Second, the applied deep learning method is based on an autoencoder where a combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) is utilized as the autoencoder ...

Additionally, utilizing stacked autoencoder architecture with layer-wise pre-training and hierarchical structures, GMScaleSAE effectively fuses multiscale features, ensuring stable training and ...

We propose a Crystal Diffusion Variational Autoencoder (CDVAE) that captures the physical inductive bias of material stability. By learning from the data distribution of stable materials, the decoder ...