Advances in Learning Theory: Methods, Models and Applications

0404817 - UIVT-O 20030182 RIV NL eng M - Část monografie knihy
Kůrková, Věra
High-Dimensional Approximation by Neural Networks.
Advances in Learning Theory: Methods, Models and Applications. Amsterdam: IOS Press, 2003 - (Suykens, J.; Horváth, G.; Basu, S.; Micchelli, C.; Vandewalle, J.), s. 69-88. NATO Science Series, 190. ISBN 1-58603-341-7
Grant CEP: GA ČR GA201/02/0428
Výzkumný záměr: AV0Z1030915
Klíčová slova: neural network learning * regularized empirical error functions * high-dimensional approximation
Kód oboru RIV: BA - Obecná matematika

Approximation of high-dimensional mappings by neural networsk is investigated in the context of nonlinear approximation theory.It is shown that the "curse of dimensionality" can be avoided when functions to be approximated have small special norms, which are tailored to the type of computational units. Properties of such norms and method of derivation of their estimates are described. Estimates of rates of nonlinear approximation are applied to neural network learning formalized as approximate minimization.
Trvalý link: http://hdl.handle.net/11104/0125050