Evolutionary training of behavior-based self-organizing map

The paper presents a novel idea of a behavior-based self-organizing map. The self-organizing map (SOM) is extended to cover 'objects' that interact with their environment. They are organized based on their behavior instead of parameterized presentation. The original SOM needs a metric to be defined, while in the new self-organizing map no metric between the parameterized presentations is needed. The neighborhood concept of the SOM algorithm is given a probability interpretation that is suitable for evolutionary computing. The behavior based SOM algorithm is presented, and the new concept is demonstrated on linear time-series models, that are identified and organized based on sample data from a simulated system.