USING REAL-TIME NEURAL NETWORKS IN INTRUSION DETECTION SYSTEMS

This paper addresses the pressing limitations of modern intrusion detection systems (NIDS), which are typically based on predefined text-based rules. Such an approach hinders the detection of new or modified attacks, as these static rules can easily be evaded by attackers using minimal modifications. As a promising direction, the study explores the implementation of neural networks equipped with online learning capabilities. Several state-of-the-art solutions are analyzed, including Online Sequential Extreme Learning Machine (OS-ELM), T-DFNN, and various incremental deep neural network models, all of which demonstrate the ability to adapt in real time. The work not only summarizes current methodologies but also emphasizes the significant potential of online learning to enhance the effectiveness and flexibility of cybersecurity systems, particularly in the dynamic detection of emerging network threats.