Visualizing changes in data collections using growing self-organizing maps

In previous work we have proposed a modification of the standard learning algorithm for self-organizing maps that iteratively increases the size of the map during the learning process by adding single neurons. The main advantage of this approach is an automatic control of the size and topology of the map, thus avoiding the problem of misclassification because of an imposed size. In this paper we discuss how this algorithm can be used to visualize changes in data collections. We illustrate our approach with some examples.