Convolution Neural Network-Based Sensitive Security Parameter Identification and Analysis

Figure 4

Result of identifying noise sources of Windows 10. In up to 8 out of 20 epochs in total, the accuracy sharply increases to approximately 0.90, and the loss rapidly decreases to approximately 0.2. After that, it gradually increases during the period and finally reaches 0.97. From epoch 21, the training loss decreases and the validation_loss tends to increase, resulting in an overfitting, and thus, the training is stopped at epoch 20.