The complexity of the brain is overwhelming. How do we identify the essential aspects of structure and connect them to function? A complete model of the brain would be a numerical quagmire. Instead, we rely on effective interactions between pairs of neurons and endeavor to relate these interactions to observed phenomena. As a model system, we focus on neurons in primary visual cortex that respond selectively to motion in one direction.

The development of direction selectivity requires visual stimulation during a critical period in the animal's life. We consider correlations among the outputs of neurons as an effective interaction that could mediate the development of direction selectivity in a new analysis of the study by Li and van Hooser, et al (2008). This dataset consists of hundreds of simultaneously recorded neurons in several animals monitored during different training paradigms. We show that correlations in a population of developing neurons are connected to the outcome of development. Further, we show that development changes the strength of correlations. Finally, this dataset shows the development of direction and orientation preference at a much finer spatial scale than in previous work. We show evidence that there can be significant shifts in the preferred orientations of single cells, which have not been detected previously.

The second half of the thesis deals with theoretical studies of mechanisms for direction selectivity. First, we study noise sensitivity in a common model for direction selectivity, which is based on feedforward network structure. We show that excitatory inputs alone cannot generate high levels of direction selectivity, even in noiseless conditions with ideal tuning. Indeed, experiments suggest that, in direction-selective, inhibitory inputs appear to have the same structure as excitatory inputs. Taking this into account, the model neuron becomes robustly direction-selective.

Finally, we show that direction-selective neurons can emerge in a random cortical network. The fraction of direction-selective neurons increases with the strength of inhibition. We consider general feature selectivity in the network and calculate the receptive field of a neuron analytically. This provides a framework for studying how recurrent connections with minimal structure shape a receptive field.