As the reviewers agree, this paper gives a nice connection between active learning and combinatorial bandits, and provides concrete performance guarantees that outperform prior approaches. The reviewers would like to ask the authors to improve clarity in the final version of the paper.