Similarity Learning and Generalization with Limited Data: A Reservoir Computing Approach

Figure 5

Fraction of image pairs correctly classified versus training set size (a and c) and reservoir size (b and d). Single reservoir results show in (a and b). Dual reservoir results shown in (c and d). Dashed curves denote fourth order polynomial best fit. Reservoir size=1000 nodes for (a and c); training size=250 pairs for (b and d). Spectral radius , sparsity = 0.9, and testing size= 500 pairs.

(a)

(b)

(c)

(d)