Stephen David
University of California, Berkeley
A major goal in visual neuroscience is to describe the nonlinear functional relationship between visual stimulus and neural response. Many computational models have been developed for primary visual cortex (V1) to describe nonlinear response properties such as cross-orientation inhibition and contrast gain control. However, this work has been completed largely using simple synthetic stimuli (e.g., sine wave gratings), and little is known about how these models generalize to natural vision.
To study this problem, we developed a new model that integrates known nonlinear properties into a traditional linear spatiotemporal receptive field framework. This model permits the characterization of basic simple and complex cell tuning as well as a wide range of nonlinear response properties. It can be estimated and validated using arbitrary stimuli, including natural scenes. We compared model fits for neurons in V1 of awake macaques using two stimulus classes: sequences of natural images and sequences of sine wave gratings. We observed a high degree of similarity in excitatory tuning between stimulus conditions but systematic differences in inhibitory tuning. The specific pattern of inhibition observed during natural visual stimulation suggests that its role is to encode natural scenes efficiently.