Selecting reliable Kohonen maps for data analysis
Extracting useful information from large high-dimensional samples normally requires some preliminary reduction of the complexity. To the extent that it may map any feature space into a two-dimensional lattice (while preserving some topological properties of the input space), the "Kohonen maps" connectionist model is an alternative to traditional techniques for unsupervised data analysis. The solutions this model provides are however sensitive to the initial random weight settings. A methodology is proposed which selects the maps giving the best representation of the data, which are suitable for retention and use in the subsequent interpretation phase.
Bibliographic Reference: Paper presented: International Conference on Artificial Neural Networks, Brighton (GB), Sept. 4-7, 1992
Availability: Available from (1) as Paper EN 36802 ORA
Record Number: 199210707 / Last updated on: 1994-12-02
Original language: en
Available languages: en