The code for paper "ART: Abstraction Refinement-Guided Training for Provably Correct Neural Networks" appearing in FMCAD'20. See tag fmcad20.

In your virtual environment, either install directly from this repository by

git clone [email protected]:XuankangLin/ART.git
cd ART
pip install -r requirements.txt

To replay the evaluation, run individual script from scripts/ by

bash scripts/...

The corresponding logs have been saved in results/xxx/... directory.

Alternatively, one can directly run via Docker at xuankanglin/art. To run jupyter notebook in Docker, use customizable commands like

docker run -p 8888:8888 --rm --gpus=all xuankanglin/art jupyter notebook --allow-root --ip 0.0.0.0 --no-browser --port 8888

then visit through localhost:8888 in browser, assuming nvidia-docker2 is installed to enable GPU.

The project is available open source under the terms of MIT License.