Bayesian network structure learning when discrete and continuous variables are present.

2015 Fiscal Year Final Research Report

• Project/Area Number: 24500172
• 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: Suzuki Joe 大阪大学, 理学(系)研究科(研究院), 准教授 (50216397)
• Co-Investigator(Renkei-kenkyūsha): Washio Takashi 大阪大学, 産業科学研究所, 教授 (00192815) ; Kano Yutaka 大阪大学, 大学院基礎工学研究科, 教授 (20201436)
• Project Period (FY): 2012-04-01 – 2016-03-31
• Keywords: ベイジアンネットワーク / 構造学習 / 事後確率最大 / 相互情報量の検定 / 独立性の検定

Outline of Final Research Achievements

We consider Bayesian network structure learning when discrete and continuous variables are present. The problem is rather hard and very few results are available. I particular, we had to assume that each continuous variable is Gaussian and no two discrete variable should be between a continuous variable. In this research, we mathematically prove consistency (the correct structure is estimated as the sample size increases). In particular, we proposed applications to independence testing and estimation of mutual information.

Free Research Field

機械学習　情報理論