Research on Principle and Methods of Large-scale Causal Infrerence Based on Nonlinearity

2019 Fiscal Year Final Research Report

• Project/Area Number: 17K00305
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Intelligent informatics
• Research Institution: Osaka University
• Principal Investigator: Washio Takashi 大阪大学, 産業科学研究所, 教授 (00192815)
• Project Period (FY): 2017-04-01 – 2020-03-31
• Keywords: 統計的因果推論 / 因果解析 / 機械学習 / データマイニング / 回帰解析 / 非線形性

Outline of Final Research Achievements

There is an increasing need to understand the mechanism of large-scale systems by analyzing big data by statistical causal inference. However, its practical principles and methods have been established only for large-scale systems that are linear and have non-Gaussian noise. This research achieved (1) establishment of a new principle for estimating the causal relationship between many observation variables in a non-linear system with high accuracy, (2) development of statistical causal inference methods for large-scale systems by further extending the new principle, (3) basic performance verification using large-scale artificial data, and (4) practical performance verification using real-world data. We developed practical principles and methods for a wide range of large-scale nonlinear systems though these studies. Furthermore, we presented these results in major international conferences and international journals, and spread the breakthrough method of statistical causal reasoning.

Free Research Field

機械学習，データマイニング

Academic Significance and Societal Importance of the Research Achievements

開発した因果推論手法は対象系の非線形性に基づき、データの非線形回帰残差の大小のみで変数間の因果関係を推定できる。これは変数とノイズの独立性の推定に基づく従来の因果推論の枠組みと全く異なる。この新原理により、ノイズの性質や変数とノイズの独立性、交絡変数の有無に係わらず因果関係を一意かつ高速に推定できる。本提案原理は学術的に独創的かつ基礎的であり、本分野の世界的研究動向に新しい方向性を与えている。
実世界の殆どの対象系は何等かの非線形性を有し、本開発手法は実用的にも広範な対象のメカニズム解析に適用可能である。今後、物理学、化学、生物学、各種産業の現象解析や設計にて重要な役割を担うと期待される。