NeurIPS 2020

Learning Global Transparent Models consistent with Local Contrastive Explanations

Meta Review

Through an active discussion, it was clear that the reviewers found the proposed ideas to be quite valuable, and the evaluation to be quite convincing. However, the scope of the work needs to be considerably clarified in the subsequent versions, for example, make it obvious that it will only work on tabular datasets, and without long conjunctions of features in the model. With appropriate framing (which we strongly encourage the authors to do for the final version), this paper will likely have much more impact.