README.md

Code for "Learning Perceptually-Aligned Representations via Adversarial Robustness"

These are notebooks for reproducing our paper "Learning Perceptually-Aligned Representations via Adversarial Robustness" (preprint, blog). Based on the robustness python library.

Running the notebooks

Steps to run the notebooks (for now, requires CUDA):

• Clone this repository
• Download our models from S3: CIFAR-10, Restricted ImageNet (standard training for comparison)
• Make a models folder in the main repository folder, and save the checkpoints there
• Install all the required packages with pip install -r requirements.txt
• Edit user_constants.py to point to PyTorch-formatted versions of the CIFAR and ImageNet datasets
• Start a jupyter notebook server: jupyter notebook . --ip 0.0.0.0

Citation

@inproceedings{engstrom2019learning,
    title={Learning Perceptually-Aligned Representations via Adversarial Robustness},
    author={Logan Engstrom and Andrew Ilyas and Shibani Santurkar and Dimitris Tsipras and Brandon Tran and Aleksander Madry},
    booktitle={ArXiv preprint arXiv:1906.00945},
    year={2019}
}