MS-SOM: Magnitude Sensitive Self-Organizing Maps

  • Enrique Pelayo
  • David Buldain
Conference paper
Part of the Advances in Intelligent Systems and Computing book series (AISC, volume 295)

Abstract

This paper presents a new neural algorithm, MS-SOM, as an extension of SOM, that maintaining the topological representation of stimulus also introduces a second level of organization of neurons. MS-SOM units tend to focus the learning process in data space zones with high values of a user-defined magnitude function. The model is based in two mechanisms: a secondary local competition step taking into account the magnitude of each unit, and the use of a learning factor, evaluated locally, for each unit. Some results in several examples demonstrate the better performance of MS-SOM compared to SOM.

Keywords

Self-Organizing Maps Magnitude Sensitive Competitive learning unsupervised learning classification surface modelling 

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. 1.
    Kohonen, T.: Self-Organizing Maps. Springer, New York (1997)MATHCrossRefGoogle Scholar
  2. 2.
    Martinez, P., et al.: Hyperspectral image classification using a self-organizing map. In: Summaries of the X JPL Airborne Earth Science Workshop (2001)Google Scholar
  3. 3.
    Arias, S., Gómez, H., Prieto, F., Botón, M., Ramos, R.: Satellite image classification by self organized maps on GRID computing infrastructures. In: Proceedings of the Second EELA-2 Conference, pp. 1–11 (2009)Google Scholar
  4. 4.
    Pelayo, E., Buldain, J.D., Orrite, C.: Magnitude Sensitive Competitive Learning. Neurocomputing 112, 4–18 (2013)CrossRefGoogle Scholar
  5. 5.
    Villmann, T., Claussen, J.C.: Magnification Control in Self-Organizing Maps and Neural Gas. Neural Computation 18, 446–469 (2006)MATHMathSciNetCrossRefGoogle Scholar
  6. 6.
    Merenyi, E., Jain, A., Villmann, T.: Explicit magnification control of self-organizing maps for ‘forbidden’ data. IEEE Trans. on Neural Networks 18(3), 786–797 (2007)CrossRefGoogle Scholar
  7. 7.
  8. 8.
    Prechelt, L.: Proben1 - a set of neural network benchmark problems and benchmarking rules. Technical Report 21/94, Fac. of Informatics, Univ. Karlsruhe (1994)Google Scholar
  9. 9.

Copyright information

© Springer International Publishing Switzerland 2014

Authors and Affiliations

  • Enrique Pelayo
    • 1
  • David Buldain
    • 1
  1. 1.Aragon Institute for Engineering ResearchUniversity of ZaragozaZaragozaSpain

Personalised recommendations