hyperbolic neural networks have shown great potential for modeling complex
data. However, existing hyperbolic networks are not completely hyperbolic, as
they encode features in a hyperbolic space yet formalize most of their
operations in the tangent space (a Euclidean subspace) at the