[NeurIPS 2023] Two Sides of The Same Coin: Bridging Deep Equilibrium Models and Neural ODEs via Homotopy Continuation

Code release for Two Sides of The Same Coin: Bridging Deep Equilibrium Models and Neural ODEs via Homotopy Continuation (NeurIPS 2023).

[paper] [project page]

Requirements

numpy                         1.21.6
torch                         1.13.1
scipy                         1.9.1

Preparation

To run the experiments, you need to first install the modified python package torchdiffeq via running pip install -e . in the current directory.

Getting started

Then, you need to first run cd scripts to enter scripts directory of the project and then simply

run python homoinit_cifar10.py to re-implement our experiments on CIFAR10 using HomoODE.

run python homoinit_cifar100.py to re-implement our experiments on CIFAR100 using HomoODE.

run python homoinit_tiny.py to re-implement our experiments on Tiny ImageNet using HomoODE.

Notably, you can obtain the Tiny ImageNet Dataset in https://www.kaggle.com/competitions/tiny-imagenet/data.

You need to download it and place it in directory scripts/data/tiny-imagenet.

Acknowledgement

The implementation of HomoODE is based on torchdiffeq.

Citation

If you find this repository useful in your research, please consider citing:

@inproceedings{
ding2023two,
title={Two Sides of The Same Coin: Bridging Deep Equilibrium Models and Neural ODEs via Homotopy Continuation},
author={Shutong Ding and Tianyu Cui and Jingya Wang and Ye Shi},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=R2rJq5OHdr}
}