An EMD-Based Adaptive Client Selection Algorithm for Federated Learning in Heterogeneous Data Scenarios.

Area of Science:

• Machine Learning
• Distributed Systems
• Data Science

Background:

• Federated learning (FL) enables distributed model training while preserving data privacy.
• FL is applicable to conjoint analysis in IoT systems, like smart plant protection.
• Statistical heterogeneity in non-IID data from IoT devices challenges FL efficiency.

Purpose of the Study:

• To enhance federated learning efficiency in scenarios with statistical heterogeneity.
• To address the decline in FL performance caused by non-IID data.

Main Methods:

• Proposed ACSFed, an adaptive client selection algorithm for FL.
• ACSFed dynamically selects clients based on local statistical heterogeneity and past performance.
• Prioritizes clients with higher heterogeneity or poorer performance for later training rounds.

Main Results:

• ACSFed improves federated learning efficiency in heterogeneous settings.
• The adaptive selection strategy accelerates global model convergence.
• Experiments on benchmark datasets validate the proposed method's effectiveness.

Conclusions:

• ACSFed effectively promotes federated learning convergence and efficiency.
• The method offers a practical solution for applying FL in real-world heterogeneous data environments.
• Improved federated model performance is achieved through intelligent client selection.