Fig. 7: Demonstration of a three-layer ANN model and performance metrics.

a Demonstration of a three-layer ANN model with 784 input, 256 hidden, and 10 output neurons using 28 × 28 pixel MNIST datasets as input signals. b The extracted non-linearity factors from the fitted potentiation curves post-plasma treatment. c The digit recognition training accuracy of our device compared with software values. d Variation of training accuracy with the increasing number of hidden neurons. e Confusion matrix of handwritten digits from 0 to 9 where the diagonal represents high-accuracy classification.