Efficient Collaborative Learning in the Industrial IoT Using Federated Learning and Adaptive Weighting Based on...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Industrial Internet of Things

Background:

• Federated learning (FL) enables collaborative AI model training while preserving data privacy.
• Data diversity across industries can hinder the efficacy of global model training in FL.
• Industrial Internet of Things (IIoT) integration offers potential for secure, collaborative AI in Industry 4.0.

Purpose of the Study:

• To enhance the robustness and accuracy of global model training in IIoT federated learning.
• To address the challenge of data diversity impacting federated learning model performance.
• To reduce the computational overhead associated with federated learning.

Main Methods:

• Proposed a Shapley value-based adaptive weighting mechanism for global model training.
• Trained the global model as a sequence of cooperative games, adjusting client weights based on Shapley contributions, dataset size, and variability.
• Introduced a quantization strategy to mitigate the computational expense of Shapley value computation.

Main Results:

• Achieved highest accuracy compared to existing methods due to efficient weight assignment.
• Demonstrated comparable accuracy with significantly lower computational cost.
• Reduced the computation overhead of Shapley value computation in each training round.

Conclusions:

• The proposed Shapley value-based adaptive weighting mechanism enhances global model performance in IIoT federated learning.
• The method effectively balances accuracy and computational efficiency.
• This approach offers a promising solution for secure and collaborative AI in Industry 4.0 environments.