A research group led by Professor Teruyasu Mizoguchi of the Institute of Industrial Science, the University of Tokyo, and Professor Koji Tsuda of the Graduate School of Frontier Sciences, the University of Tokyo, is a new method for interpreting and predicting spectra widely used for material analysis with artificial intelligence. Was developed.The interpretation speed is about 2 times faster than that of experts, and it seems that it can be widely used for semiconductor design, battery development, catalyst analysis, etc.
According to the University of Tokyo, the research group uses machine learning methods used in artificial intelligence to use two mutually correlated dendrograms, a dendrogram of material information and a dendrogram of spectra, to achieve high-speed, high-precision spectra. We have developed a new method that can be interpreted.
The two dendrograms take up the spectra and interpret them while exchanging information. Using this method, substances can be analyzed at high speed and with high accuracy without specialized knowledge.
The spectrum is information obtained from the absorption and emission of incident light, and is used for material analysis in various industrial fields such as semiconductor design.Recently, advances in measurement technology have made it possible to obtain thousands to tens of thousands of spectra in a single experiment.
However, in order to interpret the spectrum and obtain information on the atomic arrangement and electronic structure, researchers had to make full use of advanced expertise and perform theoretical calculations.Since the theoretical calculation of spectra measured using electron beams and X-rays takes several hours to several days, it is practically impossible to analyze a huge number of spectra.
Paper information:[Scientific Reports] Data-driven approach for the prediction and interpretation of core-electron loss spectroscopy
