A sweet rabbit hole by DARCY: Using honeypots to detect universal trigger's adversarial attacks

Keyphrases

• Sweet 100%
• Honeypot 100%
• Adversarial Attack 100%
• Universal Triggers 100%
• Neural Network Model 50%
• Prediction Accuracy 50%
• Trapdoor 50%
• Near-zero 25%
• Attacker 25%
• Knowledge Level 25%
• Skill Level 25%
• Source Code 25%
• Learning-based 25%
• Attack Strategy 25%
• Attack Scenario 25%
• Cybersecurity 25%
• Comprehensive Experiment 25%
• Public Dataset 25%
• Significant Harm 25%
• Potential Attacks 25%
• Defense Framework 25%

Computer Science

• Honeypots 100%
• Adversarial Machine Learning 100%
• Neural Network Model 66%
• Prediction Accuracy 66%
• Attackers 33%
• Cybersecurity 33%
• Attack Scenario 33%
• Potential Attack 33%