https://github.com/nvedant07/Fairness-Through-Robustness
Official code for FAccT'21 paper "Fairness Through Robustness: Investigating Robustness Disparity in Deep Learning" https://arxiv.org/abs/2006.12621
https://github.com/nvedant07/Fairness-Through-Robustness
Last synced: 3 months ago
JSON representation
Official code for FAccT'21 paper "Fairness Through Robustness: Investigating Robustness Disparity in Deep Learning" https://arxiv.org/abs/2006.12621
- Host: GitHub
- URL: https://github.com/nvedant07/Fairness-Through-Robustness
- Owner: nvedant07
- Created: 2021-01-14T15:47:09.000Z (about 4 years ago)
- Default Branch: master
- Last Pushed: 2021-03-09T13:30:09.000Z (almost 4 years ago)
- Last Synced: 2024-08-03T13:14:23.039Z (7 months ago)
- Language: Jupyter Notebook
- Size: 2.29 MB
- Stars: 12
- Watchers: 1
- Forks: 1
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
## Fairness Through Robustness: Investigating Robustness Disparity in Deep Learning
Official code to replicate results in [Fairness Through Robustness: Investigating Robustness Disparity in Deep Learning](https://arxiv.org/abs/2006.12621), presented at [FAccT 2021](https://facctconference.org/2021/acceptedpapers.html) in ~~Toronto, Canada~~ over Zoom.
Authors:
* [Vedant Nanda*](http://nvedant07.github.io/)
* [Samuel Dooley*](https://www.cs.umd.edu/~sdooley1/)
* [Sahil Singla](https://scholar.google.co.in/citations?user=jjjbOI4AAAAJ&hl=en)
* [Soheil Feizi](https://www.cs.umd.edu/~sfeizi/)
* [John P. Dickerson](http://jpdickerson.com/)
Adversarial attacks in our code use [`foolbox`](https://github.com/bethgelab/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](https://github.com/bethgelab/foolbox) 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](https://openaccess.thecvf.com/content_cvpr_2016/html/Moosavi-Dezfooli_DeepFool_A_Simple_CVPR_2016_paper.html) and [Carlini&Wagner](https://ieeexplore.ieee.org/abstract/document/7958570/)).
#### Running Adversarial Attacks
See `./code/experiment-adversarial.py`
#### Running Randomized Smoothing
```
cd smoothing_code
python certify.py cifar10 0.125 --batch 100 --alpha 0.001 --N0 100 --N 100000 --skip 5 --gpu
```
#### 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 `[email protected]` or `[email protected]`.