Surface Fitting via Radial and Related Basis Functions with Applications to Neural Networks
Abstract
Progress was made along several fronts during this period. Classes of RBF, periodic, and nonstationary spherical wavelets were constructed, all being capable of synthesizing and analyzing scattered data. Shape preservation problems were investigated. A framework for obtaining rates of approximation in non-traditional settings and data - i.e. compact manifolds and generalized Hermite data, was provided. This framework was used to derive approximation rates for various classes of functions on the circle and the 2-sphere. In particular, for a broad class of functions, rates of approximation were obtained for scattered data on the m-sphere. Stability questions related to these classes of functions were studied. Results from the work above were applied to neural networks.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 25, 1998
- Accession Number
- ADA342010
Entities
People
- Francis J. Narcowich
- Joseph D. Ward
Organizations
- Texas A&M University