Agenda

Résumé : Despite advances in high throughput calculations, artificial intelligence, and automated synthesis, the rate of material discovery remains the same since 1960s, hindering progress in addressing the global challenges in energy, sustainability and healthcare. One of the major bottlenecks remains the identification of new structural families with targeted properties, as opposed to variation in composition of well-established structure types. Another challenge is the discovery of materials with rare or unconventional properties, for which only small datasets and limited theoretical understanding exist. Additionally, the lack of an efficient feedback loop between failed experiments and theoretical predictions impedes progress.
Databases containing experimental results are much rarer and typically smaller than theoretical datasets. We aggregated a database of ~4000 Curie temperatures, and analyzed data trends to identify and then experimentally discover new ferromagnetic materials. I will discuss how natural language processing can rapidly expand the dataset and improve machine learning prediction of the Curie temperature. I will also illustrate how candidate materials with a rare property such as spin-triplet superconductivity can be identified despite the limited number of examples. Finally, I will provide a quantitative example of experimental verification of high-throughput calculations.