Bayesian Neural Network Uncertainty Estimation for Detecting Adversarial Attacks on Network Intrusion Detection Systems
Kaustubh Bahl, Amit Praseed · 2025
Machine learning and deep learning algorithms are extensively being employed for the detection of network intrusions. However, despite the superior performance of these algorithms, they have been shown to be vulnerable to adversarial attacks. In the context of machine learning, adversarial attacks refer to the deliberate manipulation of input data to deceive the trained models, leading to misclassifications and compromised performance. As such, understanding the impact of adversarial attacks on network intrusion detection systems (NIDS) and the development of effective defense mechanisms against these attacks is a crucial task. In this work, the use of Bayesian Neural Network Uncertainty (BNNU) estimation is proposed for the detection of adversarial attacks against NIDS and is shown to be an effective line of defense against adversarial attacks against NIDS.