Differentially Private Synthetic Data via Foundation Model APIs 1: Images

arXiv:2305.15560v2 Announce Type: replace-cross
Abstract: Generating differentially private (DP) synthetic data that closely resembles the original private data is a scalable way to mitigate privacy concerns in the current data-driven world. In contrast to current practices that train customized models for this task, we aim to generate DP Synthetic Data via APIs (DPSDA), where we treat foundation models as blackboxes and only utilize their inference APIs. Such API-based, training-free approaches are easier to deploy as exemplified by the recent surge …

aim apis arxiv cs.cr cs.cv cs.lg current data data-driven foundation images practices privacy privacy concerns private private data synthetic synthetic data task train world