Abstract: We investigate practical and scalable algorithms for training machine learning models with user-level differential privacy (DP) in order to provably safeguard all the examples contributed by ...