Optimal Differentially Private Model Training with Public Data

arXiv:2306.15056v2 Announce Type: replace
Abstract: 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 privacy concerns. In this work, we assume access to a given amount of public data and settle the following fundamental open questions: 1. What is the optimal (worst-case) error of a DP model trained over a private data set while having access to side public data? …

abstract arxiv concerns cs.cr cs.lg data differential differential privacy free leak machine machine learning machine learning model math.oc practice privacy private data public public data stat.ml training type work