EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Robust aggregation algorithms for federated learning in unreliable network environments

Ziyang Zeng, Shiyu Yang, Guanyu Ding

Journal of Computing and Electronic Information Management · 30.10.2025 · 18 (3) · 34-42

DOI: 10.54097/n0dpaf43 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Computing and Electronic Information Management, online erschienen am 30.10.2025. Verfasst von Ziyang Zeng und 2 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Federated learning (FL) allows joint model training on distributed devices without losing data locality, but its results are significantly worse in unreliable network systems where packets are dropped, clients fail, resources are heterogeneous, and adversarial (Byzantine) agents exist. The viability of FL to withstand these unfavorable conditions is keyed on the robust aggregation algorithms. The paper meticulously examines powerful methods of aggregation, which include: geometric-median methods (RFA), Krum/Multi-Krum, trimmed-mean/coordinate-wise defenses, g-divergence estimators, trust-based aggregators (FLTrust), and layer-wise aggregation methods (FedRoLA) and compares their performance on simulated unreliable networks, which model packet loss, communication delay, and malicious client actions (McMahan et al., 2017; Blanchard et al., 2017; P We examine accuracy, convergence speed, communication cost and resilience in the face of model-poisoning attacks with the help of benchmark image tasks and a set of network unreliability scenarios. We find that robust aggregators combining statistical outlier resistance and structural (layer-wise) aggregation or trust calibration (especially RFA and FedRoLA) are more accurate and converge more quickly than naive FedAvg in high packet-loss and moderate Byzantine contamination and we observe up to 12 percent improvement in test accuracy with 30 percent simulated packet loss. We also talk about the trade offs between robustness, communication overhead and privacy (secure aggregation) and present a hybrid design pattern that incorporates robust aggregation and adaptive client selection with secure aggregation so as to address both unreliable links as well as adversarial updates. The results provide prescriptive advice on the use of FL in mobile, IoT, and vehicular networks that have limited reliability and security requirements and provide future directions such as privacy-conscious robust aggregation, fairness-conscious weighting, and testbed implementation.

Zitieren

APA

Zeng, Z., Yang, S., & Ding, G. (2025). Robust aggregation algorithms for federated learning in unreliable network environments. Journal of Computing and Electronic Information Management, 18(3), 34-42. https://doi.org/10.54097/n0dpaf43

Vancouver

Zeng Z, Yang S, Ding G. Robust aggregation algorithms for federated learning in unreliable network environments. Journal of Computing and Electronic Information Management. 2025;18(3):34-42. doi:10.54097/n0dpaf43

BibTeX
@article{Zeng2025,
  author = {Zeng, Ziyang and Yang, Shiyu and Ding, Guanyu},
  title = {Robust aggregation algorithms for federated learning in unreliable network environments},
  journal = {Journal of Computing and Electronic Information Management},
  year = {2025},
  volume = {18},
  number = {3},
  pages = {34--42},
  doi = {10.54097/n0dpaf43}
}
Kostenlos starten →