$ ./guess_the_languageAI & ML
AI & ML · hard · #59

What does differential privacy guarantee about a model trained on personal data?

  1. One person barely changes output
  2. No personal data is ever stored
  3. Each user can remove their records
  4. 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 ▸