College of Graduate Studies: Theses & Dissertations
Term of Award
Summer 2026
Degree Name
Master of Science in Computer Science (M.S.)
Document Type and Release Option
Thesis (open access)
Copyright Statement / License for Reuse
This work is licensed under a Creative Commons Attribution 4.0 License.
Department
Department of Computer Science
Committee Chair
Vijayalakshmi Ramasamy
Committee Member 1
Andrew Allen
Committee Member 2
Ryan Florin
Abstract
The convergence of artificial intelligence and cybersecurity presents new opportunities for automated penetration testing capable of discovering, prioritizing, and remediating vulnerabilities at machine speed. However, deployment on resource-constrained ARM platforms remains unexplored despite ARM’s dominance in mobile, IoT, and edge computing with over 280 billion chips deployed globally. This thesis presents systematic experimental evaluation of AI-driven penetration testing across four paradigms—traditional machine learning, deep learning, large language models, and reinforcement learning—on three ARM platform tiers: Raspberry Pi 5 (8GB, Cortex-A76), Radxa ROCK 5B Plus (16GB LPDDR5 with NPU), and NVIDIA Jetson Nano (4GB with Maxwell GPU). The experimental framework generates over 8,000 data points characterizing accuracy, memory consumption, inference latency, power consumption, and ther- mal behavior under realistic deployment constraints. Results demonstrate paradigm-specific de- ployment viability. Traditional machine learning achieves 98.1% accuracy within 162MB peak memory, validating immediate deployment. Deep learning requires model compression through combined pruning, quantization, and knowledge distillation, achieving 92.6% accuracy with 93% size reduction and only a 2.2 percentage-point reduction from baseline (2.3% relative degrada- tion). Large language models face fundamental barriers exceeding platform capabilities by 10– 100x, with INT4 quantization insufficient for practical edge deployment. Reinforcement learning exhibits training-inference asymmetry, enabling edge deployment through off-device training combined with lightweight on-device inference requiring only 96–254MB memory. This research introduces the ARM AI Security Benchmark (AASB) framework providing standardized evaluation methodology, proposes platform-specific optimization guidelines achieving up to 93% memory reduction while maintaining acceptable accuracy, and establishes empirical baselines enabling evidence-based deployment decisions. The findings address the gap between laboratory AI performance and practical ARM deployment, providing a systematic roadmap for securing the billions of ARM devices comprising modern computing infrastructure.
Recommended Citation
APA
Research Data and Supplementary Material
No
Included in
Artificial Intelligence and Robotics Commons, Cybersecurity Commons, Information Security Commons, OS and Networks Commons, Systems Architecture Commons, Theory and Algorithms Commons
