| Determining how populations of neurons respond to and encode information has long been a major goal for systems neuroscience. Understanding the principles by which these populations work has implications for basic biological understanding, for targeting medical interventions, and also for understanding how biological systems solve certain difficult computational problems. In vision, significant progress has been made in describing the types of cells that are present in the different layers of visual processing, but a fuller understanding of how these cell types participate in encoding natural stimuli has been elusive. In this dissertation, we describe a modeling approach that allows us to study individual retinal output cells and output cell populations much more rapidly, by using the models to simulate retinal responses to arbitrary natural stimuli. Finally, we apply these models, and the biological computations that they represent, to the field of computer vision: we show that using retinal signals as a basis for a state-of-the-art computer vision algorithm can greatly improve its performance in being able to generalize across visual conditions. |
