Fig. 3: Inorganic materials space composed of existing ICSD materials and hypothetical materials generated by GAN-ICSD. | npj Computational Materials

Fig. 3: Inorganic materials space composed of existing ICSD materials and hypothetical materials generated by GAN-ICSD.

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

Fig. 3

The two axes correspond to the two dimensions after t-sne-based dimension reduction. The ICSD materials only occupies a tiny portion of the chemical space of inorganic materials. a Training samples (green dots) and leave-out validation samples (red dots) from ICSD; b 50,000 generated samples (blue dots) together with training and leave-out samples; c 200,000 generated samples together with training and leave-out samples.

Back to article page