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

Fig. 10.12. Initial afferent weights across prenatally trained and naive FSA networks. The RFs of every third neuron horizontally and vertically in each network are plotted. For the prenatally trained network (a), the RFs were visualized by subtracting the OFF weights from the ON (as in Figure 10.11). The RF patterns are roughly similar to faces, like the RFs of the prenatal FSA trained with V1 input (Section 10.2). In contrast, the RFs of the naive network (b) were initially uniformly Gaussian; the ON and OFF weights were identical. These networks form an unequal starting point for postnatal learning of faces, as will be shown in later figures.