Research Article

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

Figure 7

SNN perceptron performance on trained (seen) classes (a) and test (unseen) classes (b) of MNIST data as a function of training dataset size and SNN perceptron depth. (c) Classification accuracy (fraction correct) of the single and dual reservoir, base SNN, deep SNN, and convSNN on seen (trained) classes and unseen (test) classes, on (1) identifying rotation transformation in MNIST images; and (2) identifying similar visual scenes from a moving camera. Training size: 500 images; testing size: 500 images.
(a)
(b)
(c)