Simulations of Ocular Dominance and Lateral Connectivity

Figure 1. Self-organization of the afferent input weights into receptive fields. The first two plots show the afferent weights of a neuron at position (42,39) in a 60 x 60 LISSOM network before and after self-organization. This particular neuron becomes monocular with strong connections to the right eye, and weak connections to the left. A neuron at position (38,23) becomes binocular with approximately equal weights to both eyes (third plot).

Figure 2. Self-organized ocular dominance and lateral connection patterns. The ocular dominance of a neuron is measured as the difference in total afferent synaptic weight from each eye to the neuron. Each neuron is labeled with a color value from black to orange to white that represents continuously changing eye preference from exclusive left through binocular to exclusive right. Small blue dots indicate the lateral input connections to the neuron marked with a big blue dot. (a) The surviving lateral connections of a left monocular neuron predominantly link areas of the same ocular dominance. (b) The lateral connections of a binocular neuron come from both eye regions.

Figure 3. Long range inhibitory connection weights immediately before connection death. The inhibitory connection strengths of the monocular and binocular neurons marked in the previous figure are plotted from a viewpoint above the top right corner of the network. The three predominant patches in the first plot correspond to the surviving connections of the monocular neuron shown in figure 2. The strong connections of the binocular neuron shown in the second plot are more or less concentrated in a valley in its neighborhood, without strong patchy connections at a distance.

Figure 4. How OD stripe wavelength changes with input correlations. The figures above shows the ocular dominance of neurons in a 64x64 network when there are no between-eye correlations (correlation factor of 0.0 on a scale of 0-1), and when the correlation factor is 0.5. The wavelength of ocular dominance is much smaller in the latter case, and there are more OD columns in the same network. When the inputs are perfectly correlated, (correlation factor of 1.0), no ocular dominance columns develop.

Figure 5. Topographic maps for the two eyes plotted together. The blue and the white lines represent the retinal topography for each eye after ocular dominance columns have formed. Each map is drawn by plotting the center of gravity (COG) of the afferent weights of each neuron from one eye, and joining up the COGs of neighboring neurons with a line. It can be seen that the regions of rapid change in topography for one eye (lines spaced well apart) correspond to regions of slow change for the other eye (lines bunched together). When plotted separately, it can also be seen that at the borders of ocular dominance columns, the topography is distorted.

Back to title page

sirosh@hnc.com