Automatic Model Selection by Cross-Validation for Probabilistic PCA

摘要

The Mixture of Probabilistic Principal Components Analyzers (MPPCA) is a multivariate analysis technique which defines a Gaussian probabilistic model at each unit. The numbers of units and principal directions in each unit are not learned in the original approach. Variational Bayesian approaches have been proposed for this purpose, which rely on assumptions on the probability distributions of the MPPCA parameters. Here we present a different way to solve this problem, where cross-validation and simulated annealing are combined to guide the search for an optimal model selection, providing a structured strategy to escape from suboptimal configurations. This allows to learn the model architecture without the need of any assumptions other than those of the basic PPCA framework. Experimental results are presented, which show the probability density estimation and missing value imputation features of the proposal.