Recent progress of machine learning opens a possibility of data-centric approaches to design and fabrication of new materials. In order to construct automatic systems to assist designing and developing novel functional materials, this research aims to open a field of informatics especially based on material structures to establish methods to estimate materials functions from their observed images in high-speed and high-accuracy.
This research focuses on the following three objectives:
Extraction of correlation between structure and function
Machine learning is employed to extract structural features involving materials functions from observed images. While machine learning is successful in feature extraction for natural images, obtaining structural features correlated to materials functions is a challenging problem.
Proposal of optimal observation method
Feedback control of observation equipment with adaptive construction of a model for an observed sample in real-time makes observation cost minimal.
Shortening learning time and improving estimation accuracy
The long learning time and low estimation accuracy of machine learning is the bottleneck for real-time use of the constructed systems. Fast and highly accurate algorithms are required to be constructed from mathematical viewpoint.