Multi-resolution distributed ART neural networks

This paper proposes a new neural network model, Multi-Resolution Distributed ART (MRD-ART), which employsfast stable learning and efjicient parallel matching to solve complex data classification problems. The architecture of MRD-ART network preserves the prominent characteristics of the ART networks and extendr their capability to represent input patterns in a hierarchical fashion which effectively controls the category proliferation and signi9cantly improves the memory efficiency and noise tolerance. To achieve this, an MRD-ART network uses multiple ouiput layers arranged in a cascaded manner which is completely different from a conventional ART network with only one output layer. Moreover, the parallel matchingprocess enables the parallel hardware implementation of an MRD-ART. To demonstrate the data representational capability of an MRD-ART network, we applied it to two data sets and the results indicated that fine-to-coarse data representation can be achieved.