What does differential privacy guarantee about a model trained on personal data?
- One person barely changes output
- No personal data is ever stored
- Each user can remove their records
- The data is fully encrypted
Calibrated noise bounds how much any single record can affect results, limiting what can be learned about an individual.
Play AI & ML ▸