Adaptive k-means algorithm with error-weighted deviation measure

Author

Chinrungrueng, Chedsada ; Séquin, Carlo H.

Author_Institution

Dept. of Electr. Eng. & Comput. Sci., California Univ., Berkeley, CA, USA

fYear

1993

fDate

1993

Firstpage

626

Abstract

The k-means algorithm can be used in multi-module networks to partition the input domain of supervised learning problems. The traditional k-means algorithm partitions the input domain based solely on the distribution of the input vectors. A modified algorithm is presented. It also integrates into its partitioning process information about the mismatch between the network function and the goal function. It uses an efficient adaptive learning rate and an error-weighted squared Euclidean distance measure that aims at equalizing the average approximation errors in all regions of the partition

Keywords

adaptive systems; learning (artificial intelligence); neural nets; adaptive k-means algorithm; adaptive learning rate; error-weighted deviation; error-weighted squared Euclidean distance; multiple module networks; neural nets; supervised learning; Adaptive equalizers; Clustering algorithms; Computer errors; Computer networks; Electric variables measurement; Euclidean distance; Least squares approximation; Partitioning algorithms; Supervised learning; Transfer functions;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks, 1993., IEEE International Conference on

Conference_Location

San Francisco, CA

Print_ISBN

0-7803-0999-5

Type

conf

DOI

10.1109/ICNN.1993.298627

Filename

298627