Torfi et. al. have built a privacy-preserving GAN model using Renyi differential privacy for generating synthetic medical data from real health records and also captures the temporal information and feature correlations that might be present in the original data. This model is composed of convolutional autoencoders and convolutional generative adversarial networks to preserve some of the critical characteristics for the generated synthetic data. the GAN aims to generate high-fidelity synthetic data in a privacy-preserving manner while the autoencoder transforms the input space into a continuous space acting as a bridge between non-continuous data and GAN. Besides, the model uses One Dimensional Convolutional Neural Networks (1D-CNNs) to capture correlated input features’ patterns as well as temporal information. This model outperforms existing state-of-the-art models under the same privacy budget using several publicly available benchmark medical datasets in both supervised and unsupervised settings.
Input variables : health record data
Output Variables : Synthetic data
Visit Model : github.com
Additional links : arxiv.org
| Model Category | : | Public |
| Date Published | : | December, 2020 |
| Healthcare Domain | : |
Payer
Provider |
| Code | : | github.com |
| Data Privacy |
| Synthetic Data Generation |