DP-CGANs (Differentially Private Conditional Generative Adversarial Networks)

DP-CGANs (Differentially Private Conditional Generative Adversarial Networks) is an open-source Python library for generating synthetic data that protects individual privacy. It was designed with personal health data in mind. Such data is highly valuable for research and for developing AI models, but sharing it is restricted because of privacy risks. Synthetic data consists of artificial records that preserve the statistical patterns of the original data, so it can be shared more safely. However, synthetic data can still reveal information about real people unless it is generated with formal privacy protection. DP-CGANs combines synthetic data generation with differential privacy to address this. It was developed at Maastricht University, and the approach is described in a paper in the Journal of Biomedical Informatics.

The library uses a conditional generative adversarial network (GAN). In a GAN, two neural networks are trained against each other: one generates synthetic records, while the other tries to tell them apart from real ones. DP-CGANs builds on the widely used CTGAN model and modifies it to better capture the relationships between variables in the data. Differential privacy can be enabled during training, adding calibrated noise so that no single person's data has a strong influence on the model.

DP-CGANs works with tabular data, such as CSV files, and with RDF data, a standard format for linked data. Users can run it from the command line or in Python scripts. The library is still under development and is released under the MIT licence.

Auto-discovered on 2026-09-23 by OECD Catalogue Automation