Cloud-Edge-End Collaborative Federated Learning: Enhancing Model Accuracy and Privacy in Non-IID Environments.

Area of Science:

• Computer Science
• Artificial Intelligence
• Distributed Systems

Background:

• Cloud-edge-end computing architectures face challenges with non-independent and identically distributed (non-IID) data from diverse terminal nodes.
• Balancing data heterogeneity and privacy protection is critical for effective large-scale edge data processing.

Purpose of the Study:

• To propose a novel privacy-preserving federated learning method tailored for cloud-edge-end collaboration.
• To address the challenges posed by non-IID data in edge computing environments.
• To enhance both model accuracy and data privacy in distributed systems.

Main Methods:

• A privacy-preserving federated learning approach utilizing cloud-edge-end collaboration.
• Grouping terminal nodes by data distribution similarity and constructing collaborative edge subnetworks.
• Enhancing Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) using an attention mechanism for synthetic data generation.
• Implementing data resampling and loss function weighting to mitigate bias from imbalanced data.

Main Results:

• The proposed method effectively mitigates the negative impact of non-IID data on global model accuracy.
• Enhanced WGAN-GP successfully generates balanced synthetic data while preserving original data patterns and privacy.
• Strategies for data resampling and loss weighting reduce model bias.
• Experimental results show significant improvements in model accuracy and F1-score compared to existing methods.

Conclusions:

• The developed cloud-edge-end federated learning method offers a robust solution for handling non-IID data while ensuring privacy.
• The approach enhances distributed machine learning performance in heterogeneous edge environments.
• This work contributes to more accurate and secure edge data analysis through collaborative learning.