Edge Devices Clustering for Federated Visual Classification: A Feature Norm Based Framework.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Vision

Background:

• Federated learning (FL) enables collaborative model training across devices while preserving data privacy.
• Non-IID data distribution across devices in FL leads to significant performance degradation due to weight divergence.
• Edge computing environments often face challenges with heterogeneous and non-IID data.

Purpose of the Study:

• To propose a novel clustered federated learning framework, cFedFN, to mitigate performance degradation in visual classification tasks.
• To reduce weight divergence in federated models trained on non-IID data.
• To enhance the performance of federated learning in edge computing scenarios without compromising data privacy.

Main Methods:

• Introduced a clustered federated learning framework (cFedFN) for visual classification.
• Incorporated feature norm vector computation during local training.
• Developed a device grouping strategy based on data distribution similarities to minimize weight divergence.

Main Results:

• cFedFN demonstrated improved performance on visual classification tasks with non-IID data.
• The framework effectively reduced weight divergence among federated devices.
• Achieved superior results compared to existing state-of-the-art clustered federated learning methods.

Conclusions:

• cFedFN offers a robust solution for federated learning on non-IID data, particularly in visual classification.
• The proposed method enhances model accuracy and stability without requiring raw data sharing.
• This framework shows significant potential for privacy-preserving AI in edge computing.