Can Shape Structure Features Improve Model Robustness under Diverse Adversarial Settings?

Keyphrases

• Structure Feature 100%
• Shape Structure 100%
• Model Robustness 100%
• Adversarial Setting 100%
• Edge Feature 100%
• Edge Detection 75%
• Convolutional Neural Network 75%
• Model Accuracy 50%
• Detection Method 25%
• Shape Features 25%
• Texture Information 25%
• Edge Structure 25%
• Learning Task 25%
• Data Properties 25%
• Robust Features 25%
• Attack Scenario 25%
• Vanilla 25%
• Human Visual System 25%
• Feature Optimization 25%
• Adversarial Attack 25%
• Global Structural Feature 25%
• Generative Adversarial Networks 25%
• Feature Exploration 25%
• Backdoor Attack 25%
• Common Corruptions 25%
• Local Texture Features 25%
• Corruption Attack 25%
• Edge-enabled 25%

Computer Science

• Adversarial Setting 100%
• Convolutional Neural Network 100%
• edge detection algorithm 66%
• Model Accuracy 66%
• Robust Approach 33%
• Texture Information 33%
• Attack Scenario 33%
• Edge Detection 33%
• Texture Feature 33%
• Adversarial Machine Learning 33%
• Human Visual System 33%
• Backdoors 33%
• Generative Adversarial Networks 33%