approximate inference

We propose a novel Bayesian batch active learning approach motivated by approximating the complete data posterior of the model parameters.