As one of the most studied materials, perovskites exhibit a wealth of superior properties that lead to diverse applications. Computational prediction of novel stable perovskite structures has big potential in the discovery of new materials for solar panels, superconductors, thermal electric, and catalytic materials, etc. By addressing one of the key obstacles of machine learning based materials discovery, the lack of sufficient training data, this paper proposes a transfer learning based approach that exploits the high accuracy of the machine learning model trained with physics-informed structural and elemental descriptors. This gradient boosting regressor model (the transfer learning model) allows us to predict the formation energy with sufficient precision of a large number of materials of which only the structural information is available. The enlarged training set is then used to train a convolutional neural network model (the screening model) with the generic Magpie elemental features with high prediction power. Extensive experiments demonstrate the superior performance of our transfer learning model and screening model compared to the baseline models. We then applied the screening model to filter out promising new perovskite materials out of 21,316 hypothetical perovskite structures with a large portion of them confirmed by existing literature.

Computational prediction of crystal materials properties can help to do large-scale insiliconscreening. Recent studies of material informatics have focused on expert design of multidimensionalinterpretable material descriptors/features. However, successes of deep learning suchas Convolutional Neural Networks (CNN) in image recognition and speech recognition havedemonstrated their automated feature extraction capability to effectively capture the characteristicsof the data and achieve superior prediction performance. Here, we propose CNN-OFM-Magpie, aCNN model with OFM (Orbital-field Matrix) and Magpie descriptors to predict the formationenergy of 4030 crystal material by exploiting the complementarity of two-dimensional OFM featuresand Magpie features. Experiments showed that our method achieves better performance thanconventional regression algorithms such as support vector machines and Random Forest. It is alsobetter than CNN models using only the OFM features, the Magpie features, or the basic one-hotencodings. This demonstrates the advantages of CNN and feature fusion for materials propertyprediction. Finally, we visualized the two-dimensional OFM descriptors and analyzed the featuresextracted by the CNN to obtain greater understanding of the CNN-OFM model.

Superconductors have been one of the most intriguing materials since they were discovered more than a century ago. However, superconductors at room temperature have yet to be discovered. On the other hand, machine learning and especially deep learning has been increasingly used in material properties prediction and discovery in recent years. In this paper, we propose to combine the deep convolutional neural network (CNN) model with fully convolutional layers for feature extraction with gradient boosting decision tree (GBDT) for superconductors critical temperature (T c) prediction. Our prediction model only uses the elemental property statistics of the materials as original input and learns a hierarchical representation of superconductors using convolutional layers. Computational experiments showed that our convolutional gradient boosting decision tree (ConvGBDT) model achieved the state-of-the-art results on three superconductor data sets: DataS, DataH, and DataK. By visually comparing the raw elemental feature distribution and the learned feature distribution, it is found that the convolutional layers of our ConvGBDT can learn features that can more effectively distinguish cuprate and iron-based superconductors. On the other hand, the GBDT part of our ConvGBDT model can learn the sophisticated mapping relationship between extracted features and the critical temperatures to obtain good prediction performance.

In this paper, a hybrid neural network (HNN) that combines a convolutional neural network (CNN) and long short-term memory neural network (LSTM) is proposed to extract the high-level characteristics of materials for critical temperature (Tc) prediction of superconductors. Firstly, by obtaining 73,452 inorganic compounds from the Materials Project (MP) database and building an atomic environment matrix, we obtained a vector representation (atomic vector) of 87 atoms by singular value decomposition (SVD) of the atomic environment matrix. Then, the obtained atom vector was used to implement the coded representation of the superconductors in the order of the atoms in the chemical formula of the superconductor. The experimental results of the HNN model trained with 12,413 superconductors were compared with three benchmark neural network algorithms and multiple machine learning algorithms using two commonly used material characterization methods. The experimental results show that the HNN method proposed in this paper can effectively extract the characteristic relationships between the atoms of superconductors, and it has high accuracy in predicting the Tc.Symmetry 2020, 12, 262 2 of 13 same time, the prediction of the Tc of superconductors, especially high-temperature superconductors, is not very accurate. The solution of these problems depends on the discovery of superconductors or similar materials and an understanding of the physical properties of these materials. Although this was the focus of research for the past 30 years, the prediction of Tc of superconductors is still very difficult.Advances in computers, as well as the development and continuous improvement of first-principles computational quantum chemistry theories and statistical (or machine learning) methods, greatly influenced research activities related to material discovery and design [12]. Material design for high-throughput (HT) calculations made progress in determining the structure of thousands of inorganic solids [13,14]. Since 1970, density functional theory (DFT) is widely used in the calculation of solid-state physics. In most cases, compared with other methods for solving multi-body problems in quantum mechanics, DFT using local density approximation gives very satisfactory results, and solid-state computing is less expensive than experiments. DFT is the leading method for the calculation of electronic structures in various fields; however, these methods are currently not suitable for high-level calculations due to the high-cost calculation. When using standard exchanges and correlation functions such as Perdew-Berke-Ernzerhof (PBE) [15] which is currently the most widely used exchange-related functional form in the calculation of solid structures, there are cases where the system is underestimated compared to the experimental values.In addition to prediction models based on physical principles/theories, the machine learning [16-21] approach for Tc prediction is a data-driven prediction model, which exploits the relationship between material ...

No abstract