Quantization Errors in the Harmonic Topographic Mapping

We review two versions of a new topology preserving mapping, the HaToM. This mapping has previously been investigated as a data visualization tool but, in this paper, we investigate empirically the quantization errors in both versions of the mapping. We show that the more model driven version does not minimise the quantization error either when it is calculated in the usual manner or when we use the Harmonic average to do so. Somewhat surprisingly the model driven method lowers the quantization error more quickly than the data-driven method. Key-words: Clustering, Topographic maps, quantization errors.