Evolving predictive visual motion detectors

Abstract

The geometrical organization of a visual sensor is of major importance for the later processing of sensed stimuli. We present an approach to evolve artificial visual detectors which adapt their size and orientation according to the experienced sensory stimulation. The criterion for the introduced optimization method is given by a Reichardt correlation measure on the input signal. Under the described conditions, the visual receptors organize their spatial arrangement following the average luminance flow recorded by the sensor over time.