Beyond Laplace and Gaussian: Exploring the Generalized Gaussian Mechanism for Private Machine Learning

ArXi:2506.12553v2 Announce Type: replace Differential privacy (DP) is obtained by randomizing a data analysis algorithm, which necessarily We prove that the full GG family satisfies differential privacy and extend the PRV accountant to privacy loss computation for these mechanisms. We then instantiate the GG mechanism in two canonical private learning pipelines, PATE and DP-SGD. Empirically, we explore PATE and DP-SGD with the GG mechanism across the computationally feasible values of $\beta$: $\beta \in [1,2]$ for DP-SGD and $\beta \in [1,4]$ for.