Author ORCID Identifier:
Date of Graduation
12-2025
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Computer Science (PhD)
Degree Level
Graduate
Department
Computer Science & Computer Engineering
Advisor/Mentor
Panda, Brajendra
Committee Member
Poncet, Aurelie
Second Committee Member
Zhang, Lu
Third Committee Member
Le, Ngan
Keywords
Access Control; Cybersecurity; Machine Learning
Abstract
Access control is a well-established challenge in cybersecurity, with significant research focused on enhancing system autonomy and accuracy across various scenarios. Access control rules can be designed based on users’ roles, attributes, or relationships requesting access to specific resources. However, despite their benefits, these models still require human oversight. As systems expand and grow, it becomes increasingly complex for administrators to maintain precise access control rules, often necessitating extensive system updates or even a complete overhaul. This dissertation introduces a novel approach that leverages contextual embedding for user information to enable the system to autonomously authorize user requests for resources. The objective is to develop a framework that supports system administrators in decisionmaking and auditing, and temporarily assists system engineers with access control tasks until a more permanent solution is implemented. Furthermore, the framework is designed to be adaptable and applicable across various systems rather than being restricted to a single one. To achieve this, we propose a candidate framework called Token2Vec, which is rigorously tested on real-world data across different scenarios and compared with other state-of-theart (SOTA) methods. The results demonstrate that Token2Vec performs well across various conditions. Additionally, we extend our research to handle tokens the system has never encountered, enhancing our framework to incorporate this new information without reconstructing the entire embedding layer. We break each token into n-grams before generating its corresponding embedding vectors to achieve this. This method enables a more flexible encoding of values by producing multiple n-gram combinations rather than relying on a single predefined token from the system. To validate the robustness of this approach, we conduct experiments comparing our new approach with our initial method in scenarios where tokens are either fully or partially introduced to the system. Our new approach performs competitively with the original method when all tokens are in the training data. Subsequently, the new approach maintains its performance, whereas the original Token2Vec framework experiences a significant decline. We also explore an embedding-based approach to assess the reliability of nodes within a fog computing system. By analyzing the embedded values derived from each node’s access control logs, we aim to detect irregularities caused by rogue or faulty nodes that permit unauthorized access by malicious users. This method enables system administrators to isolate compromised nodes swiftly. To evaluate our approach, we generate scenarios using simulated data based on real-world logs and benchmark its performance. The results demonstrate promising effectiveness in identifying rogue nodes. Finally, we examine the challenges and future research directions associated with contextual embedding for access control, including its applications in more complex system development, considerations of human roles, issues of bias and fairness, and long-term sustainability. Additionally, we emphasize the potential to integrate our approach with advanced machine learning techniques, enhancing our model into a more sophisticated solution for evolving access control frameworks.
Citation
Bui, T. (2025). Contextual Embedding using Machine Learning for Cybersecurity: Access Control and Application. Graduate Theses and Dissertations Retrieved from https://scholarworks.uark.edu/etd/6065
