Scholarlink

<p>This study presents a resource-efficient deep neural network (DNN)-based intrusion detection system (IDS) specifically designed for resource-constrained devices, while also leveraging high-performance computing (HPC) for accelerated training and large-scale evaluation. To achieve this, a design space exploration approach is employed to extract a significantly smaller DNN model than a state-of-the-art reference, ensuring suitability for embedded and low-power systems without compromising training speed or scalability. The proposed model is comprehensively validated on two benchmark datasets, CIC-IDS2017 and CSE-CIC-IDS2018, with HPC-enabled training facilitating rapid experimentation across multiple architectures. Compared to the baseline, it exhibits only minor performance reductions of 1.48%, 0.21%, and 0.76% in precision, recall, and F1-score, respectively, on CIC-IDS2017, while outperforming the baseline by 5.87%, 1.25%, and 3.34% on CSE-CIC-IDS2018. Hardware evaluation demonstrates that the model is 275 times smaller, 37 times faster, and 5.39 times more power efficient than the GPU-based reference. These results confirm that the proposed IDS provides a robust and efficient solution for securing resource-limited technologies. By enabling fast, parallel, and power-efficient inference, the proposed IDS serves as a practical building block for real-time and HPC-driven supercomputing applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LightIDS: a lightweight neural network-based intrusion detection system

  • 期刊论文

  • 发表日期：

    2025年12月23日

查看全文
  • Ebrahim Fard,
  • Mahdi Soltani,
  • Amir Hossein Jahangir,
  • Seokbum Ko

摘要

This study presents a resource-efficient deep neural network (DNN)-based intrusion detection system (IDS) specifically designed for resource-constrained devices, while also leveraging high-performance computing (HPC) for accelerated training and large-scale evaluation. To achieve this, a design space exploration approach is employed to extract a significantly smaller DNN model than a state-of-the-art reference, ensuring suitability for embedded and low-power systems without compromising training speed or scalability. The proposed model is comprehensively validated on two benchmark datasets, CIC-IDS2017 and CSE-CIC-IDS2018, with HPC-enabled training facilitating rapid experimentation across multiple architectures. Compared to the baseline, it exhibits only minor performance reductions of 1.48%, 0.21%, and 0.76% in precision, recall, and F1-score, respectively, on CIC-IDS2017, while outperforming the baseline by 5.87%, 1.25%, and 3.34% on CSE-CIC-IDS2018. Hardware evaluation demonstrates that the model is 275 times smaller, 37 times faster, and 5.39 times more power efficient than the GPU-based reference. These results confirm that the proposed IDS provides a robust and efficient solution for securing resource-limited technologies. By enabling fast, parallel, and power-efficient inference, the proposed IDS serves as a practical building block for real-time and HPC-driven supercomputing applications.