Notebooks and helpers for the "Adversaries examples & human-ML alignment" tutorial.

Exercise notebook

Solution notebook

Based on the following works

[IST+19] Ilyas A., Santurkar S., Tsipras D., Engstrom L., Tran B., Madry A. (2019). Adversarial Examples Are Not Bugs, They Are Features. arXiv, arXiv:1905.02175

[EIS+19] Engstrom L., Ilyas A., Santurkar S., Tsipras D., Tran B., Madry A. (2019). Learning Perceptually-Aligned Representations via Adversarial Robustness. arXiv, arXiv:1906.00945

[STE+19] Santurkar S., Tsipras D., Tran B., Ilyas A., Engstrom L., Madry A. (2019). Image Synthesis with a Single (Robust) Classifier. arXiv, arXiv:1906.09453

[EIS+19] Robustness (Python Library) (2019); https://github.com/MadryLab/robustness.

Citation

If you use this material or code, please cite it as follows:

@misc{santurkar2020notes,
   title={Adversarial examples and human-ML alignment (MIT BCS tutorial)},
   author={Shibani Santurkar and Dimitris Tsipras},
   year={2020},
   url={https://github.com/MadryLab/BCS_Tutoria}
}

Contributors

• Shibani Santurkar
• Dimitris Tsipras