Material Function Estimation Based on Real Space Observation

2021 Fiscal Year Final Research Report

• Project Area: Discrete Geometric Analysis for Materials Design
• Project/Area Number: 17H06469
• Research Category: Grant-in-Aid for Scientific Research on Innovative Areas (Research in a proposed research area)
• Allocation Type: Single-year Grants
• Review Section: Science and Engineering
• Research Institution: Nagoya University
• Principal Investigator: Ichiki Akihisa 名古屋大学, 未来社会創造機構, 特任准教授 (40711156)
• Co-Investigator(Kenkyū-buntansha): 大関 真之 東北大学, 情報科学研究科, 教授 (80447549)
• Project Period (FY): 2017-06-30 – 2022-03-31
• Keywords: 統計力学 / 機械学習 / マテリアルズ・インフォマティクス / 最適化問題 / 量子アニーリング

Outline of Final Research Achievements

The use of machine learning for the development of highly functional materials is the main focus of materials informatics. However, for the safety of the materials development, it is necessary to explain the predicted functions of materials in terms of what humans can understand. In this study, we have investigated the relationship between the structure and function of a material, which are two essential elements of human understanding in materials science, and have shown that the three-dimensional structure of a material and the structure of a material suggested by its molecular formula can indeed be used to predict the physical properties that the material will exhibit. We also clarified the trade-off relation between computation time and computation accuracy for a solver of an optimization problem, which typically appears in the learning phase of machine learning. This is expected to lead to the construction of more efficient machine learning algorithms.

Free Research Field

統計力学

Academic Significance and Societal Importance of the Research Achievements

材料開発分野において機械学習の積極的利用の必要性が認識され、マテリアルズ・インフォマティクスという研究領域が開拓されて久しい．一方で、従来の機械学習は大量のデータに合わせた傾向を出力するにとどまり、材料の示す多様な物性を予測できるほどの表現力を獲得していない．また、通常の機械による学習は単なるパラメータの調節にすぎないため、背景にある理論が無視され、材料世界の科学的理解を助ける道具としては不十分である．本研究ではこの困難を解消し安心して機械学習を利用するため、(i)材料物性の理論的理解を助けるための機械学習、(ii)計算結果の精度とそれを得るまでにかかるコストの評価に焦点を当てて研究を行った．