DeepHGCN

This repository contains a PyTorch implementation of "DeepHGCN: Toward Deeper Hyperbolic Graph Convolutional Networks".

Paper: https://ieeexplore.ieee.org/abstract/document/10632071

Directories

• deephgcn: code implementation for the proposed DeepHGCN model.
• geooptplus: the augmented geoopt package for implementing key features of HNN++.
• layers: including the implementation of proposed Hyperbolic Linear Layer, as well as other compared feature transformation methods in DeepHGCN paper.
• notes: miscellaneous visualizations and basic evaluations.

Requirement

• geoopt (pip install geoopt)
• networkx
• scikit-learn

Reference Code

• geoopt: https://github.com/geoopt/geoopt
• HNN++: https://github.com/mil-tokyo/hyperbolic_nn_plusplus
• HGCN: https://github.com/HazyResearch/hgcn
• GCNII: https://github.com/chennnM/GCNII
• Fréchet Mean: https://github.com/CUAI/Differentiable-Frechet-Mean

Citation

@article{liu2024deephgcn,
  title={DeepHGCN: Toward Deeper Hyperbolic Graph Convolutional Networks},
  author={Liu, Jiaxu and Yi, Xinping and Huang, Xiaowei},
  journal={IEEE Transactions on Artificial Intelligence},
  year={2024},
  publisher={IEEE}
}