Private synthetic data generation methods allow generating data that are statistically similar to sensitive health data, while ensuring the anonymity of the data subjects. The anonymity can be guaranteed using differential privacy. The approach provides one of the fundamental building blocks for secure use of health data. This project will make private synthetic data generation practical by addressing a number of key weaknesses: improving accuracy of the data under strong privacy, and developing methods to help verify that the generated data actually have the claimed privacy properties. The developed methods that address these will be incorporated in the Twinify open source package developed in our research group.

2024

2026