A Generative Adversarial Networks (Gan) Based Efficient Sampling Method for Inverse Design of Metallic Glasses

Metallic glass has gained a lot of attention due to its unique physical properties. The complicated design space of alloy makes it costly for traditional experimental methods to develop Metallic glass. We proposed a generative adversarial network based sampling model that can rapidly generate a large number of hypothetical MG samples to accelerate the development of metallic glasses. To validate the quality of generated samples, we applied an XGBoost model to validate the hypothetical samples generated by GAN, and the result shows that 95.1% generated samples were judged to be amorphous. In addition, we have validated the extent to which the generated data overlaps with the training data and the proportion of unique samples generated from the perspective of the alloy and alloy system, respectively. Our GAN model is expected to improve the sampling efficiency of metallic glass and thus shorten its development cycle.