Neural Network Security: Live Demonstration of an Adversarial Attack
summary
Overview of adversarial attacks against neural networks, explaining how small, targeted perturbations to inputs can make embedded real-time AI systems (like ADAS) misclassify. Covers different adversarial techniques and presents a new black-box attack method, with a live realistic demonstration on an automotive driver-assistance system.
AI-written summary · about
chapters (8)
- 0:00Adversarial attack introduction
- 1:22Neural network basics
- 8:22Training and vulnerabilities
- 16:32Adversarial perturbation principle
- 18:57Autonomous vehicle attacks
- 21:45Stop sign demo
- 31:10Realistic black-box attack
- 39:14Final demo and conclusion
related talks
36mAdversarial Attacks on Deep Learning
46mA Short Introduction to Hacking Artificial Intelligence
44mAdversarial attacks on deep learning models
21mTricking Neural Networks : Explore Adversarial Attacks
38mUsing adversarial samples to robustify your Neural Network Models
21mAdversarial Attacks
35mHacking AI Systems: How to (Still) Trick Artificial Intelligence
25mTricking Neural Networks : Explore Adversarial Attacks