Development and Application of Co-nonlinearity Analysis Methods Leading to Novel Knowledge Awareness

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K12018
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: Doshisha University
• Principal Investigator: Ohsaki Miho 同志社大学, 理工学部, 教授 (30313927)
• Co-Investigator(Kenkyū-buntansha): 大西 圭 九州工業大学, 大学院情報工学研究院, 教授 (30419618) ; 片桐 滋 同志社大学, 研究開発推進機構, 嘱託研究員 (40396114)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 知識発見 / 非線形従属関係 / 共非線形性尺度 / 正則化 / ニューラルネットワーク / グループラッソ

Outline of Final Research Achievements

Knowing the complex dependent relationships (co-nonlinearity) between multiple variables is the first step to understanding and elucidating phenomena. Machine learning techniques are necessary that can inductively discover unknown co-nonlinearities, which are difficult to notice with known knowledge, from a wide variety of data. To realization that, the present study proposed, developed, and evaluated the following measure and method: A measure called NNR-GL that detects co-nonlinearity by the combination of Neural Network Regression (NNR) and Group Lasso (GL). A method NNR-GLIA that discovers the sets of co-nonlinear variables and representatives by adding the function of Information Aggregation (IA) to NNR-GL. As research results, we completed NNR-GL and NNR-GLIA and demonstrated their theoretical and practical effectiveness through experiments.

Free Research Field

知能情報学

Academic Significance and Societal Importance of the Research Achievements

共非線形性尺度NNR-GLは従来の従属関係尺度の問題(2変数間に限定，設定の難しさ等)を解決し，より高い性能を達成できる．NNR-GLで得た従属関係を集合に集約するNNR-GLIAは共非線形変数集合・代表を発見可能である．断片的な従属関係を検出する尺度や事前に絞り込んだ変数集合の因果構造を推論する手法は存在するが，変数集合を絞り込む手法は見られない．我々が知る限り，NNR-GLIAはこの役割を担う初の手法である．NNR-GLとNNR-GLIAは「まずは仮説にとらわれずに現象の理解の糸口を探す」ことを支援するため，科学や工学に広く貢献すると考えられる．