Operator-learning-inspired Modeling of Neural Ordinary Differential Equations (AAAI 2024)

Introduction

This repository contains pytorch implementation for our AAAI 2024 paper:

Operator-learning-inspired Modeling of Neural Ordinary Differential Equations

Experimental environment settings.

Run the following code before starting the experiment.

conda env create -f requirements.yaml

Training / Test

Run the following code for training / test.

python main.py --tol 1e-3 --epochs 10 --batch_size 64 --hidden_size 76 

If you want to train and test in a different environmental setting, it can be done by changing the parsers below.

[ parser ]      [ Description of parser ]
--tol          : DOPRI-5 error tolerance
--epochs       : Number of epoch
--batch_size   : Batch size
--hidden_size  : Size of hidden vector

We release code for image classification tasks (CIFAR-10 dataset).

Test (Only)

If you want to evaluate it quickly, run the following code :

python test.py --path './model/10.pt'

[ parser ]      [ Description of parser ]
--path         : The path where checkpoint exists

Oher codes

[ code ]        [ Description of code ]
main.py         : Code for training and testing.
models.py       : Our model 
utils.py        : Modules required during training
test.py         : Code for testing.

Checkpoint

We provide checkpoint of trained BFNO-NODE(our model), which are located in the ./model folder.