1019c8091693ef5c5f55970346633f92-AuthorFeedback.pdf

We thank all the reviewers for their feedback. We will incorporate them in the future version. Overall, reviewers agree that our paper is technically sound. "the learner knows the logging policy": This is a fairly standard assumption in prior work [1,5,19,22] and holds for Note that this fitting can be done from unlabeled data only. "the number of unlabeled examples is at most the size of logged data set": We look at this setting mostly for simplicity, Our results are in line with a long line of prior work on active learning theory.