Bayesian information criteria and smoothing parameter selection in radial basis function networks

Sadanori Konishi, Tomohiro Ando, Seiya Imoto

Research output: Contribution to journalArticle

70 Citations (Scopus)

Abstract

By extending Schwarz's (1978) basic idea we derive a Bayesian information criterion which enables us to evaluate models estimated by the maximum penalised likelihood method or the method of regularisation. The proposed criterion is applied to the choice of smoothing parameters and the number of basis functions in radial basis function network models. Monte Carlo experiments were conducted to examine the performance of the nonlinear modelling strategy of estimating the weight parameters by regularisation and then determining the adjusted parameters by the Bayesian information criterion. The simulation results show that our modelling procedure performs well in various situations.

Original languageEnglish
Pages (from-to)27-43
Number of pages17
JournalBiometrika
Volume91
Issue number1
DOIs
Publication statusPublished - 2004
Externally publishedYes

Fingerprint

Bayesian Information Criterion
Radial basis function networks
Radial Basis Function Network
Parameter Selection
Smoothing Parameter
Regularization
Nonlinear Modeling
Likelihood Methods
Monte Carlo Experiment
Maximum likelihood
Network Model
Basis Functions
Evaluate
Modeling
Weights and Measures
Simulation
Experiments
Smoothing
Radial basis function
Bayesian information criterion

Keywords

  • Bayes approach
  • Maximum penalised likelihood
  • Model selection
  • Neural network
  • Nonlinear regression

ASJC Scopus subject areas

  • Agricultural and Biological Sciences(all)
  • Agricultural and Biological Sciences (miscellaneous)
  • Statistics and Probability
  • Mathematics(all)
  • Applied Mathematics

Cite this

Bayesian information criteria and smoothing parameter selection in radial basis function networks. / Konishi, Sadanori; Ando, Tomohiro; Imoto, Seiya.

In: Biometrika, Vol. 91, No. 1, 2004, p. 27-43.

Research output: Contribution to journalArticle

Konishi, Sadanori ; Ando, Tomohiro ; Imoto, Seiya. / Bayesian information criteria and smoothing parameter selection in radial basis function networks. In: Biometrika. 2004 ; Vol. 91, No. 1. pp. 27-43.
@article{58c362c102334eb4aadcdf9dc19ad43c,
title = "Bayesian information criteria and smoothing parameter selection in radial basis function networks",
abstract = "By extending Schwarz's (1978) basic idea we derive a Bayesian information criterion which enables us to evaluate models estimated by the maximum penalised likelihood method or the method of regularisation. The proposed criterion is applied to the choice of smoothing parameters and the number of basis functions in radial basis function network models. Monte Carlo experiments were conducted to examine the performance of the nonlinear modelling strategy of estimating the weight parameters by regularisation and then determining the adjusted parameters by the Bayesian information criterion. The simulation results show that our modelling procedure performs well in various situations.",
keywords = "Bayes approach, Maximum penalised likelihood, Model selection, Neural network, Nonlinear regression",
author = "Sadanori Konishi and Tomohiro Ando and Seiya Imoto",
year = "2004",
doi = "10.1093/biomet/91.1.27",
language = "English",
volume = "91",
pages = "27--43",
journal = "Biometrika",
issn = "0006-3444",
publisher = "Oxford University Press",
number = "1",

}

TY - JOUR

T1 - Bayesian information criteria and smoothing parameter selection in radial basis function networks

AU - Konishi, Sadanori

AU - Ando, Tomohiro

AU - Imoto, Seiya

PY - 2004

Y1 - 2004

N2 - By extending Schwarz's (1978) basic idea we derive a Bayesian information criterion which enables us to evaluate models estimated by the maximum penalised likelihood method or the method of regularisation. The proposed criterion is applied to the choice of smoothing parameters and the number of basis functions in radial basis function network models. Monte Carlo experiments were conducted to examine the performance of the nonlinear modelling strategy of estimating the weight parameters by regularisation and then determining the adjusted parameters by the Bayesian information criterion. The simulation results show that our modelling procedure performs well in various situations.

AB - By extending Schwarz's (1978) basic idea we derive a Bayesian information criterion which enables us to evaluate models estimated by the maximum penalised likelihood method or the method of regularisation. The proposed criterion is applied to the choice of smoothing parameters and the number of basis functions in radial basis function network models. Monte Carlo experiments were conducted to examine the performance of the nonlinear modelling strategy of estimating the weight parameters by regularisation and then determining the adjusted parameters by the Bayesian information criterion. The simulation results show that our modelling procedure performs well in various situations.

KW - Bayes approach

KW - Maximum penalised likelihood

KW - Model selection

KW - Neural network

KW - Nonlinear regression

UR - http://www.scopus.com/inward/record.url?scp=3242882795&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=3242882795&partnerID=8YFLogxK

U2 - 10.1093/biomet/91.1.27

DO - 10.1093/biomet/91.1.27

M3 - Article

VL - 91

SP - 27

EP - 43

JO - Biometrika

JF - Biometrika

SN - 0006-3444

IS - 1

ER -