Structure Learning in BBNs Using Regular Vines

• Activity Number: 584
• Type: Invited
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Kjersti Aas and Ingrid Hobæk Haff and Arnoldo Frigessi
• Companies: Norwegian Computing Center and Norwegian Computing Center and University of Oslo/Norwegian Computing Center
• Keywords: Bayesian Belief Networks ; regular vines ; pair-copula constructions ; structure learning ; chordal graphs

Abstract

Learning the structure of a Bayesian network from data is an NP-hard problem and still one of the most exciting challenges in machine learning (Zhu et. al., 2012). Broadly the existing structure learning algorithms fall into two categories: score-based and constraint-based approaches. When dealing with continuous data, both types of methods usually assume that the involved random variables have a joint multivariate normal distribution. In this talk, we launch the vine selection methodology as a new approach for learning the structure of Bayesian networks. Using this methodology one is restricted to a certain sub-class of chordal graphs. However, we show through several practical applications the possibility of using non-Gaussian margins and a non-linear dependency structure more than outweighs this restriction.