Analysis of Continual Learning Models for Intrusion Detection System
1/1/2022
IEEE Access
Institute of Electrical and Electronics Engineers
Abstract

2017 Deep Learning based Intrusion Detection Systems (IDSs) have received significant attention from the research community for their capability to handle modern-day security systems in large-scale networks. Despite considerable improvement performance over machine learning-based techniques and conventional statistical models, deep neural networks (DNN) suffer catastrophic forgetting: model forgets previously learned information when trained on newer data points. This vulnerability is specifically exaggerated large scale due frequent changes network architecture behaviours, which leads distribution introduction of zero-day attacks; this phenomenon termed as covariate shift. Due these constant distribution, DNN models will not be able consistently perform at high accuracy low false positive rate (FPR) rates without regular updates. However, before we update it essential understand magnitude nature drift distribution. In paper, analyze propose an eight-stage statistics learning guided implementation framework that objectively studies quantifies changes. Further, most IDS solutions collect packets store them retrain periodically, but network's size complexity increase, those tasks become expensive. To efficiently solve problem, explore potential continual incrementally learn new patterns while also retaining previous knowledge. We experimental analytical study advanced intrusion detection using three major approaches: forgetting, experience replay, dark replay NSL-KDD CICIDS dataset. Through extensive experimentation, show our achieve improved lower FPR compared state-of-the-art works being patterns. Finally, highlight drawbacks traditional non-gradient approaches handling shift problem.

Keywords
Network Security and Intrusion DetectionAnomaly Detection Techniques and ApplicationsDomain Adaptation and Few-Shot LearningComputer Networks and CommunicationsArtificial IntelligenceArtificial Intelligence
Co-authors