Hyperbolic Graph Neural Networks

Motivatedbyrecent advances ingeometric representation learning, we propose a novel GNN architecture for learning representations on Riemannian manifolds with differentiable exponential and logarithmic maps.