Development of discovering statistical methods via sparse modeling

2023 Fiscal Year Final Research Report

• Project/Area Number: 18K18009
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 60030:Statistical science-related
• Research Institution: Keio University (2019-2023); Tokyo Institute of Technology (2018)
• Principal Investigator: Katayama Shota 慶應義塾大学, 経済学部(三田), 准教授 (50742459)
• Project Period (FY): 2018-04-01 – 2024-03-31
• Keywords: 高次元データ / スパースモデリング / 因果推論 / 多重検定

Outline of Final Research Achievements

Aiming at the development of discovering statistical methods via sparse modeling, particularly (1) difference detection in high dimensional linear regression models and (2) two sample problems with ultra high dimensional parameters are studied. In the theme (1), a method for directly and sparsely estimating difference in regression coefficient vectors is developed, and gave its prediction error, variable selection consistency and derivation of the asymptotic distribution based on de-biasing. In the theme (2), a statistical inference for the ultra high dimensional parameters that characterize the differences between two groups is provided for application to the analysis of gene data. Furthermore, the proposed procedure compared the RNA-seq data of high and low risk Covid-19 patients.

Free Research Field

高次元データ解析

Academic Significance and Societal Importance of the Research Achievements

本研究課題で実施した研究(1)(2)はどちらも基礎的なものであり，それゆえに社会的意義も大きい．(1)については医療・経済・マーケティングなどへの応用が考えられ，提案手法の解釈可能性から，個体に依存した処置や介入へと繋がる．(2)については，遺伝子データからのさらなる有益な情報抽出が可能となる．また，どちらの研究も新規の方法論を開発しており，さらにはその理論保証も与えている．