Adaptive federated learning for resource-constrained IoT devices through edge intelligence and multi-edge clustering.

Area of Science:

• Computer Science
• Artificial Intelligence
• Distributed Systems

Background:

• Federated learning (FL) is vital for Internet of Things (IoT) edge computing, offering scalability and low energy use.
• Heterogeneous edge environments present challenges in efficient edge node selection and resource allocation.
• Dynamic and resource-constrained settings exacerbate these difficulties, impacting overall system performance.

Purpose of the Study:

• To propose a novel architecture, MEC-AI HetFL, for efficient federated learning in heterogeneous edge computing.
• To address the challenges of dynamic node selection and resource optimization in IoT environments.
• To enhance the collaboration and learning efficiency of edge AI nodes.

Main Methods:

• Developed the Multi-Edge Clustered and Edge AI Heterogeneous Federated Learning (MEC-AI HetFL) architecture.
• Implemented multi-edge clustering and AI-driven node communication for dynamic node selection.
• Optimized global learning tasks with a focus on low complexity and efficient resource allocation.

Main Results:

• MEC-AI HetFL demonstrates superior performance in resource allocation, quality score, and learning accuracy compared to existing methods (EdgeFed, FedSA, FedMP, H-DDPG).
• Achieved up to 5 times better performance in heterogeneous and distributed environments.
• Validated through simulations and network traffic tests, confirming effectiveness in IoT edge deployments.

Conclusions:

• MEC-AI HetFL effectively addresses key challenges in IoT edge computing, particularly in dynamic and resource-constrained scenarios.
• The proposed architecture offers a significant advancement in federated learning for heterogeneous edge environments.
• The AI-driven approach enhances collaborative learning and resource management for improved IoT system performance.