Review for NeurIPS paper: Multi-label classification: do Hamming loss and subset accuracy really conflict with each other?

This is an interesting theoretical paper that performs a cross-analysis of three popular loss functions used in multi-label classification. Despite the fact that some reviewers found the analysis straight-forward, as it is mainly based on known results either from multi-class classification or multi-label classification, the interpretation of the results from the multi-label classification perspective is very interesting. It is worth to underline that there is still a gap in theory for multi-label classification and this paper tries to fill it. Nevertheless, the paper has several flaws that makes the paper a borderline case. The current discussion about the frameworks used in [11] and in this submission is misleading (the rebuttal makes slightly better job in this regard).