"Cybersecurity of Autonomous Vehicles" | Purdue CS529: Security Analytics, Fall 2024 скачать в хорошем качестве

"Cybersecurity of Autonomous Vehicles" | Purdue CS529: Security Analytics, Fall 2024

We had a video assignment to explain a paper/concept related to the class's focus at the intersection of machine learning and security. I chose to learn all about autonomous vehicles! Timestamps: 0:00 Security of AVs is basically security of ML classification models 0:18 Source paper 1:10 Table of Contents 2:48 Motivations - growing reliance on ML/AI in AVs = growing risk 6:00 Vocabulary 8:30 Levels of automation 13:17 Spoiler: AV pipeline is the ML attack surface 17:25 Adversarial ML, as it applies to AVs! 19:33 [COOL MATH ALERT] Gradient Descent method for making adversarial examples! 22:31 Method 2: use GANs 23:52 Transferability of adversarial examples (a magical property)! 24:58 How to ATTACK an autonomous vehicle 26:23 Example attack paper: increase transferability of traffic sign sensor spoofing 29:47 How to DEFEND 31:47 Example defense paper: improve input reconstruction cleaning of sketchy images 32:48 Autoencoders - a brief tangent 35:05 Back to the defense paper 37:55 Fuzzing - beyond adversarial examples 39:00 Conclusion - security of AVs is basically security of ML classification models