Gauss-Newton approximation to Bayesian learning

This paper describes the application of Bayesian regularization to the training of feedforward neural networks. A Gauss-Newton approximation to the Hessian matrix, which can be conveniently implemented within the framework of the Levenberg-Marquardt algorithm, is used to reduce the computational ove...

Full description

Saved in:
Bibliographic Details
Published in1997 IEEE International Conference on Neural Networks Vol. 3; pp. 1930 - 1935 vol.3
Main Authors Dan Foresee, F., Hagan, M.T.
Format Conference Proceeding
LanguageEnglish
Japanese
Published IEEE 1997
Subjects
Application software
Bayesian methods
Cities and towns
Computer networks
Feedforward neural networks
Least squares methods
Neural networks
Newton method
Recursive estimation
Testing
Online AccessGet full text
ISBN0780341228
9780780341227
DOI10.1109/ICNN.1997.614194

Cover

More Information
Summary:This paper describes the application of Bayesian regularization to the training of feedforward neural networks. A Gauss-Newton approximation to the Hessian matrix, which can be conveniently implemented within the framework of the Levenberg-Marquardt algorithm, is used to reduce the computational overhead. The resulting algorithm is demonstrated on a simple test problem and is then applied to three practical problems. The results demonstrate that the algorithm produces networks which have excellent generalization capabilities.
ISBN:0780341228
9780780341227
DOI:10.1109/ICNN.1997.614194