Scholarlink

<p>Advanced Persistent Threats (APTs) pose substantial risks to Serverless edge computing environments due to their stealthy, adaptive, and distributed nature. Traditional centralized detection schemes often struggle with privacy constraints, communication overhead (CO), and delayed response. To overcome these challenges, this paper introduces FATE-X, a Federated, Transferable, and Explainable deep learning framework specifically designed for APT detection in Serverless edge environments. FATE-X integrates federated learning (FL) and transfer learning to enhance cross-node adaptability and maintain data privacy through local model training and secure aggregation. Furthermore, a SHAP-based explainability module is embedded into the inference pipeline, allowing real-time, interpretable detection decisions at each edge node with minimal computational cost. Experimental evaluations conducted on four CIC-APT2023, APTEdge2024, NSL-KDD, and StreamSpot datasets demonstrate the superior performance of FATE-X compared to state-of-the-art methods such as ReAPT-AttP, GraphBDL, and CONTINUUM. The framework achieves up to 4.8% higher accuracy and F1-score, notably reduces the False Positive Rate (FPR), and lowers CO while ensuring explainability and scalability. For example, FATE-X attains 96.8% accuracy on CIC-APT2023 and 99.93% on APTEdge2024, confirming its robustness across heterogeneous and realistic edge environments. The source codes and experimental setup of FATE-X Serverless are accessible at <a href="https://github.com/behnam1368/FATE-X-Serverless">https://github.com/behnam1368/FATE-X-Serverless</a>.</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

FATE-X: a federated transferable and explainable deep learning framework for APT detection in serverless edge environments

  • 期刊论文

  • 发表日期：

    2025年12月2日

查看全文
  • Kusum Yadav,
  • Abdullah J. Alzahrani

摘要

Advanced Persistent Threats (APTs) pose substantial risks to Serverless edge computing environments due to their stealthy, adaptive, and distributed nature. Traditional centralized detection schemes often struggle with privacy constraints, communication overhead (CO), and delayed response. To overcome these challenges, this paper introduces FATE-X, a Federated, Transferable, and Explainable deep learning framework specifically designed for APT detection in Serverless edge environments. FATE-X integrates federated learning (FL) and transfer learning to enhance cross-node adaptability and maintain data privacy through local model training and secure aggregation. Furthermore, a SHAP-based explainability module is embedded into the inference pipeline, allowing real-time, interpretable detection decisions at each edge node with minimal computational cost. Experimental evaluations conducted on four CIC-APT2023, APTEdge2024, NSL-KDD, and StreamSpot datasets demonstrate the superior performance of FATE-X compared to state-of-the-art methods such as ReAPT-AttP, GraphBDL, and CONTINUUM. The framework achieves up to 4.8% higher accuracy and F1-score, notably reduces the False Positive Rate (FPR), and lowers CO while ensuring explainability and scalability. For example, FATE-X attains 96.8% accuracy on CIC-APT2023 and 99.93% on APTEdge2024, confirming its robustness across heterogeneous and realistic edge environments. The source codes and experimental setup of FATE-X Serverless are accessible at https://github.com/behnam1368/FATE-X-Serverless.