Hyperbolic Graph Neural Networks

Authors: Qi Liu, Maximilian Nickel, Douwe Kiela

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In our experiments, we show that hyperbolic GNNs can lead to substantial improvements on various benchmark datasets. ... 4 Experiments
Researcher Affiliation Industry Qi Liu , Maximilian Nickel and Douwe Kiela Facebook AI Research {qiliu,maxn,dkiela}@fb.com
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Code and data are available at https://github.com/facebookresearch/hgnn
Open Datasets Yes Hence, we instead use the much larger ZINC dataset [44, 24, 23], which has been used widely in graph generation for molecules using machine learning methods [25, 31]. ... A popular choice for this purpose is the QM9 dataset [37].
Dataset Splits Yes The dataset consists of 250k examples in total, out of which we randomly sample 25k for the validation and test sets, respectively.
Hardware Specification No The paper does not provide any specific hardware details such as GPU models, CPU types, or memory amounts used for running the experiments.
Software Dependencies No The paper mentions software like Deep Chem and RAMSGrad/AMSGrad, but does not provide specific version numbers for any key software components or libraries required for reproduction.
Experiment Setup Yes We use leaky Re LU as the activation function σ with the negative slope 0.5. We use RAMSGrad [4] and AMSGrad for hyperbolic parameters and Euclidean parameters, respectively. ... For Euclidean features x E, we first apply expx (x E) to map it into the Riemannian manifolds. To initialize embeddings E within the Riemannian manifold, we first uniformly sample from a range (e.g. [ 0.01, 0.01]) to obtain Euclidean embeddings... For Barabási-Albert graphs, we set the number of edges to attach from a new node to existing nodes to a random number between 1 and 100. For Erd os-Rényi, the probability for edge creation is set to 0.1 1. For Watts-Strogatz, each node is connected to 1 100 nearest neighbors in the ring topology, and the probability of rewiring each edge is set to 0.1 1.