Differential privacy
Differential privacy: A formal guarantee limiting how much one individual’s presence in data can influence the result. A randomized mechanism makes output distributions similar for datasets differing by one record. The epsilon budget measures the privacy–utility trade-off and must be tracked across operations.
In italiano: Differential PrivacyLegge V · Overfitting, bias e collasso dei modelliA randomized mechanism makes output distributions similar for datasets differing by one record. The epsilon budget measures the privacy–utility trade-off and must be tracked across operations.
Formula
Pr[M(D)∈S] ≤ e^ε · Pr[M(D′)∈S] + δ
Examples
- DP-SGD
- Statistics with calibrated noise
- Cumulative epsilon budget