Recent Applications of Machine Learning and Deep Learning in the Development of Superconductors

Machine learning and deep learnings are recently counted as state-of-the-art tools in the search for new superconducting materials and the understanding of their properties, accelerate discoveries by analyzing vast datasets, predicting new materials, and optimizing experimental processes. In this work, having studied numerous recent papers, the applications of the machine learning and deep learnings in the development of superconductors are investigated. According to the obtained results, predicting superconducting transition temperatures from existing databases are conducted by Random Forest, Super Vector Machine, and Neural Networks algorithms, while Bayesian algorithm is utilized for screening of large-scale material databases. Meanwhile, the generative models of the deep learning, such as Generative Adversarial Networks and Variational Autoencoders, are applied to generate new hypothetical superconductors with optimized compositions and crystal structures, and to suggest compositions likely to exhibit superconductivity. In most of studies, the neural networks are used to improve material quality, predict electron-phonon coupling strength, provide faster predictions, analyze spectroscopic and transport measurements, monitor the performance of superconducting magnets, predict failures, and optimize maintenance.