MLSec
Machine Learning Security and Privacy
Our work in this theme is along two dimensions:
- How to effectively apply machine learning techniques to address difficult security and privacy problems? Our work has ranged from detecting phishing websites and making security/privacy mechanisms easy to use.
- Understanding security/privacy concerns inherent in machine learning applications in general and developing ways to mitigate these concerns. Our work addresses concerns like privacy-preserving predictions and guarding against model extraction attacks.
Current Projects
- Unintended interactions among ML defenses and risks
- Model extraction attacks and defenses
- Machine learning property attestations
- Unacceptable concept filtering in text-to-image model
Past Projects
- Automated generation of deceptive text
- Privacy-preserving machine learning predictions
- Deception detection via text analysis
- Automated detection of organized eCommerce fraud
- Model evasion attacks and defenses