Review for NeurIPS paper: Subgraph Neural Networks

Additional Feedback: - Related Work: Is covered broadly. Since the method of the authors heavily relies on the three factors they propose, it would make sense to mention alternative factor representations of nodes/graphs in the literature. Yet, this section is very dense and many parts are confusing at first read. Maybe the authors could try to ease reading visually by using an enumeration in each paragraph in Sec. Although this is ok since you can use them for benchmarking, I would be interested if the authors know about if this kind of classification is actually done in practice, or if there's a need for such approaches.