Abstract:
We describe an extension of the SOM algorithm, called the Growing Self-Organizing Map (GSOM), which allows to adapt simultaneously the position of the map pointers into the input space, and the topology of the output space. In the GSOM the output space connectivity is constrained to that of a generalized hypercube, with the overall dimensionality , and the extensions along the individual dimensions being a subject of the adaptation. Results for GSOM-generated projections of synthetic and real-world data sets are discussed.