A temporal trace and SOM-based model of complex cell development

Thomas J. Sullivan, Virginia R. de Sa

Abstract: An activity-dependent model of complex cell development is described. The learning rules are a combination of a temporal trace rule, a Hebbian rule, and a self-organizing map (SOM). They are local, unsupervised, and biologically realistic. The only input that is assumed is spontaneous retinal wave activity. The results show complex cells with phase invariance and orientation tuning that form an organized map.