Rank-based search operators for learning latent variable models

Ricardo Silva

Abstract

We derive a Bayesian search algorithm for learning the structure of latent variable models. This is performed by searching for subsets of the observed variables whose covariance matrix can be represented as a sum of a matrix of low rank and a diagonal matrix of residuals. The resulting search procedure is relatively efficient by using a variational approximation. The model itself is a generalization of factor analysis and its variations, where we do not constrain the dependency structure among latents. The resulting models are often simpler and give a better fit than models where latents are constrained, e.g., to be independent.

Back to the Main Page