Decoupling direction and norm for efficient gradient-based l2 adversarial attacks and defenses

Keyphrases

• Adversarial Attack 100%
• Gradient-based 100%
• Adversarial Defense 100%
• L2 Norm 100%
• Anti-attack 100%
• MNIST 66%
• Misclassification 66%
• CIFAR-10 Dataset 66%
• Line Search 33%
• Adversarial Training 33%
• Adversarial Examples 33%
• Fewer Iterations 33%
• Adversarial Perturbation 33%
• Computer Vision Tasks 33%
• White-box Attack 33%
• ImageNet Dataset 33%
• Maximum Norm 33%
• Security Implications 33%
• Image Processing System 33%
• White-box 33%

Computer Science

• Adversarial Machine Learning 100%
• Image Processing 50%
• Adversarial Example 50%
• Processing System 50%
• Computer Vision Task 50%
• Security Implication 50%