Fig. 6

Simulation image classification using MNIST: a A handwritten “6” digit with 28 × 28-pixel input image used for the neural network training, b a schematic of the three-layer neural network with 784 input, 128, 64, 32 hidden, and 10 output neurons, respectively. c Recognition accuracy plotted against 100 training epochs. A classification accuracy of 94.6% is often attained. The inset graph displays the handwritten “6” digit recognition accuracy after using 10 learning epochs