EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Federated Learning with Asynchronous Aggregation and Differential Local Perturbation: Reconciling Utility, Efficiency, and Privacy

Javier Conti, Nils Schwarz

International Journal of Computational and Biological Sciences · 10.05.2026 · 3 (2) · 46-55

DOI: 10.66238/ijcbs81 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Computational and Biological Sciences, online erschienen am 10.05.2026. Verfasst von Javier Conti und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Federated learning has emerged as a transformative paradigm for training machine learning models across decentralized data sources while mitigating direct data exposure. However, traditional synchronous federated learning faces severe challenges in heterogeneous network environments, commonly suffering from straggler effects that drastically reduce training efficiency. Furthermore, while data remains localized, shared model updates remain susceptible to advanced inference and inversion attacks. This paper investigates a novel framework combining asynchronous aggregation with differential local perturbation to simultaneously address system efficiency and data privacy. We propose a methodology where individual clients independently perturb their local updates using dynamic noise calibration before transmitting them to the central server, which updates the global model asynchronously. By integrating staleness-aware weighting with rigorous local privacy mechanisms, the proposed framework balances the privacy-utility trade-off and mitigates the detrimental impacts of delayed updates. Extensive simulated evaluations demonstrate that this dual-faceted approach significantly improves convergence rates in heterogeneous networks compared to synchronous baselines while providing quantifiable privacy guarantees against gradient leakage. The findings underscore the viability of asynchronous federated learning in highly volatile environments where strict privacy constraints are mandatory, offering a scalable solution for privacy-preserving distributed intelligence

Zitieren

APA

Conti, J., & Schwarz, N. (2026). Federated Learning with Asynchronous Aggregation and Differential Local Perturbation: Reconciling Utility, Efficiency, and Privacy. International Journal of Computational and Biological Sciences, 3(2), 46-55. https://doi.org/10.66238/ijcbs81

Vancouver

Conti J, Schwarz N. Federated Learning with Asynchronous Aggregation and Differential Local Perturbation: Reconciling Utility, Efficiency, and Privacy. International Journal of Computational and Biological Sciences. 2026;3(2):46-55. doi:10.66238/ijcbs81

BibTeX
@article{Conti2026,
  author = {Conti, Javier and Schwarz, Nils},
  title = {Federated Learning with Asynchronous Aggregation and Differential Local Perturbation: Reconciling Utility, Efficiency, and Privacy},
  journal = {International Journal of Computational and Biological Sciences},
  year = {2026},
  volume = {3},
  number = {2},
  pages = {46--55},
  doi = {10.66238/ijcbs81}
}
Kostenlos starten →