Towards A Conceptually Simple Defensive Approach for Few-shot classifiers Against Adversarial Support Samples.

Keyphrases

• Defensive Strategy 100%
• Few-shot 100%
• Detection Method 50%
• Attacker 25%
• Novel Class 25%
• Machine Learning 25%
• Non-overlapping 25%
• Two-component 25%
• Self-concept 25%
• Self-similarity 25%
• Detection Performance 25%
• Overlapping Sets 25%
• Support Set 25%
• Attack Detection 25%
• Agnostic 25%
• Adversarial Attack 25%
• Deep Network 25%
• Poisoning Attack 25%
• Self-filtering 25%
• Integrity Attacks 25%
• Extended Evaluation 25%

Computer Science

• Attackers 100%
• Use Case 100%
• Machine Learning 100%
• Detection Method 100%
• Detection Performance 100%
• Adversarial Machine Learning 100%