B1611
Title: Scale-invariant estimations for the factor analysis model based on its geometric structure
Authors: Michiko Okudo - The University of Tokyo (Japan) [presenting]
Fumiyasu Komaki - RIKEN CBS (Japan)
Abstract: Factor analysis is an important tool of multivariate analysis, especially in psychology. Factor analysis models are latent variable models, and the observation is divided into two parts, ``common factor'' and ``specific factor''. The maximum likelihood estimation of these models can be difficult when the dimension of common factor space is high and Bayesian estimation methods have been studied. We propose new priors for factor analysis models which are invariant under scalings of observations. These priors take advantage of the model manifold's geometric structure.
