A Subspace Projective Clustering Approach for Backdoor Attack Detection and Mitigation in Deep Neural Networks

Keyphrases

• Clustering Approach 100%
• Attack Detection 100%
• Deep Neural Network 100%
• Backdoor Attack 100%
• Attack Mitigation 100%
• Projective Clustering 100%
• Attacker 66%
• Weight Vector 66%
• State-of-the-art Techniques 33%
• Projection-based 33%
• Training Data 33%
• Backdoor 33%
• Neuron Activation 33%
• ImageNet 33%
• Misclassification 33%
• German Traffic Sign Recognition Benchmark (GTSRB) 33%
• Clustering Framework 33%
• Optimized Weights 33%
• Differences of Projections 33%
• VGGFace2 33%

Computer Science

• clustering approach 100%
• Deep Neural Network 100%
• Backdoors 100%
• Attackers 50%
• Experimental Result 25%
• Training Dataset 25%