SGFL: A Federated Learning Approach for Non-IID Data Using Semi-Supervised DCGAN
AliReza Rabiee, Abolfazl Ajdarloo, Mohsen Rahmani · 2023
Federated learning (FL) has been introduced as a method of cooperative machine learning models on decentralized machines. However, the assumption of independently and identically distributed data (IID) is often not true in real scenarios because the data distribution can vary significantly from device to device. This paper introduces SGFL, a novel Federated Learning approach designed specifically to address the problems posed by non-IID data distributions. SGFL employs the power of semi-supervised learning and deep convolutional generative adversarial networks (DCGANs) to enhance the federated learning process. By leveraging the redundant data received from each device, SGFL utilizes a semi-supervised DCGAN to fine-tune a global model. This approach enables better modeling of non-IID data while preserving privacy and security through local training. We evaluate the performance of SGFL on non-IID data, and the results show its effectiveness in achieving higher accuracy compared to traditional federated learning methods. The source codes of the methods presented in the paper is available at https://github.com/apaliray03/SGFL.