Salgorithm,an adaptive versioncompetitiveoflearningthe classical(CL) has ak-meansnumberclusteringof
applications. First, it can function as an adaptive method for
clustering analysis problems encountered in statistical data
analysisor unsupervised pattern recognition [I]. Second,it can
be used for vector quantizationwhich is widely used in image
processing and speech signal processing for compressing data
and message coding [2], [3]. Third, it has also been recently
incorporated into some supervised learning methods for training multilayer feedforward nets more effectively,e.g., into the
radial basis function (RBF) nets for locating the centers of
Gaussian receptive fields [4]-[6].
However, it was found by Rumelhart et al. [7], Grossberg [8], and Hecht-Nielsen [8] that the simple classical
CL algorithm has the so called under-utilized or dead unit
problem. Many efforts have been made to solve the problem.
Grossberg’sART series [7], [lo]-[13] and Kohonen Map [14],
[15] are two main developments of the classical competitive
learning.ART provides a stable model for unsupervised pattern
Salgorithm,an adaptive versioncompetitiveoflearningthe classical(CL) has ak-meansnumberclusteringofapplications. First, it can function as an adaptive method forclustering analysis problems encountered in statistical dataanalysisor unsupervised pattern recognition [I]. Second,it canbe used for vector quantizationwhich is widely used in imageprocessing and speech signal processing for compressing dataand message coding [2], [3]. Third, it has also been recentlyincorporated into some supervised learning methods for training multilayer feedforward nets more effectively,e.g., into theradial basis function (RBF) nets for locating the centers ofGaussian receptive fields [4]-[6].However, it was found by Rumelhart et al. [7], Grossberg [8], and Hecht-Nielsen [8] that the simple classicalCL algorithm has the so called under-utilized or dead unitproblem. Many efforts have been made to solve the problem.Grossberg’sART series [7], [lo]-[13] and Kohonen Map [14],[15] are two main developments of the classical competitivelearning.ART provides a stable model for unsupervised pattern
การแปล กรุณารอสักครู่..