Convolution Networks (HGCN) model and baselines, along with our detailed reproducible training setup

We really appreciate the reviewers' time and effort. First, our paper is an empirical paper. Our main result is that HGCN achieves error reduction of up to 63.1% in ROC AUC for link prediction and of up to Simply setting GCN variables to be optimized in hyperbolic space does not yield good performance. We show in ablation analysis that these algorithmic contributions result in up to 9.9% absolute gain compared to simple GCN in hyperbolic space, an improvement that is larger than what any Euclidean GCN variant achieves. Indeed, we had run unit tests verifying that points are mapped to the correct tangent space.