Cybersecurity is a technological focus of individuals, businesses, and governments due to increasing threats, the sophistication of attacks, and the growing number of smart devices. Planning, assessment, and training in cybersecurity operations have also grown to combat these threats, resulting in a boom in cyber defense software and services, workforce development and career opportunities, and research in automated cyber technologies. However, building and maintaining a new workforce and developing innovative cyber-threat solutions are expensive and time-consuming. This thesis introduces a configurable machine-learning environment tailored for training agents that uses different reinforcement learning algorithms within the cybersecurity domain. The environment allows agents to learn simulated cyber-attacks, which act as opposition forces in a realistic, controlled setting that reduces the risk to real computer networks. The thesis also investigates relevant research on machine learning agents for cybersecurity, discusses the simulation architecture, and describes experiments utilizing the Proximal Policy Optimization and Advantage Actor-Critic algorithms. The objective of the thesis is to determine the superior algorithm for automatically identifying exploitable vulnerabilities by evaluating the performance based on accuracy, detected vulnerabilities, and time efficiency.
| Identifer | oai:union.ndltd.org:ucf.edu/oai:stars.library.ucf.edu:etd2020-2919 |
| Date | 01 January 2023 |
| Creators | Müller, Daniel |
| Publisher | STARS |
| Source Sets | University of Central Florida |
| Language | English |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | Electronic Theses and Dissertations, 2020- |
Page generated in 0.0023 seconds