TY  - EJOU
AU  - Yan, Jingfu 
AU  - Zhou, Huachun 
AU  - Fan, Xiaojing 
AU  - Huang, Aoran 

TI  - A Survey on AI-Integrated Detection Technologies for Intelligent Communication Networks
T2  - Computer Modeling in Engineering \& Sciences

PY  - 
VL  - 
IS  - 
SN  - 1526-1506

AB  - Artificial Intelligence (AI) has been widely used to detect complex attacks and abnormal behaviors in intelligent communication networks. However, existing studies are often limited to a single scenario or technical route, making it difficult to systematically explain the roles, boundaries, and collaboration mechanisms of different AI-integrated detection technologies. To address this gap, this paper reviews AI-integrated detection technologies in intelligent communication networks from both scenario and technical perspectives. From the perspective of application scenarios, the paper summarizes the requirements and characteristics of AI detection in cloud networks, edge computing, satellite and space networks, and Internet of Things (IoT) environments. From the perspective of technical implementation, the paper first reviews AI detection methods based on programmable data planes (PDPs). These methods are analyzed in terms of in-network feature extraction, lightweight inference, temporal behavior modeling, and data-plane capability boundaries. The paper then introduces knowledge-driven detection methods and explains how security knowledge modeling enhances interpretability and supports attack analysis. Next, blockchain-enabled AI detection is reviewed, focusing on distributed trust, privacy-preserving collaboration, trusted mechanisms, and auditability of detection results. Finally, the paper discusses recent studies on Large Language Models (LLMs) for security detection, highlighting their potential in contextual modeling, reasoning-based detection, explanation generation, and decision support, as well as their deployment risks. Through a comparative analysis of different technical routes, this paper clarifies that these technologies should be deployed in a layered collaborative architecture rather than simply combined at the same execution layer. Specifically, programmable switches (PSs) are more suitable for real-time sensing and preliminary filtering, knowledge-driven methods for semantic correlation and attack reasoning, blockchain (BC) for trusted evidence management and audit support, and LLMs for high-level explanation and decision support. Finally, key challenges and future directions are discussed.
KW  - AI; traffic detection; programmable switches; knowledge-driven; blockchain; large language models

DO  - 10.32604/cmes.2026.084537