Abstract
The responses of cortical neurons elicited by two perceptually separable stimuli have been shown to follow the average of the responses elicited by the constituent stimuli. Such a scheme poses a challenge in segmenting two stimuli that differ only slightly, because averaging essentially takes away the information regarding the stimulus components. Here we investigate how spatially-overlapping stimuli moving transparently in slightly different directions are encoded in the middle-temporal (MT) cortex of macaque monkeys. Visual stimuli were two overlapping random-dot patches moving simultaneously within a static aperture in two directions separated by 60째. We recorded from 155 MT neurons in two monkeys as they performed a fixation task. Based on the response averaging, the predicted tuning curve of a MT neuron to our bi-directional stimuli typically had a single peak, located when the vector-averaged direction of the stimuli was at the neuron's preferred direction (PD). However, we found that the tuning curves of half of the neurons deviated from the averaging prediction. About 1/3 of the neurons showed tuning curves that were biased toward one of the component directions. The tuning curve averaged across these neurons had a single peak located when one of the component directions was near the PD. For another 1/7 of the neurons, they showed two separate peaks in their tuning curves, which peaked when either one of the component directions was near the PD. Interestingly, these direction tuning curves evolved gradually over time. During the initial response period of ~50 ms, the tuning curve was symmetric with a single peak. Over a period of an additional 50 ms, the response tuning was either biased toward one component direction, or split into two peaks. These nonlinear response properties of MT neurons may manifest a dynamic solution of segmenting slightly different component directions of transparently moving stimuli.
Meeting abstract presented at VSS 2014