A Federated Reinforcement Learning Framework via a Committee Mechanism for Resource Management in 5G Networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Telecommunications Engineering

Background:

• Mobile Edge Computing (MEC) environments require efficient resource management for 5G core network automation.
• Decentralized learning approaches are needed to avoid centralized data sharing and enhance privacy.
• Deep Reinforcement Learning (DRL) offers potential for intelligent resource allocation but requires effective integration in decentralized settings.

Purpose of the Study:

• To propose and evaluate a novel Decentralized Federated Reinforcement Learning (DFRL) framework.
• To enhance virtual instance scaling in MEC for 5G core network automation.
• To ensure robust and cost-effective resource management against adversarial attacks.

Main Methods:

• Integration of Deep Reinforcement Learning (DRL) with Decentralized Federated Learning (DFL).
• Implementation of a committee mechanism for monitoring and ensuring reliable gradient aggregation.
• Local decision-making by DRL agents in each MEC with model parameter exchange.

Main Results:

• Demonstrated cost-effective resource usage and significantly reduced blocking rates under diverse traffic conditions.
• Showcased strong resilience against adversarial MEC nodes.
• Validated efficient and adaptive resource management in dynamic network scenarios.

Conclusions:

• The proposed DFRL framework effectively optimizes resource allocation in MEC environments.
• The framework provides a robust solution for 5G network automation, enhancing efficiency and security.
• DFRL is a promising approach for adaptive resource management in complex, distributed systems.