Figure 5

Reshaped 14 × 14 contour images of the synaptic weight of each filter with filter evaluation at a δ-value of 0.5 during 15 training numbers for (a) right-center filter to divide ‘Sandal’ and ‘Sneaker’, (b) right-top filter to divide ‘Sandal’ and ‘Boot’, and (c) right-down filter to divide ‘Sneaker’ and ‘Boot’.