Neural networks, radial basis functions and projection pursuit regression arenonlinear models which simultaneously project the m-dimensional input data into a p-dimensional space and model nonlinear functions of the linear combinations of the inputs in this new space. Previous statistical theory for estimating the true error variance s 2 constructing approximated confidence intervals seems inappropriate, since the degrees of freedom of these models do not equal the number of adaptive parameters. We show in this article that the problem maybe overcome by using the equivalent degrees of freedom(e.d. f .) based on the dimension of the projection space. We present the results of a MonteCarlo study on simulated data showing that e.d. f . give numerical stable results and seemto work reasonably well in estimating s 2 and constructing confidence intervals

Computational studies with equivalent degrees of freedoms in neural networks

INGRASSIA, Salvatore;
2007-01-01

Abstract

Neural networks, radial basis functions and projection pursuit regression arenonlinear models which simultaneously project the m-dimensional input data into a p-dimensional space and model nonlinear functions of the linear combinations of the inputs in this new space. Previous statistical theory for estimating the true error variance s 2 constructing approximated confidence intervals seems inappropriate, since the degrees of freedom of these models do not equal the number of adaptive parameters. We show in this article that the problem maybe overcome by using the equivalent degrees of freedom(e.d. f .) based on the dimension of the projection space. We present the results of a MonteCarlo study on simulated data showing that e.d. f . give numerical stable results and seemto work reasonably well in estimating s 2 and constructing confidence intervals
2007
978-88-6056-020-9
Error variance; Equivalent degrees of freedoms; Neural networks
File in questo prodotto:
File Dimensione Formato  
IngrassiaMorlini_Cladag2007b.pdf

solo gestori archivio

Licenza: Non specificato
Dimensione 388.87 kB
Formato Adobe PDF
388.87 kB Adobe PDF   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.11769/57943
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact