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

Fig. 5.22. Self-organized OR/DR map. The orientation (top row) and direction (bottom row) maps in the LISSOM OR/DR model were computed separately after self-organization. The orientation preferences are coded using the color bar key on top, and the direction preferences using the color arrow key in the middle. Selectivity is shown in gray scale in both cases, with black indicating low selectivity (as in Figure 5.9). (a) The network represents both orientation and direction in smoothly varying maps that contain all the features found in animal maps, such as linear zones, pairs of pinwheels, saddle points, and fractures (outlined as in Figure 2.4). (b) Most neurons become selective for specific orientation and direction of motion, and are therefore nearly white in the selectivity plots. (c) Overlaying the preference and selectivity plots shows that regions of low selectivity occur near pinwheel centers and along fractures in both maps. (d) The histograms are essentially flat because the training inputs were unbiased. These plots show that LISSOM can develop biologically realistic orientation and direction maps through self-organization based on abstract input patterns.