" What if a Machine Learning (ML) model could have training data generated based on its classification uncertainty and query a human-in-the-loop for their label? In doing so, would we able to construct a better ...