The paper has received largely positive reviews (with the exception of one outlier), and the author response addressed the initial concerns adequately. Thus, the paper is suitable for publication at NeurIPS without significant changes, although the authors are strongly urged to implement the changes promised in the author response.