• Dynamical Low-Rank Compression of Neural Networks with Robustness under Adversarial Attacks
• Dynamic Low-Rank Training with Spectral Regularization: Achieving Robustness in Compressed Representations

1. Clone the Github repository:

   git clone https://github.com/ScSteffen/RobustDLRT.git
   

2. Create a local python environment and install the python requirements in a local virtual environment:

   python3 -m venv ./venv
   source venv/bin/activate
   pip install -r requirements.txt
   

Set up datasets

1. put the data into .dataset/data_adversarial_rs/<testcasename>. The easiest way to get started is to run

   cd dataset
   python create_cifar10.py

sh run_example.sh

To run experiments on OReole-FM MR models, the transformers need to be converted from the timm format to transformers format.

Run convert_timm_to_hf_vit.py with the appropriate command line arguments.

Main contributors

• Schnake, Stefan
• Schotthoefer, Steffen
• Yang, Lexie H.

Student contributors

• Snyder, Thomas
• Park, Hannah