A Fast VQ Codebook Generation Algorithm Using ART2 Neural Network

We present a new fast codebook generation algorithm entitled CGART2 (Codebook Generation using ART2). ART2 is a stable and flexible neural network suitable forreal time application. This network uses relative similarity between patterns. Special attention has been paid to structuringART nets so that neural processes can control the complicatedoperation of these nets. Since we use fast learning in ART2, the data are noisy binary. For this propose, we perform apreprocessing on the data; i.e. BTC like binarization. The proposed algorithm is fundamentally different from the previousapproaches in that previous approaches focus on reducing the size of the codebook. Indeed, ART does not require that all training patterns be presented in the same order, or even with the same frequency. Our experimental results show that proposed fast codebook generation algorithm (CGART2) is up to 50% faster than the best codebook generation algorithm

Keywords:

Vector Quantization , Codebook Generation , ART2 , Neural Networks , Clustering