Convolution Neural Network-Based Sensitive Security Parameter Identification and Analysis

Figure 10

Results of identifying power consumption waveform of Arduino. For up to 4 out of the total 15 epochs, the accuracy sharply increases to approximately 0.75, and the loss rapidly decreases to approximately 0.75. After that, it gradually increases during the period and finally reaches 0.98. From epoch 16, the training loss decreases and the validation_loss tends to increase, resulting in an overfitting, and thus, the training was stopped at epoch 20.