EUVIMEDEUVIMED entdecken
Crossref

FedHCG: A Byzantine Robust Federated Learning Framework With Hierarchical Confidence Estimation and Geometric Consistency

Xiaoting Sun, Zhong Li, Wenhao Guo, Changjun Jiang

IEEE Transactions on Cognitive Communications and Networking · 2026 · 12 · 9529-9543

DOI: 10.1109/tccn.2026.3704844 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Cognitive Communications and Networking, 2026. Verfasst von Xiaoting Sun und 3 weiteren Autor:innen. Die Arbeit nennt 44 Literaturangaben. Gefördert von: National Key Research and Development Program of China, Discipline Breakthrough Pioneer Program of the Ministry of Education of China, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Federated learning (FL) has become a cornerstone of user-centric intelligence in large-scale Internet of Things (IoT) networks, where heterogeneous devices such as mobile users, sensors, and edge nodes collaboratively train shared models. However, the openness of FL exposes it to Byzantine participants that may upload corrupted updates, thereby degrading global performance. Existing Byzantine defenses fall into two categories: detection and tolerance, but both remain limited. Detection methods typically rely on distance-based heuristics that overlook hierarchical sensitivity and dynamic user trust. While tolerance methods often ignore the low-dimensional manifold structure of user updates, leading to reduced discriminability under complex attacks. To address these challenges, we propose FedHCG, a Byzantine robust FL framework that introduces reliable user selection based hierarchical confidence estimation (HCE) and geometric consistency (GC) based aggregation. The HCE quantifies each user’s hierarchical reliability across temporal and structural layers, enabling adaptive filtering of anomalous participants. The GC embeds user updates into a manifold space and measures their geometric consistency within peer clusters, yielding soft aggregation weights that suppress corrupted updates without complete exclusion. Extensive experiments on the benchmark dataset have shown that our method can improve by 58.86% compared to FedAvg when honest users make up the majority. Even though malicious users dominate, it has improved by 82.1% compared to FedAvg, demonstrating strong robustness. Theoretical analysis further proves its convergence and robustness. Overall, FedHCG provides a principled and scalable foundation for trustworthy, user-centered federated intelligence.

Förderung

  • National Key Research and Development Program of China
  • Discipline Breakthrough Pioneer Program of the Ministry of Education of China
  • National Natural Science Foundation of China
  • Interdisciplinary Frontier Innovation Team Development Special Fund of Donghua University
  • Donghua University 2024 Cultivation Project of Discipline Innovation

Zitieren

APA

Sun, X., Li, Z., Guo, W., & Jiang, C. (2026). FedHCG: A Byzantine Robust Federated Learning Framework With Hierarchical Confidence Estimation and Geometric Consistency. IEEE Transactions on Cognitive Communications and Networking, 12, 9529-9543. https://doi.org/10.1109/tccn.2026.3704844

Vancouver

Sun X, Li Z, Guo W, Jiang C. FedHCG: A Byzantine Robust Federated Learning Framework With Hierarchical Confidence Estimation and Geometric Consistency. IEEE Transactions on Cognitive Communications and Networking. 2026;12:9529-9543. doi:10.1109/tccn.2026.3704844

BibTeX
@article{Sun2026,
  author = {Sun, Xiaoting and Li, Zhong and Guo, Wenhao and Jiang, Changjun},
  title = {FedHCG: A Byzantine Robust Federated Learning Framework With Hierarchical Confidence Estimation and Geometric Consistency},
  journal = {IEEE Transactions on Cognitive Communications and Networking},
  year = {2026},
  volume = {12},
  pages = {9529--9543},
  doi = {10.1109/tccn.2026.3704844}
}
Kostenlos starten →