Everything You Need to Know about Security Issues in Today's ML Systems

Explore the critical security issues in modern machine learning systems through this comprehensive 25-minute conference talk. Gain essential knowledge for ML practitioners, including an overview of potential vulnerabilities like poisoning, evasion, and inversion attacks. Focus on test-time vulnerabilities, particularly adversarial examples, and understand their potential negative consequences. Examine real-world attacks on ML as a service platforms, face recognition systems, autonomous vehicles, and voice assistants. Learn to distinguish between genuine threats and less concerning issues, equipping yourself with practical insights for developing more secure ML systems.