Title :
Fast competitive learning algorithm for image compression neural networks
Author_Institution :
Dept. of Comput. Studies, Loughborough Univ. of Technol., UK
fDate :
7/18/1996 12:00:00 AM
Abstract :
A fast training algorithm for competitive learning neural networks is presented. The algorithm identifies the full Euclidean distance calculation as the major bottleneck. Through theoretical analysis, a simple approximate distance is derived and used as the pre-test to exclude most of the neurons in competitive learning. This provides significant efficiency improvement over the standard algorithm
Keywords :
data compression; image coding; neural nets; unsupervised learning; Euclidean distance; competitive learning; fast algorithm; image compression; neural network; training;
Journal_Title :
Electronics Letters
DOI :
10.1049/el:19960891
