Code for the paper "ExLipBaB: Exact Lipschitz Constant Computation for Piecewise Linear Neural Networks"
The package was developed under python 3.11.14 and can be installed with:
$ pip install .we recommend first creating a virtual environment, for example through
$ conda create -n exlipbab python=3.11Note: On some systems, we encountered errors when trying to install pycddlib automatically as above. In such cases, try manually installing pyddlib (e.g. version 3.0.2) through
$ python -m pip install pycddlib
before installing exlipbab.
The code for the examples of the paper can be found in the applications Folder and should mostly run as-is with an installed exlipbab package. They are named <model_name>_exlipbab_computation.py. Some data or code fragments, we cannot redistribute, so we link to the corresponding repositories.
The code for the comparative results with LipSDP can be found in the LipSDP_applications Folder, but needs the corresponding repositories to function correctly, which are described in the corresponding matlab files.
The package structure of exlipbab was created with cookiecutter and the py-pkgs-cookiecutter template.