ContextGAN: Differentially Private Synthetic Data Generation Based on a Context-Aware GAN

ContextGAN is a context-aware differentially private GAN that generates synthetic data by encoding explicit and implicit domain rules through a constraint matrix. Its constraint-aware discriminator evaluates synthetic data according to these rules, and the differential privacy mechanism protects sensitive information in the original data. The method was validated in the healthcare, security and finance domains; the generated data adheres to domain rules, with improvements in both realism and utility.