Generative Adversarial Network (GAN) provide a promising direction in the studies where data availability is limited. However, GANs can easily remeber training samples due to the complexity of deep networks. This becomes a major concern when they are applied to private or sensitive data such as patient medical records. This issue is managed by differentially private GAN (DPGAN) ,where differential privacy is achieved in GANs by adding carefully designed noise to gradients during the learning procedure. DCGAN focuses on preserving the privacy during the training procedure instead of adding noise on the final parameters directly. Thus the noise is added on the gradient of the Wasserstein distance with respect to the training data. The parameters of discriminator show differential privacy with respect to the sample training points.
Input variables : original PHI data
Output Variables : Synthetically generated deidentified data
Visit Model : github.com
Additional links : arxiv.org
| Model Category | : | Public |
| Date Published | : | February, 2018 |
| Healthcare Domain | : |
Payer
Provider |
| Code | : | github.com |
| Data Privacy |
| Synthetic Data Generation |