Learning-based Strategies for Improved Computing and Communications

Abstract

The anticipated extreme connectivity in future networks paves the way for advanced immersive applications across various domains, including education, industry, and medicine. However, complex immersive systems face significant challenges with the current state of communications and computational technology, such as limited bandwidth capacity, insufficient device computational capabilities, and the need for advanced processing technologies for immersive video transmission. To facilitate the full adoption of immersive applications, this thesis presents a set of models that can be applied at different system levels to enhance performance.

This thesis thoroughly explores immersive systems and introduces the following contributions: (i) a comprehensive overview of the virtual technology applications in the medical domain; (ii) a comparison of learning-based models for intelligent cellular infrastructure management; (iii) a proposed hybrid offloading model for resourceintensive applications; (iv) an analytical framework for delivering video in immersive applications.

From the communication perspective, the study introduces eight Machine Learning (ML) models for network traffic prediction, enabling intelligent infrastructure management in communication overlays. From the computational perspective, the research presents an interoperability framework that bridges computing paradigms and medical applications, along with the Mobile Edge Computing (MEC) – Mobile Cloud Computing (MCC) model for resource-intensive applications. A collaborative offloading strategy reduces system response time by 60% compared to local processing. From the application perspective, this work presents the evaluation of different video delivery approaches in immersive applications and introduces the Field-of-View (FoV)-based delivery model, which minimized the load by 79% compared to sphere-based video delivery.