Belt and Braces: When Federated Learning Meets Differential Privacy.

This article explores the combination of federated learning (FL) and differential privacy (DP) in machine learning to allow for minimal raw data exposure and enhanced data privacy. The article provides an overview of both as well as a look at centralized and distributed DP for FL. Tools and platforms are examined with an emphasis on several areas including clipping-bound estimation and privacy-loss composition. Lastly, challenges including vertical/transfer federation and robustness are discussed.