Deterministic error analysis of kernel regularized regression for spherical scattered data
When scattered data are approximated by deterministic methods, four aspects need to be considered carefully: namely (i) the geometry of the underlying set of placement of the data, (ii) the kernel methods that are to be applied including the particular choice of radial basis functions, (iii) the correct \textit{Ansatz} for the learning problem, and (iv) the use of the right quadrature rules.\N\NThis article addresses all these questions and supplies good error bounds for the approximations by learning and regression methods using the celebrated approximation methods by radial basis function kernels. The authors are experts for the mentioned quadrature rules on general domains too.\N\NFor this, the distance functions that are combined with mentioned kernel functions (which are formed from a radial basis functions \(\phi\), say, convolved with a distance function, call it \(d(\cdot,\cdot)\), so that we approximate by linear combinations of \(\phi(d(x,y))\)) have to be chosen judiciously.\N\NThe mentioned error bounds for the approximations by learning are identified by using the fact that the radial basis functions in use are reproducing kernels in reproducing kernel Hilbert spaces. Those spaces are also called native spaces and they are available for all radial basis functions on spheres which are (strictly) positive definite. It is a typical observation (and considered with great detail in the article) that the quality of the approximations depends on the smoothness of the radial basis function. This is because the weight functions in the Hilbert space (semi-)norms depend on the asymptotic properties of the Fourier transforms of the radial functions, and they depend in turn on the smoothness of the kernel function.\N\NThe regularization of the approximations are carried by our generalized cross-validation. This is in order to avoid over-sampling which is a routine problem with interpolation (for which otherwise radial basis functions are standard choices).
- LpError Estimates for Scattered Data Interpolation On Spheres
- \(L_p\) error estimate for minimal norm SBF interpolation
- A high-order norm-product regularized multiple kernel learning framework for kernel optimization
- Complexity of numerical integration over spherical caps in a Sobolev space setting
- Convergence analysis of deterministic kernel-based quadrature rules in misspecified settings
- Direct and inverse Sobolev error estimates for scattered data interpolation via spherical basis functions
- Introduction to Riemannian Manifolds
- Kernel techniques: From machine learning to meshless methods
- Learning a function from noisy samples at a finite sparse set of points
- Learning rates of least-square regularized regression
- Learning rates of regularized regression on the unit sphere
- Learning Theory
- Multivariate polynomial inequalities with respect to doubling weights and \(A_{\infty}\) weights
- New averaging technique for approximating weighted integrals
- Nonparametric regression using needlet kernels for spherical data
- Numerical integration over spheres of arbitrary dimension
- Numerical integration with polynomial exactness over a spherical cap
- On the K-functional in learning theory
- Optimality of the rescaled pure greedy learning algorithms
- Positive cubature formulas and Marcinkiewicz-Zygmund inequalities on spherical caps
- Regularized kernel-based reconstruction in generalized Besov spaces
- Stability results for scattered-data interpolation on Euclidean spheres
- Support Vector Machines
- The convergence rates of Shannon sampling learning algorithms
- Theory of Reproducing Kernels
This page was built for publication: Deterministic error analysis of kernel regularized regression for spherical scattered data
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6962156)
