Context-Aware, Sustainable Mobile Cloud Computing for Pervasive Health Monitorin

Recent advances on wearable and mobile technologies have made mobile devices a promising tool to manage patients' own health status and assessment through services like telemedicine. However, the inherent limitations of mobile devices make them less effective and sustainable in computation- or data-intensive tasks such as physiological monitoring and analysis. Cloud computing embraces new opportunities of transforming healthcare delivery into a more reliable and endurable manner. Firstly, we propose a hybrid mobile-cloud computational solution to enable effective personalized health monitoring. To demonstrate the efficacy and efficiency of the proposed approach, we present a case study of the real-time electrocardiograph (ECG) tele-monitoring system based upon the developed mobile-cloud computing platform. The experimental results show that the proposed approach can significantly enhance the conventional mobile-based health monitoring in terms of diagnostic accuracy, execution efficiency and energy efficiency.;However, in real-life scenarios, given the ever-changing clinical priorities, personal demands, and environmental conditions, multiple objectives, such as processing latency, energy consumption, and diagnosis accuracy usually need to be considered and fulfilled when deploying such a mobile-cloud-based telemonitoring platform. Therefore, it is imperative to explore a smart scheduling and management approach capable of dynamically adjusting the offloading strategy on this mobile-cloud infrastructure. We propose a Hidden Markov Model (HMM) based dynamic scheduling approach to allow the system to adapt to the changing requirements. Nonetheless, through further analysis, we find that the energy consumption and configuration time cost of the scheduling algorithm itself is non-trivial. Therefore, we study and deploy a model-free reinforcement learning based scheduling approach---Q-learning---to further improve the effectiveness of dynamic computation offloading and task scheduling, while minimizing the overhead. However, with the concern of complex context situation alternation and long term spanning of user behavior pattern analysis, we perceive that fixed modelling of scheduling method could not provide precise reasoning towardsing the mobile cloud system status. In order to bear such diversified circumstances of mobile cloud healthcare services, a Dynamic Bayesian Network (DBN) based sensory fusion approach has been discussed to enable the self-optimization of scheduling strategy itself. With such an adaptive scheme, the sensory infrastructure inside the mobile cloud system would be commanded to reconfigure in a timely and effective manner, to accommodate the various kinds of context condition changes and healthcare service quality requirements.