/Simplicial-neural-network-benchmark

Simplicial neural network benchmarking software

Primary LanguagePythonMIT LicenseMIT

Simplicial-neural-network-benchmark

Simplicial neural network benchmarking software as part of my final year dissertation project

Paper on SAT can be found here: https://arxiv.org/abs/2204.09455

To create the correct environments, run the following commands

conda create -n myenv python=3.8
conda activate myenv
./requirements.sh

This software features the implementation of 6 models:

  • Graph convolutional network (GCN) by Kipf and Welling
  • Graph attention network (GAT) by Veličković et al.
  • Simplicial neural network (SCN) by Ebli, Defferrard, and Spreemann
  • Simplicial 2-complex convolutional neural network (SCConv) by Bunch et al
  • Simplicial attention network (SAT) by Goh, Bodnar, and Liò
  • Simplicial attention neural network (SAN) by Giusti et al.

This software features four benchmarking tests

  • Superpixel graph classification of MNIST and CIFAR10 images (superpixel_benchmark.py)
  • Trajectory classification (orientation_flow_benchmark.py)
  • Adversarial resistance (adversarial_superpixel_benchmark.py)
  • Unsupervised representation learning (planetoid_dgi_benchmark.py)