Collaborative Multi-Expert Active Learning for Mobile Health Monitoring: Architecture, Algorithms, and Evaluation.

Area of Science:

• Cyber-Physical Systems (CPS)
• Machine Learning
• Healthcare Technology

Background:

• Mobile health monitoring is crucial for future healthcare CPS.
• Healthcare CPS face dynamic environments due to the user's multiple roles.
• Existing machine learning models struggle with accuracy changes in dynamic settings.

Purpose of the Study:

• To develop a novel learning architecture for autonomous adaptation in mobile health CPS.
• To address challenges of dynamic environments and heterogeneous knowledge sources.
• To minimize adaptation cost and data labeling uncertainty.

Main Methods:

• Proposed a collaborative multiple-expert architecture using active learning and transfer learning.
• Developed algorithms to manage heterogeneous experts with varying confidence levels.
• Incorporated expert collaboration to enrich knowledge and reduce labeling costs.

Main Results:

• Achieved over 85% accuracy on the first human activity dataset.
• Achieved over 92% accuracy on the second human activity dataset.
• Demonstrated efficacy by labeling only 15% of unlabeled data.

Conclusions:

• The proposed architecture effectively manages heterogeneous knowledge sources in dynamic healthcare CPS.
• Expert collaboration enhances model adaptation and reduces data labeling costs and uncertainty.
• The framework enables accurate activity recognition with significantly reduced labeled data.