FIST: fast iterative self-structuring and training of artificial neural networks

Author

Garvin, A.D.M. ; Rayner, P.J.W.

Author_Institution

Dept. of Eng., Cambridge Univ., UK

fYear

1994

fDate

13-16 Apr 1994

Firstpage

377

Abstract

Four new algorithms are presented to perform self-structuring of generalized single layer networks. Given a large set of nonlinear basis functions and a problem specified by the training data, the algorithms determine the optimal subset of the basis functions which may be used by the network to solve the problem. The algorithms are all fast as they use iterative update techniques to remove the need for calculating matrix inverses. Calculating which terms should be added to the network or removed from the network is done in parallel. Bayesian model selection is used to determine the optimal basis set. Theoretical and experimental results are presented

Keywords

Bayes methods; iterative methods; learning (artificial intelligence); optimisation; self-organising feature maps; Bayesian model selection; FIST; artificial neural network training; fast iterative self-structuring; generalized single layer networks; iterative update techniques; neural network self-structuring; nonlinear basis functions; optimal basis set; Artificial neural networks; Bayesian methods; Ear; Image processing; Iterative algorithms; Kernel; Neural networks; Polynomials; Speech processing; Training data;

fLanguage

English

Publisher

ieee

Conference_Titel

Speech, Image Processing and Neural Networks, 1994. Proceedings, ISSIPNN '94., 1994 International Symposium on

Print_ISBN

0-7803-1865-X

Type

conf

DOI

10.1109/SIPNN.1994.344888

Filename

344888