Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploring the Effect of DNN Depth on Adversarial Attacks in Network Intrusion Detection Systems

Mohamed ElShehaby, Ashraf Matrawy

TL;DR

This study examines how increasing DNN depth affects adversarial robustness in network intrusion detection, contrasting NIDS with computer vision. Using five fully-connected architectures trained on CSE-CIC-IDS2018 and MNIST, and evaluating under FGSM with varying $\varepsilon$, the authors reveal a domain-specific depth-robustness trade-off: deeper NIDS models become more vulnerable to adversarial perturbations, while depth has a modest effect in MNIST. The work contributes empirical evidence and analysis of why depth interacts differently with structured, low-dimensional NIDS data versus high-dimensional image data, underscoring the need for domain-aware defense strategies. Practically, the findings suggest prioritizing shallow architectures and feature-space defenses for NIDS to balance accuracy and adversarial resilience in security-critical settings.

Abstract

Adversarial attacks pose significant challenges to Machine Learning (ML) systems and especially Deep Neural Networks (DNNs) by subtly manipulating inputs to induce incorrect predictions. This paper investigates whether increasing the layer depth of deep neural networks affects their robustness against adversarial attacks in the Network Intrusion Detection System (NIDS) domain. We compare the adversarial robustness of various deep neural networks across both \ac{NIDS} and computer vision domains (the latter being widely used in adversarial attack experiments). Our experimental results reveal that in the NIDS domain, adding more layers does not necessarily improve their performance, yet it may actually significantly degrade their robustness against adversarial attacks. Conversely, in the computer vision domain, adding more layers exhibits a more modest impact on robustness. These findings can guide the development of robust neural networks for (NIDS) applications and highlight the unique characteristics of network security domains within the (ML) landscape.

Exploring the Effect of DNN Depth on Adversarial Attacks in Network Intrusion Detection Systems

TL;DR

This study examines how increasing DNN depth affects adversarial robustness in network intrusion detection, contrasting NIDS with computer vision. Using five fully-connected architectures trained on CSE-CIC-IDS2018 and MNIST, and evaluating under FGSM with varying , the authors reveal a domain-specific depth-robustness trade-off: deeper NIDS models become more vulnerable to adversarial perturbations, while depth has a modest effect in MNIST. The work contributes empirical evidence and analysis of why depth interacts differently with structured, low-dimensional NIDS data versus high-dimensional image data, underscoring the need for domain-aware defense strategies. Practically, the findings suggest prioritizing shallow architectures and feature-space defenses for NIDS to balance accuracy and adversarial resilience in security-critical settings.

Abstract

Adversarial attacks pose significant challenges to Machine Learning (ML) systems and especially Deep Neural Networks (DNNs) by subtly manipulating inputs to induce incorrect predictions. This paper investigates whether increasing the layer depth of deep neural networks affects their robustness against adversarial attacks in the Network Intrusion Detection System (NIDS) domain. We compare the adversarial robustness of various deep neural networks across both \ac{NIDS} and computer vision domains (the latter being widely used in adversarial attack experiments). Our experimental results reveal that in the NIDS domain, adding more layers does not necessarily improve their performance, yet it may actually significantly degrade their robustness against adversarial attacks. Conversely, in the computer vision domain, adding more layers exhibits a more modest impact on robustness. These findings can guide the development of robust neural networks for (NIDS) applications and highlight the unique characteristics of network security domains within the (ML) landscape.
Paper Structure (17 sections, 1 equation, 6 figures, 3 tables)

This paper contains 17 sections, 1 equation, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background
  4. ML-based Network Intrusion Detection Systems
  5. Adversarial Attacks
  6. Experimental System Description and Methodology
  7. Datasets and Preprocessing
  8. Experimental Neural Networks Architectures
  9. Experimental Scenario
  10. Attack Space in Our Experiments on NIDS Dataset
  11. Experimental Results and Analysis
  12. Performance in NIDS Domain
  13. Performance in Computer Vision Domain
  14. Results' Analysis
  15. Discussion and Implications
  16. ...and 2 more sections

Figures (6)

  • Figure 1: Network Intrusion Detection System's Deployment shehaby2023adversarial2
  • Figure 2: Comparison of adversarial perturbations in computer vision (top) and NIDS (bottom) domains, illustrating the constrained nature of network traffic features versus continuous pixel modifications. Adding adversarial perturbations to image pixels produces a visually similar picture that preserves semantic content, whereas adding perturbations to network traffic flows in the NIDS domain can alter the network and/or malicious functionality of the flow.
  • Figure 3: Comparison of continuous features of CSE-CIC-IDS2018 before and after standardization. (a) Original features exhibit extreme range variations with numerous outliers. (b) Standardized features are centered at zero with unit variance, reducing the overall scale to [-1000, 1000] while maintaining the relative structure and distributional characteristics of each feature, ensuring equal contribution of all features during model training.
  • Figure 4: Experimental Scenario
  • Figure 5: Performance vs Network Depth under FGSM Attack at Different Epsilon Values on CSE-CIC-IDS2018 Dataset
  • ...and 1 more figures