Figure 5.

First-order proximity. The normalized graph distances between the billing code networks learned from real and synthetic data with respect to first-order proximity (FOP). We compute the graph distances in 4 settings: real vs real, medGAN vs real, medWGAN vs real, medBGAN vs real and EMR-WGAN vs real. We sort the generative models according to the normalized distance values. BGAN: boundary-seeking generative adversarial network; EMR: electronic medical record; GAN: generative adversarial network; WGAN: Wasserstein generative adversarial network.