Fairness Through Robustness: Investigating Robustness Disparity in Deep Learning

Official code to replicate results in Fairness Through Robustness: Investigating Robustness Disparity in Deep Learning, presented at FAccT 2021 in Toronto, Canada over Zoom.

Authors:

• Vedant Nanda*
• Samuel Dooley*
• Sahil Singla
• Soheil Feizi
• John P. Dickerson

Adversarial attacks in our code use foolbox. Code under ./smoothing_code is taken from https://github.com/locuslab/smoothing and https://github.com/Hadisalman/smoothing-adversarial. We thank the authors of these repos for making their code public and easy to use.

Requirements

• python 3.7.x or higher
• CUDA 10.1

python dependencies

• To install all dependencies run pip install -r requirements.txt
• NOTE: Our implementation uses foolbox v2.4.0. If not available on PyPi, we recommend building it from source. Follow instructions on the Foolbox repo to build from source.

Datasets

We use CIFAR-10, CIFAR-100, UTKFace and Adience for our analysis. See ./data for more information on how to download the datasets.

Training Models

We've released pre-trained weights used in our analysis (coming soon!). You can either use these weights or train models on your own.

Working with pre-trained weights

Coming soon!

Training models

See ./code/experiment-main.py.

Evaluating Robustness Bias

This step requires running adversarial attacks (in our paper we evaluate DeepFool and Carlini&Wagner).

Running Adversarial Attacks

See ./code/experiment-adversarial.py

Running Randomized Smoothing

cd smoothing_code
python certify.py cifar10 <path to model weights> <model_name> 0.125 <output_filename> --batch 100 --alpha 0.001 --N0 100 --N 100000 --skip 5 --gpu <gpu number>

Analysis

See ./code/certificate_analysis/ for analysis of randomized smoothing and ./code/experiment_adversarial_only_plot.ipynb for analysis of adversarial attacks.

Citation

If you found our work useful, please cite it.

@inproceedings{nanda2021fairness,
    author = {Nanda, Vedant and Dooley, Samuel and Singla, Sahil and Feizi, Soheil and Dickerson, John P.},
    title = {Fairness Through Robustness: Investigating Robustness Disparity in Deep Learning},
    year = {2021},
    isbn = {9781450383097},
    publisher = {Association for Computing Machinery},
    address = {New York, NY, USA},
    url = {https://doi.org/10.1145/3442188.3445910},
    doi = {10.1145/3442188.3445910},
    booktitle = {Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency},
    pages = {466–477},
    numpages = {12},
    location = {Virtual Event, Canada},
    series = {FAccT '21}
}

Feedback

We'd love to hear from you! Write to us at vedant@cs.umd.edu or sdooley1@cs.umd.edu.