Parts | Days | Selection | Search | Downloads | Help
SYPH: Physik im Hirn - Physical Approaches to Brain Function
SYPH II: HV II
SYPH II.1: Invited Talk
Wednesday, March 13, 2002, 14:40–15:20, H2
Optimal neural processing in the analysis of visual motion — •Rob de Ruyter van Steveninck — NEC Research Institute, Priceton (NY), USA
Our sense organs receive physical signals from the world around us. These signals are transduced into electrical responses, and those in turn are interpreted by the brain. Such processing can be cast in the form of an inverse problem: The brain must abstract useful features from a generally noisy and incomplete set of input data. With this in mind we study motion estimation by the visual system of the fly. As a fly moves through the world its retina receives a wildly fluctuating spatiotemporal pattern of optical stimuli. We want to know first, what is the optimal way of extracting an estimate of self motion from such a data stream, and second, how closely does the fly approach this optimal computation? To get at these questions we sample the joint distribution of self motion and visual input in natural conditions. From this we can directly define a statistically optimal velocity estimator. By recording from fly motion sensitive neurons we characterize the fly’s motion response and its dependence on contrast and velocity. The fly neurons, as is typical for biological motion detectors, show a characteristic contrast bias in velocity estimation. The optimal estimator shows the same bias, which leads us to reinterpret the physiological findings as an optimal solution to the motion estimation problem in noisy conditions. Some consequences of this interpretation will be explored in the context of natural visual input signals.