Extensible Structure-Informed Prediction of Formation Energy with improved accuracy and usability employing neural networks

Machine learning-assisted materials development and device management in batteries and supercapacitors: performance comparison and challenges

“…Krajewski et al optimized an ANN model to predict the formation energies of structures by utilizing 271 features, including elemental and crystal structural attributes derived from Voronoi tessellations. 104 Their model achieved a test set MAE of 30 meV per atom after removing less-stable structures with formation energies above 250 meV per atom, and an MAE of 41 meV per atom when applied to a subset of more complex structures. This demonstrated the increased predictive accuracy and generalizability achieved by ML models by restricting the chemical space of the dataset.…”

“…The stability of a material depends on fundamental thermodynamic properties, such as formation energy, that rely on computationally intensive DFT calculations to determine. 104 The major challenge for developing ML models capable of predicting formation energies is in representing crystal structure data and interatomic interactions in a suitable input format. 105 MAE of 30 meV per atom aer removing less-stable structures with formation energies above 250 meV per atom, and an MAE of 41 meV per atom when applied to a subset of more complex structures.…”

Reflections on one million compounds in the open quantum materials database (OQMD)

“…On an even larger scale, OQMD data has found widespread employment in the training and validation of newly developed ML models. Focusing on formation energy prediction, Krajewski et al employed the compositional and structural features introduced by Ward et al [24,25] and tested several neural network architectures, identifying the best-performing ones and developing an open sources tool with a user interface [75]. Including more generic structural information, Jain et al trained a representation learning feed-forward neural network on the 20 most common structure types on the OQMD exploiting exclusively atomic number and crystallographic symmetry information [76], and Jørgensen et al developed a message passing neural network based only on local information about bonding and symmetry [77].…”

The ScholarNet and Artificial Intelligence (AI) Supervisor in Material Science Research

“…In recent years, the wide-ranging implementation of artificial intelligence (AI) and robotics technology in material science has been evident. − These applications primarily focus on material property prediction and reaction optimization, − enabling high-throughput experimentation and efficient optimization algorithms that significantly enhance the efficiency of chemical experiments . However, a crucial challenge lies in the fact that these applications are limited to the execution phase of scientific research, serving as lab technicians rather than contributing to the generation of research ideas.…”