L-Red: Efficient Post-Training Detection of Imperceptible Backdoor Attacks Without Access to the Training Set.
Zhen XiangDavid J. MillerGeorge KesidisPublished in: ICASSP (2021)
Keyphrases
- training set
- test set
- training examples
- supervised learning
- watermarking scheme
- training samples
- training patterns
- nearest neighbor
- false positives
- data sets
- cross validation
- training data
- detection algorithm
- active learning
- classification accuracy
- feature space
- access control
- classification algorithm
- decision trees
- detection method
- class labels
- support vector machine
- test data
- object classes
- training dataset
- countermeasures