An Information-Theoretic Explanation for the Adversarial Fragility of AI Classifiers.
Hui XieJirong YiWeiyu XuRaghu MudumbaiPublished in: ISIT (2019)
Keyphrases
- information theoretic
- mutual information
- information theory
- training data
- theoretic framework
- information theoretic measures
- information bottleneck
- feature selection
- entropy measure
- support vector
- decision trees
- training set
- kullback leibler divergence
- jensen shannon divergence
- multi modality
- log likelihood
- computational learning theory
- feature set
- nearest neighbor
- pattern recognition
- minimum description length
- training samples
- image classification
- multi class