The paper presents new methods and, for the first time, theoretical guarantees for learning in hyperbolic spaces, which has benefits over Euclidean spaces especially learning with hierarchical data. The reviewers found the work sound. The narrow empirical evaluation was seen as a minor weakness.