Accelerating the prediction of remanent polarization in multicomponent ferroelectrics by using variational autoencoder-based data augmentation

Using a variational autoencoder to expand the dataset to 20 000, six key descriptors for the remanent polarization of the (K 1− x − y − z Na x Ba y Ca z )(Nb 1− u − v − w Zr u Ti v )O 3 system were identified, enabling accurate predictions for new experimental systems.