EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

PFedKAN: A New Paradigm for FedKAN Client Selection Based on Non‐iid

Zhi Jiang, Yun-Tian Du, Xue-Bin Chen, Chang-Yin Luo

International Journal of Intelligent Systems · 2026-01 · 2026 (1)

DOI: 10.1155/int/2675763 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Intelligent Systems, online erschienen am 22.07.2026. Verfasst von Zhi Jiang und 3 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Federated learning (FL) is an emerging distributed architecture in which multiple clients collaborate to solve machine learning problems under the coordination of a central aggregator. However, FL systems face challenges such as device heterogeneity and data heterogeneity. Traditional client selection methods aim to enhance the role of high‐response devices in the model training process, while reducing the role of low‐response devices or even ignoring them entirely. Nevertheless, this approach may cause the performance of the global model to be biased toward high‐response clients. Therefore, we innovatively propose a new paradigm for client selection: aggregation–selection–aggregation (ASA). Built based on traditional aggregation methods, this paradigm first screens out and eliminates clients that have a negative impact on the global model and then performs aggregation again. The ASA paradigm features a simple structure, ease of implementation, and compatibility with most FL algorithms. We apply the ASA paradigm to the field of client selection in FL and propose a Paradigm‐based Federated KAN (PFedKAN) method for FL client selection, which mitigates the adverse effects caused by traditional client selection methods. Meanwhile, KAN is a new type of neural network architecture, boasting advantages such as a small number of parameters and strong capability in handling high‐dimensional data. Currently, research on applying KAN to FL is still in its infancy. To the best of our knowledge, this work represents the first KAN‐based client selection method for FL, as well as the first study to propose a FedKAN optimization algorithm specifically targeting non‐iid data. Our research expands the application scenarios of KAN in the field of FL. Experimental results demonstrate that compared with various traditional client selection methods, PFedKAN achieves significant improvements in both the performance and convergence speed of the global model.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Jiang, Z., Du, Y., Chen, X., & Luo, C. (2026). PFedKAN: A New Paradigm for FedKAN Client Selection Based on Non‐iid. International Journal of Intelligent Systems, 2026(1). https://doi.org/10.1155/int/2675763

Vancouver

Jiang Z, Du Y, Chen X, Luo C. PFedKAN: A New Paradigm for FedKAN Client Selection Based on Non‐iid. International Journal of Intelligent Systems. 2026;2026(1). doi:10.1155/int/2675763

BibTeX
@article{Jiang2026,
  author = {Jiang, Zhi and Du, Yun-Tian and Chen, Xue-Bin and Luo, Chang-Yin},
  title = {PFedKAN: A New Paradigm for FedKAN Client Selection Based on Non‐iid},
  journal = {International Journal of Intelligent Systems},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1155/int/2675763}
}
Kostenlos starten →