2  MedGAN

The original MedGAN article (Choi et al. 2018) was written by Choi et al. in 2017 and descrived for the first time a way to generate realistic synthetic patient records.

The original MedGAN The idea was simple train: an autoencoder on the original data and then use its decoder as part of the generator of a GAN network to gnerate synthetic data.

My initial idea was to modify the input of the network proposed in the original paper to ingest my dataset, maybe play a little with activation functions and optimizers to find the optimal implementation for my use case.

Boy, was I wrong!
Choi, Edward, Siddharth Biswal, Bradley Malin, Jon Duke, Walter F. Stewart, and Jimeng Sun. 2018. “Generating Multi-Label Discrete Patient Records Using Generative Adversarial Networks,” no. arXiv:1703.06490 (January). https://doi.org/10.48550/arXiv.1703.06490.