Quiroga, Maria del Mar. Recurrent network dynamics modulate orientation tuning in primary visual cortex. Retrieved from https://doi.org/doi:10.7282/T3XG9T0N
DescriptionAdaptation refers to the phenomenon that sensory neurons are affected not only by their immediate input, but also by the sequence of preceding inputs. In visual cortex, for example, neurons change their preferred orientation after exposure to an oriented stimulus. This adaptation is traditionally attributed to biophysical changes in the intrinsic properties of neurons and their connections to other neurons (i.e. ‘plasticity’). In this thesis, however, we propose that some effects of adaptation on neural responses instead reflect natural consequences of response dynamics in recurrent neural networks. In Chapter 2 we used computational modeling to show that recurrently connected neurons can be surprisingly slow to respond to changing environments, even when they have fast intrinsic dynamics. Consequently, adaptation effects on a time-scale of hundreds of milliseconds arose naturally in a recurrent model of sensory processing without plasticity. We showed that these adaptation effects match previously reported changes in orientation tuning in cat and monkey primary visual cortex. In addition, the model predicted a novel short-term change in orientation perception. In Chapter 3 we used quantitative psychophysics to test this prediction in human observers. These behavioral data provided additional support for the model. In Chapter 4 we explored the dynamics of neural responses to rapidly presented sequences of oriented gratings, using electrophysiological techniques to record the activity of neurons in primary visual cortex. We directly compared the recorded neural responses to the responses of the model presented in Chapter 2. We showed that some, but not all, of the changes in the response due to adaptation can be explained by recurrent connectivity within the network. We also compared the neural responses to a linear summation model proposed by Benucci et al. (2009). We found that the linear prediction failed to explain the dynamic responses, suggesting that non-linear, recurrent interactions make important contributions to the representation of orientation. Through the combination of computational, behavioral, and electrophysiological techniques, our work highlights the complexity of neural tuning properties that recurrent networks generate in dynamic sensory environments. Further, our work offers a path forward to measure and understand the complexity of recurrent neural networks in the alert brain.