Soft-competitive-growing classifier with unsupervised fine-tuning

Author

Alba, José L. ; Docío, Laura ; Ruibal, Simón

Author_Institution

Dept. de Tecnol. de las Commun., Vigo Univ., Spain

Volume

3

fYear

1997

fDate

9-12 Jun 1997

Firstpage

1418

Abstract

In this paper a method for growing a Gaussian-mixture-based network is developed. The constructive technique is based on an EM algorithm to estimate the parameters and the number of nodes is iteratively increased by means of discriminant placement. The growth control is imposed by an information theoretic criterion that prevents the network from becoming extremely complex and losing generalization capabilities. After the growing phase is finished, another EM algorithm is used with labeled and unlabeled data in order to fine-tune network parameters. This solution improves the test-performance for the applications where labeled data is insufficient and the classes are not highly overlapped. We report results on some artificially generated examples and on terrain classification over a Landsat-TM image

Keywords

feedforward neural nets; image classification; information theory; iterative methods; maximum likelihood estimation; tuning; unsupervised learning; AIC; EM algorithm; Gaussian-mixture-based network; Landsat-TM image; growth control; information theory; iterative method; parameter estimation; radial basis function neural net; soft-competitive-growing classifier; terrain classification; unsupervised fine-tuning; Clustering algorithms; Euclidean distance; Gaussian processes; Interpolation; Iterative algorithms; Neural networks; Parameter estimation; Remote sensing; Satellites; Testing;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks,1997., International Conference on

Conference_Location

Houston, TX

Print_ISBN

0-7803-4122-8

Type

conf

DOI

10.1109/ICNN.1997.614002

Filename

614002