DP-GenFL: a local differentially private federated learning system through generative data

Conclusion We have provided an enhanced privacy protection mechanism for FL. With the help of GAN, a local DP-FL framework has been proposed, and the privacy level, as well as the convergence performance, have been systemically conducted. Experimental results have shown the effectiveness and the advantages of the proposed algorithm by comparing it with the state of the art. In addition, further directions can find an optimal fake ratio and the number of training epochs for the proposed framework.