摘要
We present a deep-learning framework,CrysXPP,to allow rapid and accurate prediction of electronic,magnetic,and elastic properties of a wide range of materials.CrysXPP lowers the need for large property tagged datasets by intelligently designing an autoencoder,CrysAE.The important structural and chemical properties captured by CrysAE from a large amount of available crystal graphs data helped in achieving low prediction errors.Moreover,we design a feature selector that helps to interpret the model’s prediction.Most notably,when given a small amount of experimental data,CrysXPP is consistently able to outperform conventional DFT.A detailed ablation study establishes the importance of different design steps.We release the large pre-trained model CrysAE.We believe by fine-tuning the model with a small amount of property-tagged data,researchers can achieve superior performance on various applications with a restricted data source.
