Federated Learning Framework for Edge Devices with Reducing Communication Network Costs and Enhancing Performance

Authors : Sercan Yalçın

Pages : 11-18

Doi:10.53070/bbd.1626847

View : 32 | Download : 30

Publication Date : 2025-06-01

Article Type : Research Paper

Abstract :In this study, the effectiveness of the proposed federation learning method is evaluated through some experiments conducted on the MNIST dataset. The proposed method aims to significantly reduce the communication costs while increasing the model accuracy thanks to a hierarchical approach and adaptive weighting mechanism. The experimental results show that the proposed method significantly shortens the training time and reduces the communication cost. Especially, in scenarios where edge devices with different computational power are present in the network environments, the presented method showed better performance. It was also observed that the proposed method increased the model accuracy and provided better generalization ability to the models. The findings obtained in this study show that the proposed federation learning model is an effective and efficient solution for model training in edge computing systems. This method is considered to have great potential especially for applications where communication bandwidth is limited and privacy is important. In future studies, it is planned to evaluate the performance of this method on different datasets and more complex model architectures.
Keywords : Federasyonlu öğrenme, uç bilişim, iletişim ağları, evrişimli sinir ağları

ORIGINAL ARTICLE URL