Reviews: Hyperbolic Neural Networks

Thanks to the authors for the detailed response. The new results presented in the rebuttal are indeed convincing, hence I am updating my score to an 8 now. This is with the understanding that these would be incorporated in the revised version of the paper. Several works in the last year have explored using hyperbolic representations for data which exhibits hierarchical latent structure. Some promising results on the efficiency of these representations at capturing hierarchical relationships have been shown, most notably by Nickel & Kiela (Nips, 2017). However one big hindrance for utilizing them so far is the lack of deep neural network models which can consume these representations as input for some other downstream task.