Efficiently extract explanations, justifications, and uncertainty from black-box DNNs using a novel Bayesian approach.