Abstract
Complex cells constitute the major fraction of primate V1 neurons. Predicting the responses of these cells in natural viewing conditions is a crucial step toward understanding the mechanisms by which visual information is encoded in the brain. Yet, computational studies of V1 have virtually neglected complex cells in favor of the relatively linear responses of the less frequent simple cells. The few existing models of complex cells are based on data from anesthetized animals, where unknown artifacts of anesthesia and immobilization may alter neuronal responses. In alert animals, it is known that the sequence of inputs produced by eye movements as the monkey inspects objects and scans the scene is a major contributor to the responses of V1 neurons. Furthermore, many complex cells in alert monkeys exhibit strong response modulation to drifting gratings, a behavior that contrasts with the unmodulated responses reported for anesthetized monkeys. An unmodulated response to drifting gratings is an important feature of existing models of complex cells. As a step toward predicting the activity of cortical neurons during natural viewing, we have developed a new model of complex cells that can successfully simulate the effects of several standard stimulus variations on complex cell responses. The model consists of parallel channels, each composed of the cascade of a linear element and a static nonlinearity. Contrary to previous models, no constraint is imposed on the phase of linear subunits. Parameters of this model can be quickly tuned to replicate the activity of individually recorded complex cells by measuring cell responses to pseudo-random stimulation. We have analyzed the performances of the model in predicting the responses of complex cells to a variety of visual stimuli. Test stimuli included drifting and counter-phase gratings as well as dynamic stimuli that reproduced the temporal variation of contrast present when eye movements scan natural scenes.
This material is based upon work supported by the National Science Foundation under Grant No. EIA-0130851 and by the National Institute of Health under grant No. EY012243.