Development of a neural network algorithm for unsupervised competitive learning
Dong-Chul Park · Proceedings of International Conference on Neural Networks (ICNN'97) · 2002
An unsupervised competitive learning algorithm is proposed. The proposed centroid neural network (CNN) algorithm estimates optimal centroids of the related cluster groups to each training data. The CNN is based on the classical K-means clustering algorithm. This paper also explains algorithmic relationships between the CNN and some of the conventional unsupervised competitive learning algorithms such as Kohonen's self-organization map (SOM) and Kosko's differential competitive learning (DCL). The CNN algorithm requires neither a predetermined learning coefficient schedule nor a total number of iterations. The simulation results from an image compression problem show that the CNN converges much faster than SOM or DCL with compatible compression error.