Fig. 5

Handwritten digit recognition with ANN. a Optical potentiation and electrical depression behaviors. b The nonlinearity of weight update (NL) under the optical potentiation and electrical depression behaviors. c The recognition accuracy increases and the loss decreases with the training epochs. d The ANN network structure consists of input, hidden, and output layers. e The mapping images of the output number “2” after training. f Confusion matrix with the 1st training epoch and the 40th training epoch