A self-organizing architecture of recursive elements for continuous learning

This paper describes how recursive nodes with rich dynamics can be explored in a self-organizing artificial network for continuous learning tasks. The purpose of inserting the recursive elements is introducing chaos behavior in a modified self-organizing map (SOM). This new structure is called CSOM. It incorporates some of the main features of SOM, but it also improves the capability of cluster input patterns through increasing the winning opportunities of the units. The proposal is to use the Lyapunov exponent value to define the winner unit. In addition, the CSOM is introduced in continuous learning task, which is the capacity of learning a new pattern, without losing the patterns learned. The proposal addressed here is described, analyzed quantitatively and its performance is compared with that of conventional SOM.