Generative data augmentation with differential privacy for non-IID problem in decentralized clinical machine learning

Tianyu He, Peiyi Han, Shaoming Duan, Zirui Wang, Wentai Wu, Chuanyi Liu, Jianrun Han

Abstract: Highlights•A novel GAN-based data augmentation framework proposed to tackles the non-IID problem in swarm learning (SL).•The proposed method co-trains a differential GAN with the local target task model in SL.•We theoretically prove the convergence of the proposed GAN-based method over non-IID data in SL.•We provide the privacy guarantee for our method.•Extensive experiments are performed using three real-world clinical datasets.