Figure 4.

The protocol of input pattern presentation to the model, and the evolution of activity in the cortical sheets after presentation of a retinal wave to an untrained network. Each input pattern is presented 15 times (of which four are shown in the diagram for illustration), transformed (in the case of retinal waves, expanded, and in the case of natural images, translated in a random direction) relative to its initial position, followed by a blank stimulus. Each presentation lasts 20 time steps, during which the activity of the retinal sheet is kept constant. The input pattern presentation has two stages: 5000 iterations of expanding retinal waves, followed by 5000 iterations of natural images with randomly jittered positions. Over the 300 time steps that a given stimulus is presented, the layer 2/3 sheet activation gradually forms a pattern of blobs while the layer 4C sheet pattern has local disorder that persists only within the overall pattern of blobs matching the layer 2/3 sheet activity. These differences in activity patterns lead to different map organizations in layer 2/3 and 4C sheets, with layer 2/3 sheet developing smooth maps for orientation and phase (see Figure 6).The red rectangle outlines the region of the retinal sheet corresponding to the cortical 2/3 sheets.