Bayesian networks: generating independent samples

Abstract

Given a Bayesian network with evidence, we propose an efficient majorizing function for the Acceptance-Rejection method to generate a sequence of independent samples from the posterior distribution. Because the proposed method generates independent samples, it has none of the problems associated with MCMC. The sequence allows for the easy calculations of the confidence intervals for the posterior quantities of interest. It can also yield independent inputs for the simulations of a bigger probabilistic system that includes the BN.

Keywords:

• Bayesian networks
• logic samplings
• mutilated samplings
• independence samplings