Code release for Subhomogeneous Deep Equilibrium Models (ICML 2024).
Can install with pip install -r requirements.txt.
run main.py to re-do our experiments on MNIST, CIFAR100, SVHN, and Tiny ImageNe using the SubDEQ.
run train.py to re-do our experiments on Cora citation, Cora author, CiteSeer, DBLP, and PubMed using the SubDEQ GNN.
The implementation of SubDEQ is based on Deep Implicit Layers - Neural ODEs, Deep Equilibirum Models, and Beyond and A PyTorch implementation of Predict then Propagate: Graph Neural Networks meet Personalized PageRank.
If you find this repository useful in your research, please consider citing:
@inproceedings{sittoni2024subhomogeneous,
title = {Subhomogeneous Deep Equilibrium Models},
author = {Pietro Sittoni and Tudisco, Francesco},
booktitle={International Conference on Machine Learning (ICML)},
year = {2024}
}