How fast can neuronal algorithms match patterns?

Abstract

We investigate the convergence speed of the Self Organizing Map (SOM) and Dynamic Link Matching (DLM) on a benchmark problem for the solution of which both algorithms are good candidates. We show that the SOM needs a large number of simple update steps and DLM a small number of complicated ones. A comparison of the actual number of floating point operations hints at an exponential vs. polynomial scaling behavior with increased pattern size. DLM turned out to be much less sensitive to parameter changes than the SOM.