Managing Distributed Machine Learning Lifecycle for Healthcare Data in the Cloud

Abstract

The main objective of this paper is to highlight the research directions and explain the main roles of current Artificial Intelligence (AI)/Machine Learning (ML) frameworks and available cloud infrastructures in building end-to-end ML lifecycle management for healthcare systems and sensitive biomedical data. We identify and explore the versatility of many genuine techniques from distributed computing and current state-of-the-art ML research, such as building cognition-inspired learning pipelines and federated learning (FL) ecosystem. Additionally, we outline the advantages and highlight the main obstacles of our methodology utilizing contemporary distributed secure ML techniques, such as FL, and tools designed for managing data throughout its lifecycle. For a robust system design, we present key architectural decisions essential for optimal healthcare data management, focusing on security, privacy and interoperability. Finally, we discuss ongoing efforts and future research directions to overcome existing challenges and improve the effectiveness of AI/ML applications in the healthcare domain.