Federated learning for 6G-enabled secure communication systems: a comprehensive survey.

Area of Science:

• Artificial Intelligence
• Distributed Systems
• Cybersecurity

Background:

• Conventional centralized training of machine learning (ML) and deep learning (DL) models faces limitations in distributed applications due to storage and performance bottlenecks.
• Federated learning (FL) offers a decentralized approach, enabling collaborative model training on edge devices while preserving data privacy.

Purpose of the Study:

• To analyze vulnerabilities and threats in federated learning environments across diverse application areas.
• To review existing defensive algorithms and strategies against security and privacy threats in FL.
• To provide a structured assessment for successful real-world FL deployment.

Main Methods:

• Literature survey of federated learning vulnerabilities and threats.
• Categorization of FL applications into space, air, ground, and underwater communications.
• Review and comparison of recent defensive algorithms and strategies.

Main Results:

• Identified challenges in FL including data heterogeneity, client mobility, scalability, and data aggregation.
• Cataloged security threats across different FL application domains.
• Compared existing defense mechanisms based on approach, model, datasets, and evaluation metrics.

Conclusions:

• Federated learning presents a promising paradigm for privacy-preserving distributed ML but requires robust security measures.
• Addressing FL vulnerabilities is crucial for its successful adoption in real-world scenarios, especially in communication-centric applications.
• Further research is needed to overcome existing drawbacks and explore future directions in FL security and defense.