Data normalization with self-organizing feature maps
Alfred Ultsch, Günter Halmans · 2002
The authors present a method to find a suitable transformation using a self-organizing feature map. The feature map's learning algorithm was suitably modified in order to predict the parameter for a transformation. The authors generated different distributions with different skewness and trained a modified Kohonen self-organized feature map with a description of the data. First results point out that the net is able to recall the training set almost exactly. Furthermore, the model is able to generalize to different transformations and to estimate the transformation parameter for unknown distributions with promising error rates.>