Materials informatics for 2d materials combined with sparse modeling and chemical perspective: Toward small-data-driven chemistry and materials science

Application of data-scientific approaches to conventional sciences, such as chemo-informatics, bio-informatics, and materials informatics (MI), has attracted much interest toward datadriven research. The concept enables accelerated discovery of new materials, enhancement of performance, and optimization of processes. However, sufficient bigdata is not always prepared to apply to machine learning. For example, experimental scientists have their own small data including success and failure in their laboratory, whether in academia or industry. If such small data is effectively utilized with a data-scientific approach, research activities can be accelerated without energy, resource, and cost consumption. This account focuses on MI for small data, a recent concept for application of small data, with introduction of model cases, such as control of exfoliation processes to obtain 2D materials. Combination of machine learning and chemical perspective is effective for construction of straightforward and interpretable predictors through the extraction of a limited number of descriptors from small dataset. Although the prediction accuracy is not so precise, the model has enough accuracy to be a guideline reducing the number of the next experiments. The present MI for small data opens potentials of small-data-driven chemistry and materials science.