Quartet-Based Learning of Hierarchical Latent Class Models: Discovery of Shallow Latent Variables.
Tao Chen, Nevin Lianwen Zhang · Rare & Special e-Zone (The Hong Kong University of Science and Technology) · 2006
Hierarchical latent class (HLC) models are tree-structured Bayesian networks where leaf nodes are observed while internal nodes are hidden. The currently most efficient algorithm for learning HLC models can deal with only a few dozen observed variables. While this is sufficient for some applications, more efficient algorithms are needed for domains with, e.g., hundreds of observed variables. With this demand in mind, we explore quartet-based methods. The basic idea comes from phylogenetic tree reconstruction: One first learn submodels for quartets — groups of four observed variables— and then derive an overall model from those quartet submodels. As the first step in the new direction, this paper assumes that there is a way to find the “true” submodel for any quartet and investigate how to identify shallow latent variables efficiently by using the minimum number of quartet submodels. By shallow latent variables, we mean latent variables that are connected to at least one observed variable.