Towards Robust Learning Using Diametrical Risk Minimization For Network Intrusion Detection
Abstract
Currently, deep neural networks (DNNs) show great promise in the detection of malicious network traffic at machine speed. However, these networks are typically trained using Empirical Risk Minimization (ERM), which is not robust to misclassified or altered training data. We propose applying Diametrical Risk Minimization (DRM), which is shown to lead to more robust optimization solutions, to train DNNs to classify malicious network traffic. Using two different network traffic datasets, we find that when state-of-the-art DNNs are trained on partially mislabeled data, utilizing DRM results in higher accuracy compared to equivalent models trained with ERM in 13 of 20 cases examined, with ERM being more accurate in only 5 of the 20 cases. More importantly, when models are tested against previously unseen cyber-attack types, models trained with DRM correctly identify the previously unseen cyber-attacks more often. Of the 46 cases we examine, models trained with DRM show better performance compared to models trained with ERM in 25 cases and equal performance in an additional 10 cases. We show that these DNNs are computationally tractable to deploy in real-time on edge computing systems utilizing commercial-off-the-shelf hardware.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2023
- Accession Number
- AD1213565
Entities
People
- Kelson J. Mccollum
Organizations
- Naval Postgraduate School