Scholarlink

This paper proposes Cyber-Hawk, an adaptive multimodal threat detection system that detects cyber threats across heterogeneous data sources by means of graph-based representation learning coupled with dynamic optimization. For spatial-temporal features, the architecture uses a hybrid Graph Convolutional Transformer (HGCT), and a Refined Multi-Contrastive Learner (RMCL) with Transformer-based fusion to align semantic embeddings from modalities including network traffic, dark web content, and user behavior logs. Based on Harris Hawks Optimization and augmented with reinforcement learning and cyber-memory for dynamic parameter tuning, a new Cyber-Hawk Optimization (CHO) mechanism is presented to guarantee real-time deployment and adaptability. Experiments on integrated multimodal inputs show strong cross-fold resilience, high precision and 99.6% classification accuracy. Cyber-Hawk supports low-latency scenarios’ lightweight classifiers including XGBoost and Logistic Regression. Cyber-Hawk is a scalable, interpretable, real-time solution for intrusion detection outside conventional unimodal or rule-based systems since the results are validated by convergence analysis and macro-averaged performance metrics.

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

Adaptive Multimodal Threat Detection Using Hybrid Graph Convolutional Transformers and Dynamic Harris Hawks Optimization

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Arindam Bhar,
  • Arindam Sarkar

摘要

This paper proposes Cyber-Hawk, an adaptive multimodal threat detection system that detects cyber threats across heterogeneous data sources by means of graph-based representation learning coupled with dynamic optimization. For spatial-temporal features, the architecture uses a hybrid Graph Convolutional Transformer (HGCT), and a Refined Multi-Contrastive Learner (RMCL) with Transformer-based fusion to align semantic embeddings from modalities including network traffic, dark web content, and user behavior logs. Based on Harris Hawks Optimization and augmented with reinforcement learning and cyber-memory for dynamic parameter tuning, a new Cyber-Hawk Optimization (CHO) mechanism is presented to guarantee real-time deployment and adaptability. Experiments on integrated multimodal inputs show strong cross-fold resilience, high precision and 99.6% classification accuracy. Cyber-Hawk supports low-latency scenarios’ lightweight classifiers including XGBoost and Logistic Regression. Cyber-Hawk is a scalable, interpretable, real-time solution for intrusion detection outside conventional unimodal or rule-based systems since the results are validated by convergence analysis and macro-averaged performance metrics.