PatchUp: A Feature-Space Block-Level Regularization Technique for Convolutional Neural Networks.
Mojtaba FaramarziMohammad AminiAkilesh BadrinaaraayananVikas VermaSarath ChandarPublished in: AAAI (2022)
Keyphrases
- convolutional neural networks
- feature space
- convolutional network
- high dimensionality
- feature vectors
- feature selection
- mean shift
- high dimensional
- data points
- image representation
- input space
- principal component analysis
- dimensionality reduction
- training samples
- classification accuracy
- mercer kernels
- training set
- regularization parameter
- feature extraction
- dimension reduction
- kernel function
- kernel methods
- regularization term
- parameter selection
- riemannian manifolds
- high dimensional feature space
- optimal kernel
- input data
- sparse coding
- linear discriminant analysis
- high dimension
- regularization framework
- visual words