Creating Machine Learning-Driven Material Recipes Based on Crystal Structure

Abstract

Determining the manner in which crystal structures are formed is considered a great mystery within materials science. Potential solutions have the possibility to be uncovered by revealing hidden patterns within the material data. Data science is therefore implemented in order to link the material data to the crystal structure. In particular, unsupervised and supervised machine learning techniques are used where the Gaussian mixture model is employed in order to understand the data structure of the materials database while random forest classification is used to predict the crystal structure. As a result, the unsupervised and supervised machine learning techniques reveal descriptors for determining the crystal structure via the materials database. In addition, predicting atomic combinations from the crystal structure is also achieved using a trained machine where the first-principles calculations confirm the stability of predicted materials. Thus, one can consider that the estimation of the crystal structure can be achieved in principle via the combination of material data and machine learning, thereby leading to the advancement of crystal structure prediction.

Machine Learning for Computational Heterogeneous Catalysis

“…Ward et al [64] combined composition-based descriptors with Voronoi tessellation derived attributes, employed OQMD as training data and decision tree model to map formation enthalpies with mean absolute error of 80 meV/atom. Takahashi et al [118] utilized the Gaussian mixture model and random forest classification to reveal the important descriptors that determine the crystal structure of single and binary compounds in OQMD, based on descriptor importance rankings. Almost at the same time, Seko et al [119] proposed to use descriptive statistics as descriptorsto represent a compound for predicting cohesive energies.…”

QM-symex, update of the QM-sym database with excited state information for 173 kilo molecules

“…Another vital application of accelerated development is artificial intelligence. Checking the excited-state properties of each molecule experimentally is time and energy consuming, and thus the use of quantum mechanical computation (QM) or machine learning algorithm (ML) is necessary in enabling scholars to study the structure and properties of material molecules more efficiently [11][12][13][14][15][16][17] and to compile large databases. However, quantum mechanical computation and machine learning algorithms, especially neural networks, are able to come up with relatively good performance only if large databases are utilized in training and debugging models.…”

Machine Learning in Magnetic Materials

“…Machine learning (ML) has been succesfully applied in screening and processing of large databases with materials data with the aim to detect cases with potential for use in various applications. [62][63][64][65][66][67][68][69][70][71][72][73][74] There are relatively few published works in utilizing machine learning targeting magnetic materials compared to other disciplines but there is increasing interest in the field, for the reasons we have already mentioned.…”