Optimal Differentially Private Model Training with Public Data

Andrew Lowy, Zeman Li, Tianjian Huang +1

Differential privacy (DP) ensures that training a machine learning model does not leak private data. In practice, we may have access to auxiliary public data that is free of privac…