Forward invariance of neural ODEs

Causal manipulation of neural ODEs (via model parameters or external inputs) to achieve performance guarantees, such as safety

There are four simple modelling demos using neural ODEs with performance specifications (spiral curve regression, convexity portrait, Mujoco, and end-to-end lidar-based autonomous driving).

Setup

$ conda create -n invODE python=3.8
$ conda activate invODE
$ pip install torch==1.10.1+cu113 torchvision==0.11.2+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
$ pip install pytorch-lightning==1.5.8 opencv-python==4.5.2.54 matplotlib==3.5.1 ffio==0.1.0  descartes==1.1.0  pyrender==0.1.45  pandas==1.3.5 shapely==1.7.1 scikit-video==1.1.11 scipy==1.6.3 h5py==3.1.0
$ pip install qpth cvxpy cvxopt
$ pip install torchdiffeq

If you find this helpful, please cite our work:

@inproceedings{xiao2023inv,
  title = {On the Forward Invariance of Neural ODEs},
  author = {Wei Xiao and Tsun-Hsuan Wang and Ramin Hasani and Mathias Lechner and Yutong Ban and Chuang Gan and Daniela Rus},
  booktitle = {International Conference on Machine Learning},
  year = {2023}
}