Trust-Augmented Deep Reinforcement Learning for Federated Learning Client Selection.

Area of Science:

• Distributed machine learning
• Artificial intelligence in healthcare
• Internet of Things (IoT) security

Background:

• Federated learning (FL) addresses privacy concerns in distributed machine learning by training models locally on client devices.
• A key challenge in FL is selecting appropriate clients for efficient and effective model training.
• Data scarcity and varying learning capabilities across servers can hinder FL performance.

Purpose of the Study:

• To propose a novel trust-based deep reinforcement learning approach for client selection in federated learning.
• To integrate transfer learning to address data scarcity and enhance model performance in FL.
• To apply and evaluate the proposed FL solution in a healthcare context for COVID-19 detection using IoT devices.

Main Methods:

• Developed a trust-based deep reinforcement learning algorithm for intelligent client selection based on resource consumption and training time.
• Implemented a transfer learning strategy to mitigate data scarcity issues and improve learning across different servers.
• Utilized a federated learning framework for collaborative training of a COVID-19 detection model on edge servers and IoT devices.

Main Results:

• The proposed trust-based FL approach demonstrated effective client selection, optimizing resource usage and training duration.
• Transfer learning successfully compensated for data scarcity, improving overall model robustness.
• The system achieved a favorable balance between COVID-19 detection accuracy and model execution time compared to existing methods.

Conclusions:

• The developed federated learning approach enhances privacy and efficiency in distributed machine learning, particularly in sensitive domains like healthcare.
• The combination of trust-based client selection and transfer learning offers a robust solution for data-scarce environments.
• This method provides a practical and effective framework for privacy-preserving COVID-19 detection using IoT devices.