Accurate prediction of material properties, largely facilitated by machine learning, is crucial for discovering novel materials with tailored functionalities, yet traditional deep learning models require large annotated datasets. Here, the authors introduce a self-supervised pretraining approach using surrogate labels, achieving up to 6.67% improvement in prediction accuracy, thus setting a benchmark in material property predictions and enhancing methodological robustness.
- Chowdhury Mohammad Abid Rahman
- Aldo H. Romero
- Prashnna K. Gyawali
