Function smartnoise_runtime::utilities::mechanisms::gaussian_mechanism

pub fn gaussian_mechanism(
    value: f64, 
    epsilon: f64, 
    delta: f64, 
    sensitivity: f64, 
    analytic: bool, 
    enforce_constant_time: bool
) -> Result<f64>

Returns noise drawn according to the Gaussian mechanism.

If using the standard guassian, Let c = sqrt(2ln(1.25/delta)). Noise is drawn from a Gaussian distribution with scale sensitivityc/epsilon and centered about 0. For more information, see the Gaussian mechanism in C. Dwork, A. Roth The Algorithmic Foundations of Differential Privacy, Chapter 3.5.3 Laplace versus Gauss p.53. August 2014.

If using the analytic gaussian, the noise scale is derived using Balle (2018).

NOTE: this implementation of Gaussian draws in likely non-private due to floating-point attacks See Mironov (2012) for more information on a similar attack of the Laplace mechanism.

Arguments

• value - Statistic to be privatized.
• epsilon - Multiplicative privacy loss parameter.
• delta - Additive privacy loss parameter.
• sensitivity - Upper bound on the L2 sensitivity of the function you want to privatize.
• enforce_constant_time - Whether or not to enforce the algorithm to run in constant time

Return

A draw from Gaussian distribution with scale defined as above.

Examples

use smartnoise_runtime::utilities::mechanisms::gaussian_mechanism;
let n = gaussian_mechanism(22.3, 0.1, 0.0001, 2.0, false, false);