Skip to main content Skip to main navigation


DP-CTGAN: Differentially Private Medical Data Generation Using CTGANs

Mei Ling Fang; Devendra Singh Dhami; Kristian Kersting
In: Martin Michalowski; Syed Sibte Raza Abidi; Samina Abidi (Hrsg.). Artificial Intelligence in Medicine - 20th International Conference on Artificial Intelligence in Medicine. Conference on Artificial Intelligence in Medicine (AIME-2022), June 14-17, Halifax, NS, Canada, Pages 178-188, Lecture Notes in Computer Science, Vol. 13263, Springer, 2022.


Generative Adversarial Networks (GANs) are an important tool to generate synthetic medical data, in order to combat the limited and difficult access to the real data sets and accelerate the innovation in the healthcare domain. Despite their promising capability, they are vulnerable to various privacy attacks that might reveal information of individuals from the training data. Preserving privacy while keeping the quality of the generated data still remains a challenging problem. We propose DP-CTGAN, which incorporates differential privacy into a conditional tabular generative model. Our experiments demonstrate that our model outperforms existing state-of-the-art models under the same privacy budget on several benchmark data sets. In addition, we combine our method with federated learning, enabling a more secure way of synthetic data generation without the need of uploading locally collected data to a central repository.

Weitere Links