Efficient implementation of dynamic hierarchical structure for ART2 network

In traditional ART2 network, class representation is contained in a F2 node instead of being dispersed on several nodes. As a result, any harm to that F2 node leads to the loss of that class. The proposed network is capable of maintaining categories in a dynamic hierarchical structure. The new architecture is characterized by: (a) maintains all the characters of traditional ART2 network; (b) Creates better clustering of patterns; (c) preserve learnt patterns in a dynamic structure. The proposed networks provide a few advantages, quick learning and reduced search time. To demonstrate their capabilities, the proposed networks were applied to solve the image classification problems. The proposed network is believed to achieve effective unsupervised task and it has been experimentally found to perform well in images classification.