A Recurrent Self-Organizing Map for Temporal Sequence Processing

This paper presents a recurrent self-organizing map (RSOM) for temporal sequence processing. The RSOM uses the history of a pattern (i.e., the previous elements in the sequence) to compute the best matching unit and to adapt the weights of the map. The RSOM is similar to Kohonen's original SOM except that each unit has an associated recursive differential equation. The experimental results show that the RSOM is able to learn and distinguish temporal sequences, and that it can improve EEG-based epileptic activity detection.