Improving the Privacy and Practicality of Objective Perturbation for Differentially Private Linear Learners

In the arena of privacy-preserving machine learning, differentially private stochastic gradient descent (DP-SGD) has outstripped the objective perturbation mechanism in popularity and interest.