Click on the image to see a PDF version (for zooming in)

Fig. 5.29. Combined OR/OD/DR map trained with Gaussians. Based on oriented, moving Gaussian patterns with different brightnesses in each eye, LISSOM develops overlaid orientation, ocular dominance, and direction maps simultaneously. This plot shows the orientation preferences in color coding, the boundaries of the OD stripes in black, and the direction preferences and selectivities as white arrows, as in Figures 5.23 and 5.27. The network develops a realistic orientation and direction map, with OR patches subdivided into areas preferring the opposite directions of motion. Ocular dominance boundaries tend to cross linear zones at right angles, rather than following the orientation map. These results are similar to the ones with individual input dimensions, complicated by the fact that multiple dimensions are being mapped at once. Similar results have been observed experimentally with the cat visual cortex (Hubener et al. 1997; Löwel et al. 1988).