This paper is close to a borderline but I think it should be accepted. The question in the paper, whether the lottery tickets are transferable across datasets, is a very obvious one to ask and I am certain that many people are thinking about it. I think there is a value in getting this paper out so that others can build upon these results. The work is mostly well executed and, with minor exceptions, it is clearly written. While suggesting this paper’s acceptance, I do suggest that the authors run the experiments suggested by the reviewer 2 and put the results in the camera-ready version.