F-036
Some Pitfalls in Infinite Relational Data Analysis

○中野允裕（NTT）

This paper discusses two case studies of Bayesian nonparametric relational data analysis, and clarifies some pitfalls in infinitely exchangeable array models. We first briefly review a variety of conventional relational models in this field, including the infinite relational model, the Mondrian process,the rectangular tiling process, and the binary space partitioning-tree model. Then we clarify some technical nuisances for the conventional models to fail to solve. Finally, we discuss two case studies for alternative strategies to overcome them.