Modeling Studies of Spatiotemporal Processing

The data collected in our neurophysiological studies is complex and often open to alternative interpretations. We maintain an active program in neural modeling of distributed systems to provide a better understanding of the neurophysiological data and to narrow the possible interpretations of this data. The current model of the saccadic system employs a dynamic network of 400 interconnected units that represents the topological motor map present in the superior colliculus (SC). This dynamic network is connected by distributed feedforward connections to a simulated saccadic burst generator in the brain stem and by feedback connections of eye velocity and eye displacement signals from the brain stem. These connections, and interconnections within the simulated SC, are optimized so that the model produces accurate saccadic behavior with realistic unit behavior for a variety of saccade conditions. Unit behavior in the model can be compared to the neural recordings that we have made in all of these structures.

The figure shows an emergent property of the optimized model. When a focal lesion is placed in the distributed model, it produces hypermetric saccades for some target directions and amplitudes. This property, observed in the actual saccadic system following the placement of small lesions in the SC, has been claimed as proof that lower structures use a vector averaging step to decode the distributed output of the SC. Our results show that a vector summation model (which is much easier to envision in terms of real neuronal processing) can produce hypermetric movements when dynamic neural networks are used to represent structures like the SC with known recurrent connections.