Deep ResNIDS: A Multistage AI Framework for Novelty Detection in Network Traffic

Soumyadeep Hore, Ankit Shah, Nathaniel D. Bastian, Jalal GhadermaziOpenReview 2024 · Submission to ICLR 2024 · Desk rejected · OpenReview

Abstract

Ensuring computer and network system security is crucial in today's digital landscape. Network intrusion detection systems (NIDS) monitor network traffic to identify potential threats. However, traditional NIDS struggle to adapt to evolving cyberattack tactics. To address this, we propose an AI-enabled novelty detection framework to handle zero-day, out-of-distribution, and adversarial evasion attacks. Our framework comprises three sequential deep neural network architectures: one for the classifier and two for specific autoencoders, designed to effectively detect both known attack patterns and novel, previously unseen samples. We use innovative transfer learning, unfreezing specific neurons, and layer combinations to enhance resilience. Leveraging the one-shot learning approach in the transfer learning component of the framework, we demonstrate continuous improvement in detection accuracy for both known and novel network traffic patterns. Our experiments on benchmark intrusion detection data sets achieved, on average, 98.5% accuracy in detecting various attacks.
BibTeX
@misc{hore2024deep,
  author = {Soumyadeep Hore and Ankit Shah and Nathaniel D. Bastian and Jalal Ghadermazi},
  title = {{Deep ResNIDS: A Multistage AI Framework for Novelty Detection in Network Traffic}},
  year = {2024},
  url = {https://openreview.net/forum?id=2DJMtdfgfH},
  howpublished = {OpenReview},
  note = {Submitted to ICLR 2024},
}
Figure