Adversarial Noises Are Linearly Separable for (Nearly) Random Neural Networks.
Huishuai ZhangDa YuYiping LuDi HePublished in: AISTATS (2023)
Keyphrases
- linearly separable
- neural network
- linear separability
- input space
- linear classifiers
- high dimensional feature space
- hyperplane
- feature space
- pattern recognition
- data points
- kernel function
- back propagation
- convex hull
- infinite dimensional
- kernel methods
- decision boundary
- training process
- machine learning
- soft margin
- decision trees
- sample set
- high dimensional