Feature Space Mapping network for classification

Wlodzislaw Duch and Rafal Adamczak
Department of Computer Methods, 
Nicholas Copernicus University, 
Grudziadzka 5, 87-100 Torun, Poland

The Feature Space Mapping (FSM) network is based on multidimensional
separable localized or delocalized functions centered around data clusters.
The network learns by memorizing the noisy training data in the input
(feature) space and classifies new data by searching for the nearby memory
traces. Preliminary results for several classification problems are given.
