| dc.description.abstract | The entire Internet of Things (IoT) ecosystem is directing towards a high volume
of diverse applications. From smart healthcare to smart cities, every ubiquitous digital sector provisions automation for an immersive experience. Augmented/Virtual
reality, remote surgery, and autonomous driving expect high data rates and ultra-low
latency. The Network Function Virtualization (NFV) based IoT infrastructure of decoupling software services from proprietary devices has been extremely popular due
to cutting back significant deployment and maintenance expenditure in the telecommunication industry. Another substantially highlighted technological trend for delaysensitive IoT applications has emerged as multi-access edge computing (MEC). MEC
brings NFV to the network edge (in closer proximity to users) for faster computation.
Among the massive pool of IoT services in NFV context, the urgency for efficient edge service orchestration is constantly growing. The emerging challenges are
addressed as collaborative optimization of resource utilities and ensuring Quality-ofService (QoS) with prompt orchestration in dynamic, congested, and resource-hungry
IoT networks. Traditional mathematical programming models are NP-hard, hence inappropriate for time-sensitive IoT environments. In this thesis, we promote the need
to go beyond the realms and leverage artificial intelligence (AI) based decision-makers
for “smart” service management. We offer different methods of integrating supervised and reinforcement learning techniques to support future-generation wireless
network optimization problems. Due to the combinatorial explosion of some service
orchestration problems, supervised learning is more superior to reinforcement learning performance-wise. Unfortunately, open access and standardized datasets for this
research area are still in their infancy. Thus, we utilize the optimal results retrieved by
Integer Linear Programming (ILP) for building labeled datasets to train supervised
models (e.g., artificial neural networks, convolutional neural networks). Furthermore,
we find that ensemble models are better than complex single networks for control
layer intelligent service orchestration. Contrarily, we employ Deep Q-learning (DQL)
for heavily constrained service function chaining optimization. We carefully address
key performance indicators (e.g., optimality gap, service time, relocation and communication costs, resource utilization, scalability intelligence) to evaluate the viability
of prospective orchestration schemes. We envision that AI-enabled network management can be regarded as a pioneering tread to scale down massive IoT resource
fabrication costs, upgrade profit margin for providers, and sustain QoS mutually | en_US |
