Table 1 Novelty check of generated samples by GANs.

From: Generative adversarial networks (GAN) based efficient sampling of chemical composition space for inverse design of inorganic materials

  GAN-OQMD GAN-MP GAN-ICSD
Training sample # 251,368 57,530 25,323
Leave-out sample # 27,929 6392 2813
Generated sample # 2,000,000 2,000,000 2,000,000
Recovery % of training samples 60.26% 47.36% 59.54%
Recovery % of leave-out sample 60.43% 48.82% 60.13%
New samples 1,831,648 1,969,633 1,983,231