The Continuous Interpolating Self-organizing Map

摘要

The self-organizing map (SOM) [5] provides a general data approximation method which is suitable for several application domains. The topology preservation is an important feature in data-analysis and may also be advantageous for the evaluation of the data in a function approximation or regression task. For this reason the interpolated self-organizing map (I-SOM) adds an output layer to the SOM architecture which computes a real valued output vector. This paper presents an extension of I-SOM towards a continuous interpolation. It is compared to RBF and to the parametrized self-organizing map.