Reviews: Hyperbolic Graph Neural Networks

Originality This paper is an interesting generalization of neural networks on Riemannian manifolds that are agnostic to space. It introduces the idea of working on the tangent space and using a pair of functions that are inverse to each other and move points from and to the tangent space. What is not clear in the paper is why we should use functions that move points to and from the tangent space and not any pair of inverse functions. If there is no theoretical reason and empirical would have been sufficient. It is also not clear why they claim generality on any network and they picked graph nets.