Consider an active learning setting in which a learner has a training set
with few labeled examples and a pool set with many unlabeled inputs, while a
remote teacher has a pre-trained model that is known to perform well for the
learner's task. The learner actively transmits batches of