Metal hydride composition-derived parameters as machine learning features for material design and H2 storage

Simple local environment descriptors for accurate prediction of hydrogen absorption and migration in metal alloys

“…The development of machine learning (ML) and other surrogate models presents a promising alternative to expensive high-throughput ab initio search platforms. 34,35,52,53 Once sufficiently trained, regression models are many orders of magnitude faster to execute than experiments or traditional ab initio calculations. They can be used to quickly screen materials composition space and discover novel, high-performing candidates.…”

“…34,35 A variety of ML tools have been used to predict the properties of metal alloys, from classical lightweight models like random forests or support vector machines to state-of-the-art deep learning architectures. 53,54 The main drawback of traditionally used ML models is the lack of interpretability and transferability to unknown atomic environments. 12 An alternative approach proposed in this work is based on the identification of physically motivated descriptors (arising from the electronic, atomic, or dynamic traits of the material) that strongly correlate with the target properties.…”