| Project/Area Number |
18K04716
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 26020:Inorganic materials and properties-related
|
|---|
| Research Institution | National Institute for Materials Science |
|---|
Principal Investigator |
KOYAMA Yukinori 国立研究開発法人物質・材料研究機構, 統合型材料開発・情報基盤部門, 主幹研究員 (20437247)
|
|---|
| Project Period (FY) |
2018-04-01 – 2021-03-31
|
| Project Status |
Completed (Fiscal Year 2020)
|
| Budget Amount *help |
¥4,420,000 (Direct Cost: ¥3,400,000、Indirect Cost: ¥1,020,000)
Fiscal Year 2020: ¥1,690,000 (Direct Cost: ¥1,300,000、Indirect Cost: ¥390,000)
Fiscal Year 2019: ¥1,690,000 (Direct Cost: ¥1,300,000、Indirect Cost: ¥390,000)
Fiscal Year 2018: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
|
|---|
| Keywords | マテリアルズ・インフォマティクス / 機械学習 / 多重信頼度モデル / 多重代入法 |
|---|
| Outline of Final Research Achievements |
Amount of data in materials science is small for specific material classes and properties, and this causes difficulty in machine learning. This is the small data problems. In this study, several machine learning techniques that can take account of relationship among multiple properties have been examined to overcome the small data problems. Cokriging is confirmed as an effective technique for multiple properties having hierarchical relationship and datasets of monotonic missing patterns. Missing patterns of real data are, however, not monotonic in general, and cokriging is not effective for such datasets. Multiple imputation techniques have been examined for datasets whose missing patterns are not monotonic and target properties have no explicit hierarchical relationship. The results suggest that the multiple imputation techniques are useful even for properties having no explicit hierarchical relationship.
|
| Academic Significance and Societal Importance of the Research Achievements |
機械学習を用いた物質の物性値推定はマテリアルズ・インフォマティクスの主要な課題のひとつである。しかし、個々の研究者が注目する特定の材料群や物性に限るとデータ数は少なく、これが材料研究における機械学習を困難にする「スモールデータ問題」が存在している。本研究では、データの階層構造や物性の相関関係を考慮し、複数の物性を同時に取り扱う機械学習が「スモールデータ問題」に対して有効な対応策となることを示すことができた。
|
