Structure-Aware DropEdge Towards Deep Graph Convolutional Networks (TNNLS)

Jiaqi Han, Wenbing Huang, Yu Rong, Tingyang Xu, Fuchun Sun, Junzhou Huang

In IEEE Transactions on Neural Networks and Learning Systems, 2023.

[Paper]

In Structure-Aware DropEdge, we enhance graph edge dropping technique with two structure-aware samplers, the layer-dependent sampler and feature-dependent sampler, to further relieve the over-smoothing issue in deep graph networks.

Dependencies

Please check out the Python environment depicted in requirements.txt.

Data

The semi-supervised setting strictly follows GCN, and the full-supervised setting follows DropEdge. The co-author and co-purchase datasets can be downloaded from https://github.com/shchur/gnn-benchmark.

Running the Experiments

The code has been tested in the above-mentioned environment with Python=3.6.2. We recommend using conda.

conda create -n xxx python=3.6.2
conda activate xxx
pip install -r requirements.txt

To reproduce our results, just run the scripts in the scripts folder. For example,

sh scripts/semi/citeseer_appnp.sh

Citation

If you find our work helpful, please cite as:

@ARTICLE{10195874,
  author={Han, Jiaqi and Huang, Wenbing and Rong, Yu and Xu, Tingyang and Sun, Fuchun and Huang, Junzhou},
  journal={IEEE Transactions on Neural Networks and Learning Systems}, 
  title={Structure-Aware DropEdge Toward Deep Graph Convolutional Networks}, 
  year={2023},
  volume={},
  number={},
  pages={1-13},
  doi={10.1109/TNNLS.2023.3288484}}

Contact

If you have any questions, feel free to reach us at:

Jiaqi Han: alexhan99max@gmail.com