This article discusses a self-supervised probabilistic model (SSPM) designed to explore and discover shape memory alloys (SMAs) by leveraging machine learning. The model autonomously learns atomic representations from unlabeled crystal structures, enhancing the accuracy of downstream predictions about material properties. The SSPM’s effectiveness is demonstrated through the identification of new SMA compounds, including MgAu. The approach combines graph neural networks and probabilistic modeling to predict stable material compositions and crystal structures, offering a more efficient method for discovering new materials.
For more details, you can access the full article here: