2022 Fiscal Year Final Research Report
Development of statistical analysis methods focusing on curvature information latent in data
Project/Area Number |
19K03642
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Review Section |
Basic Section 12040:Applied mathematics and statistics-related
|
Research Institution | Keio University |
Principal Investigator |
Kobayashi Kei 慶應義塾大学, 理工学部(矢上), 准教授 (90465922)
|
Project Period (FY) |
2019-04-01 – 2023-03-31
|
Keywords | 幾何学的データ解析 / 機械学習 / データ埋め込み / 距離変換 / クラスタリング |
Outline of Final Research Achievements |
In this study, we proposed a new statistical analysis method that combines two types of distance transformations, namely α and β distances, by focusing on the distance and curvature of the data space. Furthermore, we applied this method to calculate generalized variances that take advantage of geometric features caused by annual cycles, successfully capturing new abnormal variations in rainfall data. Additionally, we applied the theory of metric cones, which was introduced for the first time in data analysis during the proposal of β distance transformation, to graph embedding, proposing a new method that automatically extracts the hierarchical structure of network data. These results were presented at international conferences and published as academic papers.
|
Free Research Field |
統計科学
|
Academic Significance and Societal Importance of the Research Achievements |
本研究成果の学術的意義は,まずその新規性にある.従来のデータ解析は,与えられたデータ間の距離をそのまま用いるか,もしくは次元削減等により得られた特徴ベクトル間の距離を用いることがほとんどであったが,本研究では,データやそれが分布する空間の距離をうまく変換した上でデータ解析を行うという点が大きく異なる.これにより,曲率等の幾何学的な特徴量に着目したデータ解析手法の開発および改良が可能となった.また,これまで用いられたことがなかった計量錐をデータ解析に応用したことも今後の発展につながる大きな新規性を含んでおり,学術的意義の高い研究成果と言える.
|