A Unified Detection Framework for Inference-Stage Backdoor Defenses

Keyphrases

• Detection Framework 100%
• Backdoor Attack 100%
• Unified Detection 100%
• False Positive Rate 40%
• Detection Power 40%
• Detection Rules 40%
• Backdoor 40%
• Computer Vision 20%
• Natural Language Processing 20%
• Benchmark Dataset 20%
• Application-oriented 20%
• Detection Problem 20%
• Effective Approach 20%
• State-of-the-art Techniques 20%
• Real-world Application 20%
• ROC-AUC 20%
• Optimal Detection 20%
• Latent Representation 20%
• Learning Scenario 20%
• Vision-language 20%
• Deep Nets 20%
• Backdoor Trigger 20%
• Classical Learning 20%
• Specific Behavior 20%
• Unified Inference 20%
• Backdoor Detection 20%

Computer Science

• Backdoors 100%
• Backdoor Attack 80%
• False Positive Rate 40%
• Natural Language Processing 20%
• World Application 20%
• Effective Approach 20%
• Computer Vision 20%