Scholarlink

<p>In recent years, with the expansion of Internet of Things (IoT) technologies and the increase in the volume of data generated in distributed environments, edge computing has been proposed as an effective solution for fast and optimal data processing near its source. This approach enables real-time services by reducing latency, improving security, and preserving privacy. However, challenges such as hardware resource limitations, energy consumption, and the need to preserve privacy in serverless edge environments have made it difficult to develop efficient attack detection systems. In this paper, a proposed architecture based on federated self-supervised learning (FedSSL) and lightweight deep learning (LWDL) is presented, which is designed to detect attacks in serverless edge computing (SEC) environments. This architecture first trains local models on unlabeled data using FedSSL to extract meaningful representations. Then, using a LWDL algorithm based on Gradient-based One-Side Sampling (GOSS) and Gradient Boosting Trees (GBT) optimization techniques, the final model is trained and used for attack detection. This architecture significantly reduces energy consumption and the training and execution time of the model, while maintaining data privacy. The valid DoW and UNSW-NB15 datasets have been used to evaluate performance. Experimental results show that the proposed method has significantly improved in terms of accuracy (99.56%), precision (99.78%), recall (99.51%), and F1-Score (99.64%) compared to the FODWNNDoWAD methods and other baseline algorithms. Additionally, model training and execution times were reduced by an average of 4.5 milliseconds, and latency and energy consumption were reduced by 98.08% and 62.90%, respectively, indicating effective structural and algorithmic optimizations in the proposed architecture.</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

AFedSLL-LDL: a framework based-on federated self-supervised learning and lightweight deep learning for attack detection in serverless edge computing

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Mahdi Bazargani,
  • Sina Tarkesh Esfahani,
  • Behnam Heidari,
  • Reza hosseinzadeh shabestary,
  • Mohammadreza Forghani

摘要

In recent years, with the expansion of Internet of Things (IoT) technologies and the increase in the volume of data generated in distributed environments, edge computing has been proposed as an effective solution for fast and optimal data processing near its source. This approach enables real-time services by reducing latency, improving security, and preserving privacy. However, challenges such as hardware resource limitations, energy consumption, and the need to preserve privacy in serverless edge environments have made it difficult to develop efficient attack detection systems. In this paper, a proposed architecture based on federated self-supervised learning (FedSSL) and lightweight deep learning (LWDL) is presented, which is designed to detect attacks in serverless edge computing (SEC) environments. This architecture first trains local models on unlabeled data using FedSSL to extract meaningful representations. Then, using a LWDL algorithm based on Gradient-based One-Side Sampling (GOSS) and Gradient Boosting Trees (GBT) optimization techniques, the final model is trained and used for attack detection. This architecture significantly reduces energy consumption and the training and execution time of the model, while maintaining data privacy. The valid DoW and UNSW-NB15 datasets have been used to evaluate performance. Experimental results show that the proposed method has significantly improved in terms of accuracy (99.56%), precision (99.78%), recall (99.51%), and F1-Score (99.64%) compared to the FODWNNDoWAD methods and other baseline algorithms. Additionally, model training and execution times were reduced by an average of 4.5 milliseconds, and latency and energy consumption were reduced by 98.08% and 62.90%, respectively, indicating effective structural and algorithmic optimizations in the proposed architecture.