Materials Informatics Laboratory
Staff & Contact
|Educational Staff||Prof. Mikiya Fujii|
Education and Research Activities in the Laboratory
Recent progress in the research filed of the machine learning and artificial intelligence bring us a new filed of materials sciences, called "materials informatics". This is attracting attention as the fourth science following experimental sciences, theoretical sciences, and computing sciences. We will promote studies on materials informatics, aiming to build new principles in material sciences and accelerate materials development.
The basic concept of materials informatics is a closed loop consisting of data collection, construction of learning models, search in design space, and experiments. Here, we will build a learning theory that extracts the essential principles common to multiple materials and multiple physical properties, and enable the essential understanding of materials and prediction of physical properties of new materials. Furthermore, we will develop methodologies to design new materials with desired properties by advancing Bayesian optimization and sequential learning, which are one of the design of experiments, and generative models. In addition to developing these new theories and algorithms, our laboratory will promote joint studies with experimental research groups and demonstrate the acceleration of material development.
1. Learning methodology for multiple materials and multiple physical properties: The recent progresses of the materials informatics in this decade enabled to learn and predict individual properties. We will advance the learning models to multiple materials and multiple physical properties, which resolve limited data problem in the material sciences and promote materials development.
2. Theoretical and computational designing of new materials: Material design with a desired physical property is a milestone in the material sciences. We will develop methodologies for expressing and generating material information leading to the material design based on crystal structures. Here, in addition to the materials informatics, theoretical and computational chemistry techniques are also incorporated to develop the methodologies.
Explanatory Pictures of Research Activities
Fig. 1: Basic concept of Materials Informatics that is a closed loop consisting of data collection, construction of learning models, search in design space, and experiments.
Fig. 2: Schematic view of (a) learning and predicting physical properties and (b) generative models (e.g., autoencoder) of crystal structure