Differential Privacy Simulator
Inject Laplace/Gaussian-style noise scaled by the privacy budget epsilon into model coefficients and trace the resulting privacy-utility frontier.
- differential privacy
- privacy-utility tradeoff
- governance
- epsilon
Mean DP AUC
0.778
SD of DP AUC
6.32e-5
True AUC (no DP)
0.778
Utility loss
-1.07e-5
AUC distribution under DP (ε = 1.0)
Privacy-utility frontier
There is no free lunch
Moderate privacy: AUC is degraded but usually still above chance.