IBN - 9723846 Complex Predictive Pursuit by the Eye Compared to a Cerebellar Model of Pursuit - R. E. Kettner. One of the most remarkable characteristics of all animal life is the fluidity and coordination of movement that reaches high levels in skilled athletes and dancers, but is also present in everyday action. This ability is particularly remarkable given the slowness of individual neurons in the brain. The brain increases its effective processing speed by performing many computations in parallel, but this does not appear to remove long delays in processing visual information. The problem is particularly acute when one attempts to explain how the eye is able to track a target moving rapidly along a complex trajectory with essentially zero lag. If eye motion were controlled solely by changes in the current location of the target, one would expect the eye to lag the target by the 100 ms delay required to process visual input. Rather, it appears that the system is able to compensate for delays by predictive control. That is, it predicts how the eye should move based on highly delayed information. This project will conduct experiments in monkeys to determine the limits of eye movement prediction under a variety of conditions: (1) motion along circular and complex trajectories when target velocity is either constant or variable, (2) motion along a circular trajectory interrupted by a right-angle change in target direction at either predictable times and locations compared with identical target deviations at unpredictable times and locations, and (3) motion along circular and complex trajectories when the target is briefly turned off. The project will also continue the development of a biologically-realistic neural-network model of predictive eye control based on regions of the brain's cerebellum known to be involved in pursuit eye movements. This model uses a much larger number of internal units than other pursuit models (440 input mossy fibers, 6 000 internal granule cells, 2 Purkinje cell outputs) to generate complex predictive pursuit. The model learns new trajectories in a biologically- reasonable fashion by modifying granule-to-Purkinje cell synapses using visual error signals (from climbing fiber inputs). The behavioral data, neural response properties, and anatomical connections are all based on experiment. Data obtained in the above experiments will be used to test, and if necessary, modify the model. In addition, model performance will be compared with results from studies in other laboratories. Random target motions will be tested that can only be performed using visual input. New simulations will also test how well the model performs when the frequency of a learned trajectory is changed, and whether the model can learn more than one trajectory at the same time. All of this work should provide important information about the role of prediction in motor control, as well as increase our understanding of how brain systems accomplish predictive control.
