| （英） |
In recent years, the importance of NIDS (Network Intrusion Detection Systems), which detects unauthorized access, has been increasing, and NIDS using machine learning has been actively studied.
In particular, CNN-NIDS using CNN (Convolutional Neural Network) has been proposed and it is shown to detect anomalies with high accuracy. However, in the field of machine learning, studies of AEs (Adversarial Examples) that intentionally causes misidentification is in progress, and if AEs are applied to CNN-NIDS, they could be a significant security threat. In this presentation, we apply the majority decision method to counter to AEs, where multiple CNN-NIDS are trained on the same training data and the one with the most identification results from each CNN-NIDS is adopted as the decision result. In this presenation, we prepare multiple CNN-NIDSs trained on the KDD CUP 99 data set, and for AEs generated using the one pixel method, we evaluate whether the majority decision method with these CNN-NIDS is validate. |