EAD: an ensemble approach to detect adversarial examples from the hidden features of deep neural networks.
Francesco CraigheroFabrizio AngaroniFabio StellaChiara DamianiMarco AntoniottiAlex GraudenziPublished in: CoRR (2021)
Keyphrases
- neural network
- training examples
- feature vectors
- detection method
- feature extraction
- genetic algorithm
- image features
- feature detection
- co occurrence
- fuzzy logic
- feature space
- pattern recognition
- classification accuracy
- low level
- sufficient conditions
- multi agent
- training data
- back propagation
- false positives
- self organizing maps
- high level
- distinctive features