LAUSR

Materials informatics has significantly accelerated the discovery and analysis of materials in the last decade. Spectroscopic data provide essential information about materials and hence are widely used for materials analysis. However, data analysis, that is, the extraction of physical parameters, of spectra is often conducted by manually comparing spectra and on-the-fly data analysis has not been realized yet. Considering that more than 100,000 X-ray absorption spectra (XAS) can be measured per day using the scanning transmission X-ray microscopy system at the Photon Factory [1], an automated analysis methodology is urgently required [2, 3]. If physical parameters are to be estimated from the spectra, the on-the-fly analysis can be realized by space mapping of these parameters using a high-throughput spectromicroscopy experiment capable of adaptive measurements [4]. XAS often shows complex spectral features with a few hundred or more “high-dimensional” data points, and the physical parameters, such as element, charge, and symmetry can be extracted from the XAS [5]. Feature extraction, a popular machine learning approach to treat high-dimensional datasets, involves the projection of data onto a few features (parameters), thus retaining the relevant information [6]. In this study, we propose a methodology to estimate the physical parameters from XAS using feature extraction with dimensionality reduction, as shown in Fig. 1.