This paper proposes an interesting approach for differentiable, interpretable rule mining given a knowledge base. The major pro of the approach is its in an inductive setting without the need for negative examples, which excited the reviewers. Initially the paper lacked important comparisons to many related works, but the author did a good job in rebuttal. Please include the comparison results in the final version and the results on other datasets pointed out by the reviewers. I would like to recommend an acceptance to NeurIPS.