Revealing Backdoors, Post-Training, in DNN Classifiers via Novel Inference on Optimized Perturbations Inducing Group Misclassification.
Zhen XiangDavid J. MillerGeorge KesidisPublished in: ICASSP (2020)
Keyphrases
- training process
- training set
- test set
- training data
- training examples
- training samples
- structured prediction
- classifier training
- decision trees
- class labels
- trained classifiers
- labeled training data
- naive bayes
- bayesian networks
- classification algorithm
- supervised training
- training set size
- classification models
- training stage
- linear svm
- multi layer perceptron
- feature selection
- training and testing data
- svm classifier
- error rate
- support vector
- supervised learning
- misclassification rate
- neural network
- boosted classifiers
- discriminative classifiers
- training phase
- linear classifiers
- bayesian inference
- active learning
- feature space
- machine learning