CTRAIN 0.1.1

CTRAIN is an unified, modular and comprehensive package for the training and evaluation of certified training techniques.

Implementation of Certified Training Methods

• IBP (Gowal et al. , On the Effectiveness of Interval Bound Propagation for Training Verifiably Robust Models, 2019)
• Improved IBP (Shi et al., Fast Certified Robust Training with Short Warmup, 2021)
• CROWN-IBP (Zhang et al., Towards Stable and Efficient Training of Verifably Robust Neural Networks)
• SABR
• TAPS
• STAPS

TODO: Multiple Recent Papers by De Palma: https://arxiv.org/pdf/2206.14772 https://arxiv.org/pdf/2305.13991 https://arxiv.org/pdf/2410.01617

Setup

1. Create Virtual Environment and activate it

python3 -m venv ./venv
source venv/bin/activate

2. Clone auto_LiRPA

git clone [e-mail]:Verified-Intelligence/auto_LiRPA.git

3. For complete verification, install $\alpha\beta$-CROWN

git clone [e-mail]:Verified-Intelligence/alpha-beta-CROWN.git

Then, adjust the folder name and add an init file, s.t. we can import it as a module.

mv ./alpha-beta-CROWN ./abCROWN
touch ./abCROWN/__init__.py

4. Install Dependencies

On Linux:

pip3 install -r requirements_linux.txt

On MacOS:

pip3 install -r requirements_macos.txt

And install auto_LiRPA

pip3 install ./auto_LiRPA

ToDos

• Investigate Crit. Eps for different certified training methods
• Investigate number of stable/unstable/active/inactive neurons