Goto

Collaborating Authors

 gcn 2



Review for NeurIPS paper: Natural Graph Networks

Neural Information Processing Systems

Additional Feedback: Specific notes: Line 132, "local symmetries form a superset of the global symmetries": I'm not sure what this means. Local and global symmetries are different types of object, right? Is this just intended as meaning that global symmetries restrict to local ones? Line 147: "it is necessary that the feature vector ... transforms ... rather than remain invariant" Is this actually true, and if so, what is the justification? It seems to me like using a non-transforming feature vector with a powerful edge kernel could still be possible?