Transforming Threats to Assets: Utilizing Backdoor Attack Models in Federated Learning
IEEE Transactions on Consumer Electronics · 2026-02 · 72 (1) · 170-181
DOI: 10.1109/tce.2025.3647772 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Consumer Electronics, 2026-02. Verfasst von Au Ashley Hoi-Ting und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 33 Literaturangaben. Gefördert von: U.K. Engineering and Physical Sciences Research Council, National Edge AI Hub for Real Data: Edge Intelligence for Cyber-disturbances and Data Quality.
Abstract (Originalsprache)
Federated Learning (FL) is recognized as a privacy-preserving machine learning technique suitable for distributed devices in consumer electronics. However, FL is still vulnerable to backdoor attacks, where malicious clients poison the global model to misclassify triggered inputs while performing normally on clean data. While existing defence techniques focus on eliminating malicious clients, this paper proposes a novel technique called FedAssets. In FedAssets, we uncover distinctive parameter patterns in malicious local models and introduce a clustering-based approach to differentiate between malicious and benign clients. Further, we develop a technique to extract useful information from malicious local models and integrate it effectively, together with benign models, into the global model. Extensive experiments demonstrate that FedAssets outperforms existing techniques, improving model accuracy on clean data up to 7% while achieving up to 35% accuracy in eliminating poisoned information in the global model.
Förderung
- U.K. Engineering and Physical Sciences Research Council
- National Edge AI Hub for Real Data: Edge Intelligence for Cyber-disturbances and Data Quality
Zitieren
APA
Hoi-Ting, A. A., He, L., Nezhad, M. M., & Maple, C. (2026). Transforming Threats to Assets: Utilizing Backdoor Attack Models in Federated Learning. IEEE Transactions on Consumer Electronics, 72(1), 170-181. https://doi.org/10.1109/tce.2025.3647772
Vancouver
Hoi-Ting AA, He L, Nezhad MM, Maple C. Transforming Threats to Assets: Utilizing Backdoor Attack Models in Federated Learning. IEEE Transactions on Consumer Electronics. 2026;72(1):170-181. doi:10.1109/tce.2025.3647772
BibTeX
@article{HoiTing2026,
author = {Hoi-Ting, Au Ashley and He, Ligang and Nezhad, Mahshid Mehr and Maple, Carsten},
title = {Transforming Threats to Assets: Utilizing Backdoor Attack Models in Federated Learning},
journal = {IEEE Transactions on Consumer Electronics},
year = {2026},
volume = {72},
number = {1},
pages = {170--181},
doi = {10.1109/tce.2025.3647772}
}
EUVIMED