When not to classify: Anomaly detection of attacks (ADA) on DNN classifiers at test time

Computer Science

• Anomaly Detection 100%
• Deep Neural Network 100%
• Attackers 40%
• Detection Accuracy 20%
• Detection Rate 20%
• Application Domain 20%
• Performance Measure 20%
• Network Layer 20%
• Reverse Engineering 20%
• Image Database 20%
• Attack Strategy 20%
• Classification Accuracy 20%
• Multiple Measure 20%
• Application Scenario 20%
• Detection Method 20%
• Confusion Matrix 20%
• Learning System 20%
• Machine Learning 20%
• Leibler Divergence 20%
• Decision Statistic 20%
• Anomaly Detector 20%
• ReLU Function 20%

Keyphrases

• Deep Neural Network Classifier 100%
• Evasion Attack 50%
• Curvature Detection 25%
• Neural Network Weights 25%
• Matrix Neural Network 25%
• CIFAR 25%
• CW Attack 25%
• Attack Strength 25%
• Class Uncertainty 25%
• Rectified Linear Unit (ReLU) 25%