Understanding the role of top-down input in visual information processing

Understanding the interaction between sensory visual input and top-down information in visual processing is crucial for comprehending how the brain effortlessly creates complex visual perceptions. This thesis explores the influence of top-down input on visual information processing using a computational model of the visual cortex. To investigate this, an extended version of the mouse Primary Visual Cortex (V1) model from the Allen Institute [1] is used, which integrates V1, the Lateromedial Area (LM), and the Lateral Geniculate Nucleus (LGN). Methodologically, the study examines response latencies, receptive field areas, and size-tuning functions, comparing the outcomes of the model with experimental data. The
results demonstrate that the model successfully reproduces several experimental observations, including response latencies and receptive field size differences between visual areas. It also generates some neurons with inverse receptive fields and surround suppression. Additionally, we evaluate the impact of feedback projections on these effects and determine the degree to which they are influenced. However, certain limitations were identified, such as an excessive synchronization of the network and an unrealistic dependency of some V1 neurons’ receptive fields on LM feedback. These results emphasize the significance of top-down input in shaping visual processing and highlight areas for future refinement of the model. This work establishes a foundation for improving the V1-LM computational model of the visual cortex, contributing to an understanding of its current neural dynamics.