An efficient personalized federated learning approach in heterogeneous environments: a reinforcement learning...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Distributed Systems

Background:

• Personalized federated learning (PFL) aims to improve model performance on diverse client data.
• Existing PFL methods struggle with data and system heterogeneity, reducing efficiency and performance.
• Heterogeneous environments pose significant challenges for current PFL approaches.

Purpose of the Study:

• To propose FedPRL, a novel personalized federated learning approach for heterogeneous environments.
• To address both data heterogeneity and system heterogeneity in federated learning.
• To enhance the efficiency and model performance of personalized federated learning.

Main Methods:

• FedPRL uses a personalized strategy with local data storage for feature extraction tailored to client data distributions.
• A reinforcement learning-based client selection mechanism optimizes client choice based on data quality and training time.
• Local training incorporates global knowledge distillation of non-target classes to mitigate catastrophic forgetting.

Main Results:

• FedPRL effectively resolves data and system heterogeneity challenges in personalized federated learning.
• The approach significantly enhances the efficiency and model performance in heterogeneous environments.
• Experiments show FedPRL outperforms state-of-the-art methods in accuracy and training efficiency.

Conclusions:

• FedPRL offers a robust solution for personalized federated learning in heterogeneous settings.
• The method improves personalized model performance and global model generalization.
• FedPRL demonstrates superior effectiveness compared to existing methods on standard and real-world datasets.