REINFORCEMENT LEARNING BASED APPROACHES FOR RESOURCE ALLOCATION IN SMART HEALTH SYSTEMS.

Abstract

With the emergence of smart health (s-health) applications and services, several requirements for quality have arisen to foresee and react instantaneously to emergency circumstances. Such conditions demand adaptive fast-acting wireless networks and efficient medical IoT devices. Yet, this requires implementing intelligent network selection and resource management schemes that account for heterogeneous networks characteristics and applications' QoS requirements. Although much literature works to solve these two problems, almost none has considered optimizing both sides intelligently at once (Network's and IoT device's side). Thus, In this thesis, we aim to fill this gap by firstly adopting an intelligent Reinforcement Learning (RL)-based network selection scheme on the Internet of Medical Things (IoMT) device. This will enable the IoMT to be more efficient in adjusting the compression ratio and select the most suitable radio access network (RAN) to transfer the acquired data while considering patient state, battery life, and network dynamics. Secondly, we extend the work by optimizing the network side resources using intelligent network slicing. In which we propose a cost-efficient DRLbased network slicing framework that sustains a high level of network's operational performance by supporting diverse and heterogeneous services, while considering key performance indicators (KPIs), e.g., reliability, energy consumption, and data quality. Specifically, In the second contribution, we aim to find the least-cost route and resources per route for different service flows. Our results from the first contribution show an improvement in the IoT device efficiency demonstrated in longer battery life in addition to a reasonable delay and distortion levels. On the other hand, our approach in the second contribution outperformed the optimal resource allocation algorithm in finding the least cost path and resources per service flow.