First, the relation to GD18 had no bearing on decision, and none of the reviewers had a COI. This is a borderline paper. The reviewers liked the notion of differential submodularity, and the avenues of future work it may open. That said, there are several things that need to be improved before the paper is ready for publication: 1. take all reviewer comments into account 2. add truly large scale experiments that really test the new method and compare it to existing ones. Only this will show the true performance. Please see the scale of experiments e.g. in these papers: http://proceedings.mlr.press/v48/mirzasoleiman16.html http://proceedings.mlr.press/v70/mirzasoleiman17a.html http://proceedings.mlr.press/v80/mitrovic18a.html http://proceedings.mlr.press/v97/kazemi19a.html These types of experiments are state of the art and need to be included. 5000 is too small. 3. Add more examples and applications. To cite from the discussion: "My main concern/question is the applicability to other problems. That is, with extra applications to other problems it would be more convincing that this notion is interesting for further investigation." These improvements are a 'must'.