A1034
Title: Minimum information dependence modeling for mixed-domain data analysis
Authors: Keisuke Yano - The Institute of Statistical Mathematics (Japan) [presenting]
Tomonari Sei - The University of Tokyo (Japan)
Abstract: A method of constructing a joint statistical model for mixed-domain data is proposed to analyze their dependence. Multivariate Gaussian and log-linear models are particular examples of the proposed model. The model is characterized by two orthogonal sets of parameters: the parameters of dependence and those of marginal distributions. The existence and uniqueness theorem is presented for the proposed model. To estimate the dependence parameter, Conditional inference is established and its consistency is shown. Also, the information-geometrical structure and the connection to the entropic optimal transport and the Schrodingerbridge problems are discussed. Finally, an application is illustrated to the earthquake data.
