A new Backdoor Attack in CNNs by training set corruption without label poisoning.
Mauro BarniKassem KallasBenedetta TondiPublished in: CoRR (2019)
Keyphrases
- image labeling
- training set
- data corruption
- test set
- training data
- classification accuracy
- cellular neural networks
- cross validation
- training samples
- supervised learning
- countermeasures
- classification algorithm
- active learning
- nearest neighbor
- data sets
- training examples
- secret key
- machine learning
- test data
- unlabeled data
- model selection
- feature space
- decision trees
- clustering algorithm
- error rate
- test images
- class distribution
- decision boundary
- website
- search engine
- training patterns
- pairwise