Iteratively reweighted least squares based learning

Author

Warner, Bradley A. ; Misra, Manavendra

Author_Institution

Dept. of Math. Sci., US Air Force Acad., Colorado Springs, CO, USA

Volume

2

fYear

1998

fDate

4-9 May 1998

Firstpage

1327

Abstract

We demonstrate a method to obtain maximum likelihood weight estimates for a multi-layered feedforward neural network using least squares. The proposed method uses the Fisher´s information matrix instead of the Hessian matrix to compute the search direction. Since this matrix is formulated as an inner product, it is guaranteed to be positive definite. The formulation used by the method also provides an interesting way of highlighting the multicollinearity problem in multilayered feedforward networks

Keywords

feedforward neural nets; iterative methods; learning (artificial intelligence); least squares approximations; matrix algebra; maximum likelihood estimation; multilayer perceptrons; Fisher´s information matrix; inner product; iteratively reweighted least squares based learning; maximum likelihood weight estimates; multi-layered feedforward neural network; multicollinearity problem; positive definite matrix; search direction; Artificial neural networks; Convergence; Feedforward neural networks; Feedforward systems; Least squares approximation; Least squares methods; Maximum likelihood estimation; Military computing; Multi-layer neural network; Neural networks;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks Proceedings, 1998. IEEE World Congress on Computational Intelligence. The 1998 IEEE International Joint Conference on

Conference_Location

Anchorage, AK

ISSN

1098-7576

Print_ISBN

0-7803-4859-1

Type

conf

DOI

10.1109/IJCNN.1998.685967

Filename

685967