Scholarlink

<p>The increasing complexity and scale of wireless sensor networks (WSNs), particularly in the context of 6G and intelligent edge computing, call for adaptive, secure, and cyber-resilient resource optimization solutions. Traditional heuristic and rule-based methods often fall short in dynamic environments due to limited generalization, lack of real-time adaptability, and vulnerability to cyber threats. In this work, we propose a novel framework that leverages Large Language Models (LLMs) and Generative AI techniques to address the dual objectives of cybersecurity and efficiency in resource allocation. Specifically, we develop a Retrieval-Augmented Generation (RAG)-based architecture that integrates LLMs with contextual network knowledge to perform intelligent, data-driven resource allocation while ensuring resilience against cyberattacks. To strengthen security, we incorporate a Generative Adversarial Network (GAN)-based detection mechanism that identifies and mitigates adversarial cyber threats, such as data spoofing, injection, or model poisoning, thereby enhancing system robustness. We evaluate three LLM variants—CodeLLaMA, LLaMA-2-Instruct, and base LLaMA-2—under multiple network conditions and adversarial scenarios. Two allocation schemes are examined: an LLM-only strategy and a hybrid LLM + binary power control model. Comparative analysis against baselines such as exhaustive search, random allocation, and standalone control demonstrates that our RAG-GAN framework achieves up to 95.2% of optimal energy efficiency, maintains strong spectral efficiency, and detects cyberattacks with over 93% accuracy. These results highlight the potential of GenAI-driven methods to enable cybersecure, adaptive, and efficient wireless resource management in future-generation sensor networks.</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

Integrating LLM and GenAI for CyberSecure and Adaptive Resource Management in Large-Scale Wireless Sensor Networks

  • 期刊论文

  • 发表日期：

    2025年12月2日

查看全文
  • Divya Gupta,
  • Shalli Rani

摘要

The increasing complexity and scale of wireless sensor networks (WSNs), particularly in the context of 6G and intelligent edge computing, call for adaptive, secure, and cyber-resilient resource optimization solutions. Traditional heuristic and rule-based methods often fall short in dynamic environments due to limited generalization, lack of real-time adaptability, and vulnerability to cyber threats. In this work, we propose a novel framework that leverages Large Language Models (LLMs) and Generative AI techniques to address the dual objectives of cybersecurity and efficiency in resource allocation. Specifically, we develop a Retrieval-Augmented Generation (RAG)-based architecture that integrates LLMs with contextual network knowledge to perform intelligent, data-driven resource allocation while ensuring resilience against cyberattacks. To strengthen security, we incorporate a Generative Adversarial Network (GAN)-based detection mechanism that identifies and mitigates adversarial cyber threats, such as data spoofing, injection, or model poisoning, thereby enhancing system robustness. We evaluate three LLM variants—CodeLLaMA, LLaMA-2-Instruct, and base LLaMA-2—under multiple network conditions and adversarial scenarios. Two allocation schemes are examined: an LLM-only strategy and a hybrid LLM + binary power control model. Comparative analysis against baselines such as exhaustive search, random allocation, and standalone control demonstrates that our RAG-GAN framework achieves up to 95.2% of optimal energy efficiency, maintains strong spectral efficiency, and detects cyberattacks with over 93% accuracy. These results highlight the potential of GenAI-driven methods to enable cybersecure, adaptive, and efficient wireless resource management in future-generation sensor networks.