Learning Adaptive Neighborhoods for Graph Neural Networks

This repository contains a PyTorch implementation of the ICCV 2023 paper "Learning Adaptive Neighborhoods for Graph Neural Networks".

This repository is based on https://github.com/chennnM/GCNII, we use their repository structure and their training scripts. We thank the authors for their code.

Dependencies

• CUDA 10.1
• python 3.6.9
• pytorch 1.9.1
• networkx 2.1
• scikit-learn
• torch_geometric 2.1.0

Datasets

We have maintained the datasets from the GCNII repository, with the data folder containing three benchmark datasets(Cora, Citeseer, Pubmed).

Models

We have integrated our Differentiable Graph Generator (DGG) into the following graph neural network models:

1. GCNII
2. GAT
3. SAGE
4. GCN

These models are available in the models.py file. Pretrained models are coming soon!

Usage

To run the training pipelines, use the following command:

python train_small_graphs.py --dataset cora --model GCN_DGG 

The arguments for the DGG can be found in the training scripts.

Citation

@InProceedings{Saha_2023_ICCV,
    author    = {Saha, Avishkar and Mendez, Oscar and Russell, Chris and Bowden, Richard},
    title     = {Learning Adaptive Neighborhoods for Graph Neural Networks},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {22541-22550}
}