Publicação
Cyber Red team bot with RL
| dc.contributor.author | Neto, João Pires Ferreira | |
| dc.contributor.institution | Faculty of Sciences | |
| dc.contributor.institution | Department of Informatics | |
| dc.contributor.supervisor | Sá, Alan Oliveira de | |
| dc.date.accessioned | 2026-01-15T16:15:04Z | |
| dc.date.available | 2026-01-15T16:15:04Z | |
| dc.date.issued | 2025 | |
| dc.description | Tese de Mestrado, Engenharia Informática, 2025, Universidade de Lisboa, Faculdade de Ciências | |
| dc.description.abstract | The ever-evolving field of cybersecurity continuously faces numerous challenges, demanding the development of innovative and sophisticated techniques to defend against intrusions, breaches, and data leaks. An increasingly adopted way to identify existing vulnerabilities in a complex networked computing system is through penetration testing. This approach requires a highly skilled group of people called penetration testers to come up with innovative and effective techniques, serving as the barrier of defense between attackers and companies or private individuals valuable information, which makes the time-consuming penetration testing process expensive and prone to human error. To address these issues, this study explores the application of Deep Reinforcement Learning (DRL) algorithms in automating penetration testing procedures and helping penetration testers. By investigating existing works and introducing a novel approach, the research proposes a solution based on the Double Deep Q-Networks (DDQN) algorithm and Graph Neural Networks (GNN) layers to enhance the automation of the penetration testing process. The proposed approach employs a synthetic data-driven environment using property graphs, serving as the foundation for adaptive automated decision-making. Key challenges such as modeling state spaces, training environments, defining actions, and handling uncertainties inherent in penetration testing are critically addressed. Experimental results show that the agent, when trained and evaluated under the same environmental conditions as a human tester, it can outperform this expert-level human counterpart, showcasing it’s potential in advancing current methodologies in penetration testing. The system’s outcomes demonstrate the potential for automating and optimizing penetration testing, ultimately contributing to more effective cybersecurity practices. | en |
| dc.format | application/pdf | |
| dc.identifier.tid | 204173710 | |
| dc.identifier.uri | http://hdl.handle.net/10400.5/116631 | |
| dc.language.iso | eng | |
| dc.subject | Artificial Intelligence | |
| dc.subject | Deep Reinforcement Learning | |
| dc.subject | Property Graphs | |
| dc.subject | Penetration Testing | |
| dc.subject | Cybersecurity | |
| dc.title | Cyber Red team bot with RL | en |
| dc.type | master thesis | |
| dspace.entity.type | Publication | |
| rcaap.rights | openAccess |
Ficheiros
Principais
1 - 1 de 1