Personalized Federated Learning for Intelligent IoT Applications: A Cloud-Edge based Framework

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Intelligent Internet of Things (IoT) services are rapidly expanding, utilizing vast user data.
• Federated learning (FL) enables model training on IoT devices while preserving data privacy.
• Traditional FL struggles with device, statistical, and model heterogeneity in complex IoT ecosystems.

Purpose of the Study:

• To propose a personalized federated learning (PFL) framework tailored for intelligent IoT applications.
• To address and mitigate the inherent heterogeneity challenges in IoT environments.
• To leverage cloud-edge architecture for efficient and low-latency IoT services.

Main Methods:

• Investigated emerging personalized federated learning techniques.
• Developed a PFL framework within a cloud-edge architecture.
• Conducted a case study on IoT-based human activity recognition.

Main Results:

• Demonstrated the effectiveness of PFL in overcoming heterogeneity issues.
• Showcased improved model performance and suitability for IoT environments.
• Validated the framework's capability to meet fast-processing and low-latency requirements.

Conclusions:

• Personalized federated learning is a viable solution for intelligent IoT applications.
• The proposed cloud-edge PFL framework effectively handles IoT heterogeneity.
• PFL enhances the performance and applicability of federated learning in diverse IoT scenarios.