A primary goal of sensory neuroscience is to be able to predict neural response to any arbitrary stimulus. Progress toward this goal has important implications for analysis of large data sets as biological sensory systems are adept at tasks of classification and pattern recognition. To understand the information processing capabilities of the visual system we study cortical architecture and conduct large-scale simulations with a battery of stimuli. To meet the computation challenges of such an approach, we employ parallelization by stimulus which leads to highly efficient and loosely coupled simulations. As an application of visual processing, we show the self-organization of memory of a machine learning system.