Complex cells - a question of time

Konrad Köerding

Instistute of Neuroinformatics, Zürich

In primary visual cortex two major cell types can be distinguished, simple and complex cells. Simple cells respond well to oriented edges and bars of appropriate contrast polarity; if a bright bar on dark background efficiently drives the cell a dark bar on bright background usually does not. Complex cells to the contrary are invariant with respect to contrast polarity as well as insensitive to small translation of the stimulating bars or edges. Here we explore a strictly local learning rule that exploits the output of the neurons over time. Extracting slowly varying features from natural videos leads to neuron properties that closely match those of complex cells.