Researchers from MIT have developed a machine learning approach that could replace the intuition-based methods typically used in the creation of catalysts. The team, led by graduate student Xiaochen Du, devised a system that offers more detailed insights than conventional techniques, identifying previously undiscovered atomic configurations in a material that had been researched for three decades. The machine learning model begins by examining a single clean surface, then uses an active learning system with a Monte Carlo algorithm to help select sites to examine on that surface. This method reduces the number of calculations required compared to the multivariate approaches currently in use. The model, named the Automatic Surface Reconstruction framework, requires very little human input and represents a transformation of previous intuition-led processes. The researchers have made the system openly accessible and hope that it will be employed in the development of materials for new catalysts, battery components, and fuel-cell parts.
Skip to content
Skip to footer
