Abstract
The self-organizing map (SOM), a biologically inspired, learning algorithm from the field of artificial neural networks, is presented as a self-organized critical (SOC) model of the extremal dynamics family. The SOM’s ability to converge to an ordered configuration, independent of the initial state, is known and has been demonstrated, in the one-dimensional case. In this ordered configuration it is now indicated by analysis and shown by simulation that the dynamics of the SOM are critical. By viewing the SOM as a SOC system, alternative interpretations of learning, the organized configuration, and the formation of topograhic maps can be made.

This publication has 17 references indexed in Scilit: