Transactions on Cognitive Communications and Networking, 12, 6698-6709. https://doi.org/10.1109/tccn.2026.3670272
Vancouver
Shi X, Zhang W, Li Y, Wu M, Li R, He S. Federated Learning With Stage-Wise Trajectory Matching for AAV Networks on Non-IID Data. IEEE Transactions on Cognitive Communications and Networking. 2026;12:6698-6709. doi:10.1109/tccn.2026.3670272
BibTeX
@article{Shi2026,
author = {Shi, Xiufang and Zhang, Wei and Li, Yuheng and Wu, Mincheng and Li, Rushi and He, Shibo},
title = {Federated Learning With Stage-Wise Trajectory Matching for AAV Networks on Non-IID Data},
journal = {IEEE Transactions on Cognitive Communications and Networking},
year = {2026},
volume = {12},
pages = {6698--6709},
doi = {10.1109/tccn.2026.3670272}
}
EUVIMED