International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

LightFedSelect: A Lightweight Framework for Byzantine-Robust Federated Learning

Authors :

Seyed Saeed Razavi¹ Seyed Arsalan Vasegh Rahim Parvar² Soroosh Dadashi Pakdeh³ Mohammad Matin Rezaeifard⁴ Morteza Mollaie Chafi⁵ Reza Ebrahimi Atani⁶

1- University of Guilan 2- University of Guilan 3- University of Guilan 4- University of Guilan 5- University of Guilan 6- University of Guilan

Keywords :

Federated Learning،Privacy،Secure aggregation،Adversarial Attacks،Byzantine Attack

Abstract :

Federated Learning (FL) enables collaborative model training across decentralized devices while preserving data privacy. However, its decentralized nature exposes FL to Byzantine attacks (e.g., data/model poisoning) and non-IID data heterogeneity, which degrade model performance and compromise robustness. Existing defenses either incur prohibitive computational overhead or fail to adapt to dynamic adversarial strategies. To address these challenges, we propose LightFedSelect, a lightweight framework that enhances FL robustness through client selection and layered aggregation. Experiments on Fashion-MNIST dataset demonstrate LightFedSelect’s superiority over recent works, achieving 90.52\% accuracy under label-flipping, backdoor, and noise attacks. Our framework reduces computational overhead compared to client-validation schemes and maintains stability in non-IID settings. LightFedSelect bridges the gap between robustness, efficiency, and adaptability, offering a scalable solution for secure FL deployments.