The Adaptive-Subspace Self-Organizing Map (ASSOM)
Teuvo Kohonen, Helsinki University of Technology, Neural Networks Research Centre,
Samuel Kaski, Helsinki University of Technology, Neural Networks Research Centre,
Harri Lappalainen, Helsinki University of Technology, Neural Networks Research Centre,
Jarkko Salojärvi, Helsinki University of Technology, Neural Networks Research Centre
Email: Teuvo.Kohonen@hut.fi
Abstract:
A special type of a Self-Organizing Map (SOM), Adaptive-Subspace
SOM, learns to identify input patterns invariantly of transformations
acting on them. It is demonstrated that the ASSOM can form
translation-, rotation- and scale-invariant filters, and that it can
be used as a learning preprocessing stage of, for instance, a
texture classifier. The ASSOM can also function as a
translation-invariant novelty filter.
Paper in PostScript
WSOM'97