An incremental growing neural gas learns topologies

An incremental and growing network model is introduced which is able to learn the topological relations in a given set of input vectors by means of a simple Hebb-like learning rule. We propose a new algorithm for a SOM which can learn new input data (plasticity) without degrading the previously trained network and forgetting the old input data (stability). We report the validation of this model on experiments using a synthetic problem, the IRIS database and the handwriting digit recognition problem over a portion of the NIST database. Finally we show how to use this network for clustering and semi-supervised clustering.