Is the Retina Optimized for Prediction?

In order to guide future behavior, nervous systems need to make predictions. This problem is so fundamental the we find evidence of predictive phenomena even in the retina. This thesis explores the hypothesis that prediction is a potential "design principle" for understanding the structure and function of the retina. We approach this hypothesis in several ways: First, we characterize statistically the motion of objects in a collection of natural movies, allowing us to quantify predictability in a natural setting. Then we test the predictive capabilities of the retina by recording population responses to artificial stimuli whose statistics are informed by natural object motion statistics; we find that responses are close to optimal when the stimulus statistics are in a naturalistic range. Finally, we examine neural responses to natural movie stimuli to determine if their structure is in line with our theory. A few mathematical derivations related to our theory are given in the appendices.