Stealing Knowledge from Protected Deep Neural Networks Using Composite Unlabeled Data.
Itay MosafiEli DavidNathan S. NetanyahuPublished in: CoRR (2019)
Keyphrases
- unlabeled data
- labeled data
- semi supervised learning
- semi supervised
- neural network
- active learning
- co training
- prior knowledge
- text classification
- training data
- supervised learning
- semi supervised classification
- domain adaptation
- learning algorithm
- labeled examples
- text categorization
- labeled training data
- data points
- class labels
- training examples
- labeled and unlabeled data
- positive examples
- pattern recognition
- supervised learning algorithms
- machine learning
- number of labeled examples
- semisupervised learning
- knowledge transfer
- unsupervised learning
- knowledge representation
- object recognition
- transfer learning
- background knowledge
- maximum likelihood
- d objects
- training set
- sample selection bias
- similarity measure
- small set of labeled