Algorithms and models toward novel graph problems and applications
Restricted (Penn State Only)
- Author:
- Hung, Hui Ju
- Graduate Program:
- Computer Science and Engineering (PHD)
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- February 23, 2024
- Committee Members:
- Sy-Miin Chow, Outside Unit & Field Member
Ting He, Major Field Member
Wang-Chien Lee, Co-Chair & Dissertation Advisor
Sencun Zhu, Major Field Member
Zhen Lei, Co-Chair of Committee
Chitaranjan Das, Program Head/Chair - Keywords:
- Network intervention
optimization algorithms
social networks
graph optimization problem
network intervention
social network
reinforcement learning
transfer learning
node representation - Abstract:
- Graph theory models entities and their relationships using nodes connected by edges, serving as an essential tool in various domains. With the rich information that graphs encapsulate, there is a growing interest in exploring the underlying complex structures and phenomena they contain. This exploration is vital for advancing knowledge and developing innovative solutions in these fields. Thus, this dissertation presents an exploration of novel graph optimization problems and their applications in various domains, leveraging both algorithmic and machine learning approaches. We begin by addressing an application of network intervention in improving individuals' health outcomes. More specifically, we propose algorithms that optimize key metrics that are related to individuals' health outcomes. The proposed algorithms could serve as a supplemental tool for specialists and practitioners. Moving beyond traditional algorithmic approaches, we then investigate the usage of machine learning techniques for solving graph optimization problems, highlighting the challenges of graph representation learning. Our proposed model uniquely captures both local and global structural information and accurately represents the information of intermediate solutions. Last, we address the question of knowledge transferability across different graph optimization problems. Our proposed framework demonstrates the ability of models to generalize and rapidly adapt across various graph optimization challenges. This dissertation not only contributes to the field of graph theory and machine learning but also demonstrates the practical implications of these advanced techniques in addressing graph optimization problems.