Experimental Evaluation for Risk Assessment of Privacy Preserving Synthetic Data
摘要
Privacy Preserving Data Synthesis (PPDS) has attracted attention as a means of processing personal data while maintaining privacy and utility. Although the PPDS has already been put to some practical use, a proper privacy risk assessment is essential to protect the rights and interests of individuals. We experimentally evaluate the privacy level of synthetic data using TAPAS (a Toolbox for Adversarial Privacy Auditing of Synthetic Data). In particular, we examine the causes and remedies for the gap between theoretical and experimental levels of privacy. It is expected that an understanding of the gap will help to identify theoretical errors and bugs in a PPDS method/code.