Adaptive Differentially Private Empirical Risk Minimization

Xiaoxia Wu, Lingxiao Wang, Irina Cristali, Quanquan Gu, Rebecca Willett

Keywords: Non-convex optimization, Gradient Perturbation, Differentially Private Learning, Adaptive Gradient Methods, Stochastic Gradient Descent, Theory

Abstract: We propose an adaptive (stochastic) gradient perturbation method for differentially private empirical risk minimization. At each iteration, the random noise added to the gradient is optimally adapted to the stepsize; we name this process adaptive differentially private (ADP) learning. Given the same privacy budget, we prove that the ADP method considerably improves the utility guarantee compared to the standard differentially private method in which vanilla random noise is added. Our method is particularly useful for gradient-based algorithms with non-constant learning rate, including variants of AdaGrad (Duchi et al., 2011). We provide extensive numerical experiments to demonstrate the effectiveness of the proposed adaptive differentially private algorithm.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2110.07435/code)