Cause and Effect

Cause and Effect: Concept Based Explanation of Neural Networks

article link

Download

For downloading the datasets run the scripts:

    ./script/dlbroden.sh
    ./script/dlplaces.sh
    ./script/dlzoo.sh

Requirements

• Python Environments

    pip3 install numpy matplotlib torch torchvision tqdm progressbar2 pandas opencv-python

and pull submodules:

    git pull --recurse-submodules

Run Cause and Effect

• For running main analysis run:

    python3 forgetting.py

• For running deepdream based on the trained concepts run:

    python3 deepdream.py

• For evaluation of scatter plots and finding the most necessary concepts for a classifier run:

    python3 scatter3.py

Reference

If you find the codes useful, please cite this paper

@article{zaeem2021cause,
  title={Cause and Effect: Concept-based Explanation of Neural Networks},
  author={Zaeem, Mohammad Nokhbeh and Komeili, Majid},
  journal={arXiv preprint arXiv:2105.07033},
  year={2021}
}