Gaussian kernels on non-simply-connected closed Riemannian manifolds are never positive definite.
Siran LiPublished in: CoRR (2023)
Keyphrases
- positive definite
- reproducing kernel hilbert space
- riemannian manifolds
- euclidean space
- kernel function
- loss function
- kernel methods
- geometric structure
- learning theory
- special case
- kernel matrix
- density estimation
- real valued
- domain adaptation
- input space
- data dependent
- covariance matrix
- gaussian process
- learning problems
- distance measure
- feature space
- low dimensional
- support vector
- shape analysis
- linear model
- vector space
- covariance matrices
- image processing
- machine learning
- training samples
- semi supervised learning
- euclidean distance
- support vectors
- support vector machine