Scholarlink

As cybersecurity faces increasingly complex threats, traditional defense mechanisms prove inadequate while attack methods continue to evolve. This paper introduces SecureTrust-MM, an innovative security-specific large model framework that enhances performance and reliability of AI in cybersecurity through five dimensions: data processing, model architecture, training strategy, defense mechanisms, and runtime monitoring. Our framework features a trustworthy multimodal security data pipeline addressing data scarcity and privacy concerns, a Security Knowledge-Enhanced Multi-Head Attention mechanism for improved understanding of complex security data, and a game-theoretic adversarial defense strategy enhancing robustness through threat-aware adversarial example generation. Experimental evaluation across multiple security datasets demonstrates that SecureTrust-MM outperforms state-of-the-art models by an average of 5.5% in F1-score while reducing false positives by over 40% and false negatives by 38%. The framework excels particularly in challenging tasks like APT detection and threat attribution, establishing a foundation for next-generation intelligent security systems capable of adapting to evolving threats while maintaining high standards of trustworthiness and performance.

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

Secure Trust-MM: A Multimodal Large Model Framework with Game-Theoretic Adversarial Defense for Trustworthy Cybersecurity

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tianxiang Xu,
  • Jiahao Li,
  • Chang Liu,
  • Yuting Zhao,
  • Jianhe Li,
  • Kangsheng Wang,
  • Zexu Huang

摘要

As cybersecurity faces increasingly complex threats, traditional defense mechanisms prove inadequate while attack methods continue to evolve. This paper introduces SecureTrust-MM, an innovative security-specific large model framework that enhances performance and reliability of AI in cybersecurity through five dimensions: data processing, model architecture, training strategy, defense mechanisms, and runtime monitoring. Our framework features a trustworthy multimodal security data pipeline addressing data scarcity and privacy concerns, a Security Knowledge-Enhanced Multi-Head Attention mechanism for improved understanding of complex security data, and a game-theoretic adversarial defense strategy enhancing robustness through threat-aware adversarial example generation. Experimental evaluation across multiple security datasets demonstrates that SecureTrust-MM outperforms state-of-the-art models by an average of 5.5% in F1-score while reducing false positives by over 40% and false negatives by 38%. The framework excels particularly in challenging tasks like APT detection and threat attribution, establishing a foundation for next-generation intelligent security systems capable of adapting to evolving threats while maintaining high standards of trustworthiness and performance.