Statistical properties of lattices affect topographic error in self-organizing maps

Authors:
Antonio Neme;Pedro Miramontes
Affiliations:
Universidad Autónoma la Ciudad de México, México D.F;Universidad Nacional Autónoma de México, Cd Universitaria, México D.F
Venue:
ICANN'05 Proceedings of the 15th international conference on Artificial Neural Networks: biological Inspirations - Volume Part I
Year:
2005

Citing 6
Cited 0

C4.5: programs for machine learning

C4.5: programs for machine learning
Self-organized formation of various invariant-feature filters in the adaptive-subspace SOM

Neural Computation
GTM: the generative topographic mapping

Neural Computation
Neural maps and topographic vector quantization

Neural Networks
Neural Networks: A Comprehensive Foundation

Neural Networks: A Comprehensive Foundation
Topology preservation in self-organizing feature maps: exact definition and measurement

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

Self-organizing maps (SOM) have been obtained mainly on regular lattices, embedded in euclidean or non-euclidean spaces [1]. We present preliminar results that show SOM can be formed on non-regular lattices by giving evidence that topographic error (TE) is influenced by a few statistical parameters of the neuron lattice, such as the characteristic path length (L), the cluster coefficient (C) and the characteristic connectivity length (Lg). TE is lower not in regular lattices, but in lattices that present a particular set of statistical parameters. In an attempt to identify that set of statistical parameters, we applied mutual information function between the parameters and the TE as well as C4.5 algorithm to obtain rules that identify lattices in which SOMs show low TE.