Which privacy-preserving ML technique adds noise to protect individual data?

• A. Federated learning
• B. Encryption in use
• C. Differential privacy
• D. Secure multiparty computation

Answer: (C)

Differential privacy protects individual data by adding calibrated randomness to the results of computations or released statistics. This noise makes it difficult to infer whether a specific person’s data was included, while still allowing accurate insights about the overall population. The amount of noise is carefully tuned (often controlled by a privacy parameter) to balance privacy and accuracy, so individual records remain protected as a whole.

Other privacy-preserving ML techniques take different approaches. Federated learning keeps data on local devices and only shares model updates, reducing data centralized exposure but not by injecting noise. Encryption in use protects data during processing without altering the results with noise. Secure multiparty computation uses cryptographic methods to compute results without revealing inputs, rather than masking individual contributions with noise.