Explore a 20-minute conference talk from USENIX's ScAINet '20 summit that delves into the critical topic of adaptive attacks on adversarial example defenses. Join Stanford University researcher Florian Tramer as he examines the challenges and intricacies of this cybersecurity issue. Gain insights into whitebox attacks, adaptive attacks, and the reasons behind the persistent vulnerability of defense mechanisms. Discover key observations and meta points that shed light on the underlying problem. Learn why this subject matters and its implications for the field of artificial intelligence and security. Conclude with a comprehensive understanding of the current state of adversarial example defenses and the ongoing challenges they face.
Overview
Syllabus
Intro
Why should we care
Challenges
Background
Whitebox Attacks
Adaptive Attacks
Why were all defenses still broken
First Observation
Second Observation
Meta Points
The Problem
Conclusion
Taught by
USENIX
Related Courses
-
Is "Adversarial Examples" an Adversarial Example?
-
Lessons Learned from Evaluating the Robustness of Defenses to Adversarial Examples
-
Adaptive Attacks to Adversarial Example Defenses - CAP6412 Spring 2021
-
The Devil is in the GAN - Defending Deep Generative Models Against Adversarial Attacks
-
Protecting the Protector - Hardening Machine Learning Defenses Against Adversarial Attacks
