Scholarlink

With global IPv6 adoption accelerating to address IP address exhaustion, critical security vulnerabilities have emerged, particularly in the Internet Control Message Protocol version 6 (ICMPv6) which performs essential network discovery and management functions. This paper presents a novel framework for detecting ICMPv6-based attacks in dynamic network environments using a hybrid approach that combines federated autoencoders with reinforcement learning. The proposed system enables distributed learning across multiple network nodes while preserving privacy and adapting to evolving attack patterns. Our federated autoencoder model captures normal traffic patterns without sharing raw network data, while the reinforcement learning component optimizes detection policies based on environmental feedback. Experimental evaluation across diverse network topologies demonstrates that our approach achieves 96.8% detection accuracy with a false positive rate of 2.3%, outperforming conventional centralized approaches by 18.7% in dynamic environments. Furthermore, the system exhibits resilience against adversarial attacks and requires 76% less communication overhead compared to centralized learning methods. These findings establish the efficacy of our hybrid approach for securing IPv6 networks against sophisticated ICMPv6-based attacks while accommodating the constraints of distributed and dynamic network environments.

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

ICMPv6-Based Attacks Detection Using Federated Autoencoders and Reinforcement Learning in Dynamic Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hammad Khan,
  • Himanshi Tripathi,
  • Lisha Singh,
  • Utkarsh Singh,
  • Anu Chaudhary,
  • Akhilesh Verma

摘要

With global IPv6 adoption accelerating to address IP address exhaustion, critical security vulnerabilities have emerged, particularly in the Internet Control Message Protocol version 6 (ICMPv6) which performs essential network discovery and management functions. This paper presents a novel framework for detecting ICMPv6-based attacks in dynamic network environments using a hybrid approach that combines federated autoencoders with reinforcement learning. The proposed system enables distributed learning across multiple network nodes while preserving privacy and adapting to evolving attack patterns. Our federated autoencoder model captures normal traffic patterns without sharing raw network data, while the reinforcement learning component optimizes detection policies based on environmental feedback. Experimental evaluation across diverse network topologies demonstrates that our approach achieves 96.8% detection accuracy with a false positive rate of 2.3%, outperforming conventional centralized approaches by 18.7% in dynamic environments. Furthermore, the system exhibits resilience against adversarial attacks and requires 76% less communication overhead compared to centralized learning methods. These findings establish the efficacy of our hybrid approach for securing IPv6 networks against sophisticated ICMPv6-based attacks while accommodating the constraints of distributed and dynamic network environments.