MM-BD: Post-Training Detection of Backdoor Attacks with Arbitrary Backdoor Pattern Types Using a Maximum Margin Statistic

Keyphrases

• Maximum Margin 100%
• Backdoor 100%
• Pattern Type 100%
• Backdoor Attack 100%
• Embedding Function 27%
• Attacker 18%
• Learning Classifiers 9%
• State-of-the-art Techniques 9%
• Number of Sources 9%
• Training Set 9%
• Attack Scenario 9%
• Attack Detection 9%
• Mitigation Approaches 9%
• Embedding Mechanism 9%
• Anomaly Detector 9%
• Deep Neural Network Classifier 9%
• Softmax Layer 9%

Computer Science

• Backdoors 100%
• Attackers 11%
• Deep Neural Network 5%
• Arbitrary Number 5%
• Anomaly Detector 5%