tff.analytics.differential_privacy.analytic_gauss_stddev

Compute the stddev for the Gaussian mechanism with the given DP params.

Calibrate a Gaussian perturbation for differential privacy using the analytic Gaussian mechanism of [Balle and Wang, ICML'18].

Reference: http://proceedings.mlr.press/v80/balle18a/balle18a.pdf

epsilon Target epsilon (0 < epsilon <= 500). The epsilon value is limited to at most 500 in this implementation because large epsilons causes overflow in python.
delta Target delta (0 < delta < 1).
norm_bound Upper bound on L2 global sensitivity (norm_bound >= 0).
tol Error tolerance for binary search (tol > 0).

sigma Standard deviation of Gaussian noise needed to achieve (epsilon,delta)-DP under the given norm_bound.