Reviews: Hierarchical Graph Representation Learning with Differentiable Pooling

The paper is very well written, the method is simple (which is not a bad thing!) and widely applicable, and the results are favourable. Building hierarchical representations is an important feature of CNNs and extending this to more general graph datatypes is clearly an important contribution. Overall I strongly support publication of this paper. I only have minor additional thoughts: 1. When the edges have labels (e.g.