Semi-supervised Federated Learning for Digital Twin 6G-enabled IIoT: A Bayesian estimated approach

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Network Engineering

Background:

• Industrial Internet of Things (IIoT) and 6G networks generate vast data, increasing Digital Twin (DT) usage.
• Securing sensitive data within DTs is critical.
• Federated Learning (FL) offers privacy but requires substantial labeled data, a challenge in DT environments.

Purpose of the Study:

• To present a Semi-supervised Federated Learning (SSFL) framework to address the scarcity of labeled data in DTs.
• To enhance the security and efficiency of model training for DT-empowered FL settings.

Main Methods:

• Developed the SSFL-MBE algorithm combining Mix data augmentation and Bayesian Estimation consistency regularization.
• Integrated a Bayesian-estimated pseudo-label loss to leverage probabilistic knowledge.
• Focused on scenarios with segregated labeled and unlabeled data across server and clients.

Main Results:

• Evaluated on CIFAR-10 and MNIST datasets.
• The proposed SSFL-MBE algorithm outperformed mainstream SSFL baseline models.
• Achieved model performance enhancements of 0.5% to 1.5%.

Conclusions:

• Contributes to efficient and secure model training for DT-empowered FL.
• Crucial for deploying Industrial Internet of Things (IIoT) in 6G-enabled environments.
• Advances privacy-preserving machine learning in complex digital twin ecosystems.