Adds Laplace(0, \(\Delta_1/\epsilon\)) noise, where \(\Delta_1\) is
the L1 sensitivity of the query.
Usage
morie_dp_laplace_mechanism(y, sensitivity = 1, epsilon = 1, seed = NULL)
Arguments
- y
the true query answer; scalar or vector.
- sensitivity
L1 sensitivity of the query, positive.
- epsilon
privacy budget, positive.
- seed
optional integer seed.
Value
list with release, noise_scale, noise_sd
(\(\sqrt{2}b\)), epsilon, sensitivity.
Details
Sensitivity is a property of the QUERY, never of the data in hand.
Estimating it from the dataset is itself a non-private query, and a
sensitivity that is too small does not degrade the guarantee – it
voids it.
References
Dwork, C., McSherry, F., Nissim, K. and Smith, A. (2006).
Calibrating noise to sensitivity in private data analysis.
TCC, 265-284.
Examples
morie_dp_laplace_mechanism(42,
sensitivity = 1, epsilon = 0.5,
seed = 1
)$noise_scale
#> [1] 2
