Federated deep reinforcement learning enabled hierarchical Edge-Fog-Cloud architecture for intelligent task...

Area of Science:

• Computer Science
• Artificial Intelligence
• Distributed Systems

Background:

• Increasing demand for latency-sensitive, computation-intensive, and mobility-aware applications in Edge Fog Cloud environments.
• Limitations of traditional, heuristic, and centralized offloading methods in handling heterogeneous workloads, non-stationary environments, and privacy concerns.

Purpose of the Study:

• To propose a Federated Deep Q-Learning (FDQL)-based task offloading framework for adaptive, decentralized, and privacy-conscious decision-making in hierarchical Edge Fog Cloud architectures.
• To overcome the drawbacks of existing offloading methods by incorporating deep reinforcement learning and federated learning.

Main Methods:

• Task offloading is modeled as a Markov Decision Process.
• FDQL framework trains execution decisions considering latency, bandwidth, queue length, computational load, energy state, and user mobility without raw data sharing.
• Collaborative learning between distributed edge nodes using federated model aggregation.

Main Results:

• FDQL framework demonstrates improved performance in distributed, resource-constrained environments compared to baseline approaches.
• Achieved shorter latency, increased energy efficiency, and more predictable execution-layer selection.
• Ablation studies confirm the significance of federated learning, mobility awareness, and bandwidth-aware optimization.

Conclusions:

• The designed FDQL framework is a successful, interpretable, and scalable approach to intelligent task offloading.
• FDQL decisions are based on SLA-relevant features like latency, bandwidth, and resource use, enhancing transparency and trustworthiness.
• The framework is suitable for future 6G-enabled applications such as smart cities, industrial IoT, and autonomous systems.