Finding Global Homophily in Graph Neural Networks When Meeting Heterophily

Authors: Xiang Li, Renyu Zhu, Yao Cheng, Caihua Shan, Siqiang Luo, Dongsheng Li, Weining Qian

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments to compare our models against 11 other competitors on 15 benchmark datasets in a wide range of domains, scales and graph heterophilies. Experimental results show that our methods achieve superior performance and are also very efficient.
Researcher Affiliation Collaboration 1School of Data Science and Engineering, East China Normal University, Shanghai, China 2Microsoft Research Asia, Shanghai, China 3School of Computer Science and Engineering, Nanyang Technological University, Singapore.
Pseudocode Yes Finally, we summarize the pseudocodes of Glo GNN in Algorithm 1 (Section A of the appendix).
Open Source Code Yes We provide our code and data at https://github.com/RecklessRonan/GloGNN.
Open Datasets Yes For fairness, we conduct experiments on 15 benchmark datasets, which include 9 small-scale datasets released by (Pei et al., 2020) and 6 large-scale datasets from (Lim et al., 2021). We use the same training/validation/test splits as provided by the original papers.
Dataset Splits Yes We use the same training/validation/test splits as provided by the original papers.
Hardware Specification Yes Meanwhile, we run the experiments of 6 large-scale datasets on a single Tesla V100 GPU with 32G memory and use Adam W as the optimizer following (Lim et al., 2021).
Software Dependencies No The paper mentions "We implemented Glo GNN by Py Torch." but does not provide a specific version number for PyTorch or any other software dependencies.
Experiment Setup Yes We perform a grid search to tune hyper-parameters based on the results on the validation set. Details of these hyper-parameters are listed in Tables 3 and 4.