|dc.description.abstract||The recent advances in the Internet of Things (IoT), Big Data, and Machine Learning (ML)
have contributed to the rise of a growing number of complex and intelligent applications.
Examples of such applications are real-time disease detection, self-driving vehicles, drone
package delivery, and smart manufacturing. These emerging applications are often realtime, data-intensive, and delay-sensitive, and ensuring Quality of Service (QoS) for these
applications is a challenge. Fog computing is seen as one promising solution for providing
QoS for these applications, as it puts compute, storage, and networking resources closer to
In this dissertation, we showcase how fog computing can improve the QoS in several emerging IoT and ML applications. To do so, we first provide a tutorial on fog computing and its
related computing paradigms, such as edge computing and cloudlets, discussing their similarities and differences. Secondly, we propose two novel schemes, ResiliNet and deepFogGuard,
for improving the QoS of deep learning applications that are deployed over network devices,
such as fog nodes and edge devices. Specifically, the two schemes are for improving the
failure-resiliency of inference in distributed neural network in the presence of physical node
failures. deepFogGuard’s design is based on the concept of skip connections, whereas ResiliNet uses a novel technique, introduced in this dissertation, called failout, which is inspired
by dropout in neural networks. Next, we propose two fog-based frameworks, FogPlan and
FogOffload, that can improve the QoS in terms of the reduced service delay. FogPlan
is a planning framework, for the dynamic deployment or release of application services on
fog nodes, in order to meet the low latency and the QoS requirements of applications while
minimizing the cost. FogOffload is a delay-minimizing collaboration and offloading policy
for fog-capable devices that aims to reduce the service delay of IoT applications.
The above four frameworks are designed to improve the QoS in emerging IoT and ML
applications through Fog Computing. The former two frameworks improve failure-resiliency,
whereas the latter two reduce the service delay.
The outcomes of this research have appeared in the proceedings of a conference and a workshop, three journal papers, and a paper in preparation. A website has been created to serve
as a hub for the details of fog computing conference and journals. This research has also
attracted the attention of our industry partner, Fujitsu, who partially funded this research.
Additionally, two graduate Master’s students and seven undergraduate students were mentored and actively involved in various research projects, which were defined based on the
topics of this dissertation. This dissertation has also been selected to be funded by the
Dissertation Research Award at UT Dallas.||