Orbital learning: a novel, actively orchestrated decentralised learning for healthcare

Area of Science:

• Artificial Intelligence
• Machine Learning
• Healthcare Informatics

Background:

• Decentralized learning in healthcare faces challenges like data bias and security concerns.
• Existing methods such as federated learning and swarm learning have limitations in performance and accuracy.
• Issues include non-uniform data accumulation, varied patient demographics, and potential data tampering.

Purpose of the Study:

• To propose a novel collaborative and continual learning framework for decentralized healthcare units.
• To address shortcomings of current decentralized learning methodologies, enhancing performance and accuracy.
• To introduce Orbital Learning, a system designed for robust and secure supervised model performance.

Main Methods:

• Developed Orbital Learning, a system using selective grouping and displacement of network actors for intra-group and inter-group learning.
• Incorporated features from split learning and ensemble learning for improved model security and robustness.
• Utilized a digital twin approach to model information quality and flow within the decentralized network.

Main Results:

• Orbital Learning achieved an average Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.819, outperforming federated learning's 0.714 in ECG arrhythmia classification.
• Demonstrated improved performance and robustness against data tampering and demographic variations.
• A scalability demonstration confirmed the platform's versatility with large language models.

Conclusions:

• Orbital Learning effectively addresses key issues in existing decentralized learning methods.
• The proposed system offers a robust, secure, and scalable solution for collaborative learning in healthcare.
• Orbital Learning shows significant potential for advancing AI applications in decentralized healthcare networks.