For Heads-of · Practitioner
Differential Privacy
A mathematical framework ensuring that algorithms protect individual privacy by adding carefully calibrated noise, making it impossible to infer whether a specific individual's data was used.
- differential privacy
- privacy
- noise
- statistical guarantees
Guarantee
An output of a differentially private algorithm should be almost the same whether a specific individual's data is included or not, guaranteeing limited privacy leakage.
Tradeoff
Differential privacy requires adding noise, reducing model accuracy. There is a tradeoff between privacy and utility that organizations must navigate.
Application
Differential privacy is valuable in federated learning, when publishing aggregate statistics, and when organizations want to certify privacy guarantees to regulators or users.