A Method for Privacy-Safe Synthetic Health Data

Private health records are important for medical research but hard to get because of legal rules. This shortage of data can be solved by using generative models like GANs, which make new, similar data. But GANs might leak private information. To fix this, we made a new kind of GAN with a privacy protection part called DP-ACTGAN. It uses differential privacy to keep the original data safe. We also put a classifier in the GAN to make sure the new data is very close to the real data. Experiments show that DP-ACTGAN can make good quality data without giving away private information. This means we can use data well without breaking privacy, which is good for ethical research and making new things while keeping privacy.