Is AI Taking Over the World? No, but It's Making it Less Private

Abstract

This talk highlights challenges and opportunities for trustworthy AI with a focus on privacy attacks and countermeasures. AI and machine learning have no future if their privacy and security concerns are not addressed. Machine learning models could hide malicious code or back doors, and leak private information about users. We will explore inference attacks against machine learning models and frameworks (e.g., federated learning), and set out the requirements for privacy-preserving AI systems.