Metrics
1 Record Views
Dissertation
Integrating Transformers for Cyber Defense Under Unseen and Evolving Threats with Deep Packet Inspection, Zero-Shot Backdoor Detection, and Continual Learning
University of West Florida Libraries
Doctor of Philosophy (PHD), University of West Florida
2026
Abstract
The deployment of deep learning, more specifically transformer-based architectures, in cyber applications faces critical challenges that undermine its transformative potential. These include models’ limited ability to generalize to unseen malware types, the vulnerability of deep neural networks to stealthy backdoor attacks, and the persistent issue of adapting to novel class information when models are incrementally updated with emerging knowledge. These obstacles are further exacerbated by the heavy reliance on large, annotated datasets and traditional defense mechanisms that lack the ability to address the dynamic nature of modern cyber adversaries. To address these challenges, this work introduces a comprehensive approach that spans multiple dimensions of cybersecurity, neural network model defense, and computer vision. First, we present a novel malware detection framework that employs transformer-based architectures for full deep packet inspection across diverse network environments. By integrating self-supervised and few-shot learning techniques, the framework adapts quickly to known and unseen malware variants with minimal labeled data. Next, the research tackles the vulnerability of deep neural networks to backdoor attacks by leveraging vision-language models combined with prompt tuning for zero-shot, open-world detection of unseen compositional threats. A novel compositional zero-shot learning algorithm is developed to identify previously unseen pairings of backdoor triggers and target objects within compromised datasets. Finally, to contribute to continually learning new classes without losing prior information on old tasks, the work utilizes transformers alongside parameter-efficient strategies that preserve previously acquired knowledge while accommodating emerging threat patterns. Collectively, these contributions lay the groundwork for adaptive, resilient, and scalable transformer-based systems capable of countering both emergent adversarial threats and inherent vulnerabilities in intelligent systems.
Details
- Title
- Integrating Transformers for Cyber Defense Under Unseen and Evolving Threats with Deep Packet Inspection, Zero-Shot Backdoor Detection, and Continual Learning
- Resource Type
- Dissertation
- Contributors
- Guillermo Francia III (Committee Co-Chair)Andrew A Mahyari (Committee Co-Chair)Eman El-Sheikh (Committee Member)Hakki E Sevil (Committee Member)Kristen B Venable (Committee Member)
- Publisher
- University of West Florida Libraries
- Format
- pdf
- Number of pages
- 174
- Copyright
- © Kyle Richard Stein 2026
- Identifiers
- 99381761451006600
- Academic Unit
- Intelligent Systems and Robotics; Hal Marcus College of Science and Engineering
- Language
- English
- Awarding Institution
- University of West Florida; Doctor of Philosophy (PHD)
- Theses and Dissertations
- Doctor of Philosophy (PHD), University of West Florida