Premium
Construction of self‐organizing algorithms for vector quantization
Author(s) -
Maeda Michiharu,
Miyajima Hiromi,
Murashima Sadayuki
Publication year - 1999
Publication title -
electrical engineering in japan
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.136
H-Index - 28
eISSN - 1520-6416
pISSN - 0424-7760
DOI - 10.1002/(sici)1520-6416(19990415)127:1<47::aid-eej6>3.0.co;2-7
Subject(s) - vector quantization , learning vector quantization , linde–buzo–gray algorithm , artificial neural network , algorithm , quantization (signal processing) , computer science , weight , distortion (music) , artificial intelligence , mathematics , amplifier , computer network , bandwidth (computing) , lie algebra , pure mathematics
Vector quantization is used for both storage and transmission of speech and image data, and an algorithm that minimizes the distortion error is often required. To obtain the minimum distortion error in neural networks for vector quantization, corrective competitive learning has been introduced. In a large number of algorithms, self‐creating neural networks and self‐deleting neural networks have performed well. In this paper, we improve the self‐deleting neural network and propose a generalized algorithm combining the creating and deleting neural networks. First, a few weight (reference) vectors are prepared, the self‐creating algorithm is applied, and vectors are created automatically. Next, the self‐deleting algorithm is applied, and weight vectors are deleted sequentially to reach a predetermined number. Experimental results show the effectiveness of the proposed algorithm. © 1999 Scripta Technica, Electr Eng Jpn, 127(1): 47–55, 1999