Hybrid quantum enhanced federated learning for cyber attack detection.

Area of Science:

• Cybersecurity
• Network Security
• Artificial Intelligence

Background:

• Conventional cyber-attack detection methods face challenges with data privacy and scalability.
• Centralized systems struggle to adapt quickly to evolving cyber threats.
• Limitations in traditional approaches necessitate innovative, decentralized solutions.

Purpose of the Study:

• To develop a novel federated learning solution for enhanced cyber-attack detection.
• To address limitations of centralized methods concerning data privacy and communication overhead.
• To improve the adaptability and performance of network security systems.

Main Methods:

• Integration of a spatio-temporal attention network (STAN) for pattern recognition.
• Implementation of a quantum-inspired federated averaging (QIFA) optimization procedure.
• Utilizing hierarchical model aggregation and multi-stage refinement with privacy preservation.

Main Results:

• The proposed model achieved high performance metrics: 98.2% precision, 98.5% recall, 98.35% F1-score, 98.2% specificity, and 98.34% accuracy.
• Demonstrated superior performance compared to traditional CNN, LSTM, RNN, and baseline federated learning models.
• Effective detection of various network anomalies using the UNSW-NB15 dataset.

Conclusions:

• The novel federated learning approach significantly enhances cyber-attack detection capabilities.
• The integration of STAN and QIFA offers a robust and privacy-preserving solution.
• The proposed model represents a significant advancement in adaptive and efficient network security.