Hyperplane Arrangements of Trained ConvNets Are Biased.
Matteo GambaStefan CarlssonHossein AzizpourMårten BjörkmanPublished in: CoRR (2020)
Keyphrases
- hyperplane
- feed forward
- svm classifier
- support vector machine
- feature space
- linear classifiers
- input space
- support vector
- training samples
- data points
- training set
- linearly separable
- maximal margin
- principal components
- kernel space
- support vector machine svm
- back propagation
- kernel function
- high quality
- classification procedure
- convex hull
- locality sensitive
- average distance
- artificial neural networks
- incremental learning algorithm
- normal vectors
- neural network
- nearest neighbor
- feature vectors
- linear separability