A Discussion of 'Adversarial Examples Are Not Bugs, They Are Features': Adversarial Example Researchers Need to Expand What is Meant by 'Robustness'

1. 1Recognize that adversarial examples are a feature of machine learning models, not a bug
2. 2Understand the limitations of current robustness measures
3. 3Expand the definition of robustness to include distributional shifts and other real-world scenarios
4. 4Develop new methods to improve model robustness and evaluate their effectiveness

Machine learning researchers and engineers benefit from understanding this concept to improve model robustness, and data scientists can apply this knowledge to develop more reliable models

💡 Adversarial examples are a natural consequence of machine learning models and should be considered a feature, not a bug

💡 Adversarial examples are not bugs, they're features! Rethink robustness in ML models