Summary: Pattern Recognition Lerrers r3 (r992) 153-159
Linear mappings of local d,ata structures
Mayer Aladjem and Its'hak Dinstein
Deportment of Electrical and computer Engineering, Ben-Gurion university of the Negev, Beer-sheva g4105, Israet
Received 23 September l99l
Aladjem, M' and I' Dinstein, Linear mappings of local data structures, Pattern Recognition Letters l3 (1992) 153-159.
Two methods for linear mapping of multidimensional data in the case of unsupervised learning are proposed. The first merhod
maximizes the mean square density gradient of the projected samples with the intention of compressing the clusters. The second
method is based on the k-NN technique and obtains a map of the scatter of the neighbo. clusters. An experiment with the
classical Iris data shows the mapping accuracy of the latter method.
Keywords' Interactive pattern recognition, mapping of multidimensiona! data, cluster analysis, local data structure.
Mapping methods (siedlecki et al. (t9gg)) are
especially important in the analysis of multivariate
data. In general, the mapping consists of finding a