Unintended Interactions Between Machine Learning Defenses and Security, Privacy, and Fairness Risks

Overfitting and memorization are the underlying causes of unintended interactions between machine learning defenses and security, privacy, and fairness risks. Different factors influence these causes, leading to complex trade-offs between defenses and risks.