Quantum-Resilient Federated Learning for Multi-Layer Cyber Anomaly Detection in UAV Systems

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Quantum Computing

Background:

• Unmanned Aerial Vehicles (UAVs) are vulnerable to cyber attacks, with quantum computing posing a future threat to their communication security.
• Existing cryptographic methods in UAV networks are susceptible to quantum decryption, necessitating quantum-resilient solutions.
• The need for robust, privacy-preserving security frameworks for UAVs is critical.

Purpose of the Study:

• To propose a quantum-resilient federated learning framework for multi-layer cyber anomaly detection in UAV systems.
• To integrate advanced deep learning, differential privacy, and post-quantum cryptography for enhanced UAV security.
• To address the vulnerabilities of UAV communication systems against both current and future cyber threats.

Main Methods:

• A hybrid deep learning architecture combining a Variational Autoencoder (VAE) for anomaly detection and a neural network classifier for attack categorization.
• Federated learning with differential privacy for privacy-preserving model training on sensitive UAV data.
• Byzantine-robust aggregation and CRYSTALS-Dilithium post-quantum digital signatures for security and authentication.

Main Results:

• Achieved 98.67% detection accuracy for cyber anomalies in UAV systems.
• Demonstrated a low computational overhead of 6.8% compared to classical cryptographic methods.
• Maintained high robustness against Byzantine attacks in experiments using real UAV attack data.

Conclusions:

• The proposed framework offers effective zero-day anomaly detection and precise attack classification for UAVs.
• Integrated Byzantine-robust and privacy-preserving federated learning significantly enhances UAV security.
• The practical post-quantum security design is validated and suitable for real-world UAV communication data.