2021 Fiscal Year Final Research Report
Development of machine learning method for distribution data based on information geometry
Project/Area Number |
17H01793
|
Research Category |
Grant-in-Aid for Scientific Research (B)
|
Allocation Type | Single-year Grants |
Section | 一般 |
Research Field |
Intelligent informatics
|
Research Institution | National Institute of Advanced Industrial Science and Technology |
Principal Investigator |
Akaho Shotaro 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 上級主任研究員 (40356340)
|
Co-Investigator(Kenkyū-buntansha) |
藤木 淳 福岡大学, 理学部, 教授 (10357907)
日野 英逸 統計数理研究所, モデリング研究系, 教授 (10580079)
村田 昇 早稲田大学, 理工学術院, 教授 (60242038)
|
Project Period (FY) |
2017-04-01 – 2022-03-31
|
Keywords | 機械学習 / ノンパラメトリック / 情報幾何 / 学習アルゴリズム |
Outline of Final Research Achievements |
An effective way to handle large amounts of data by machine learning is to reduce the data to the parameters of a probability distribution. In this project, we have been working on the development of machine learning for such data. Originally, there was a study of extending principal component analysis to distributional data, which had been developed by a project member. The significant contribution of this project was to extend it to a more flexible nonparametric framework, which was achieved through information geometry of Gaussian process regression and other methods. We have also been able to apply information geometry to neuroscience and geophysics through the application of matrix factorization.
|
Free Research Field |
機械学習
|
Academic Significance and Societal Importance of the Research Achievements |
機械学習では、通常実数値ベクトルを入力として学習が行われることが前提であるが、大量のデータを集約した分布のパラメータなどは実数値として扱うことが必ずしも適切ではない。本研究課題では、そのような分布データに関する機械学習の手法を理論・応用の両面から深化させ、さまざまな応用課題(神経科学や地球科学をはじめとする多くの分野)に適用するための方法論を確立したことにある。
|