Search

This thesis presents a linear regression technique applied to non-linear features extracted from a scaled Gaussian mixture model (SGMM) to describe the receptive field (RF) behavior of neurons in the Medial Superior Temporal region of the brain of Rhesus monkeys which viewed visual stimuli on a projector screen. The stimuli consist of simple and complex combinations of planar, circular, and radial motion and neuron responses were recorded in spikes/second. It is our goal to understand the neuronal responses of the primates to these stimuli and to create a model capable of predicting how the RFs in the primates' neurons respond to novel stimuli simulating the effects of self- motion. The SGMM is trained in stages using a competitive algorithm where a speciated genetic algorithm competes against a custom greedy algorithm. The regions of each subject's visual field which produce the most active neuron responses are identified as hotspot regions for that neuron. We strongly believe that the non-linear interaction between RFs can be explained by the relationship between hotspot and non-hotspot region behavior based on the direction of stimulus motion within each region. Our results show that the SGMM is capable of reasonably modeling and predicting firing rate response values for simple visual stimuli for a moderate number of data points, and that the linear model is capable of very accurately modeling and predicting firing rate response values for more complex stimuli for a small number of data points.