Advancing Penetration Testing Simulations

Reinforcement Learning has been shown to excel at sequential decision making tasks. Agents can learn from experience while interacting with an environment. Some of the research here at Cylab aims at developing agents that are capable of performing penetration tests for small networks. While current methods look promising, further improvements need to be made in simulating more advanced environments and scenarios, to augment the capabilities. You’ll be able integrate into the team and provide meaningful additions that will be used in the future by us and potentially other research labs.

This domain of research contains several separate projects that students may choose from. Some of the choices are:

Programming new simulations to train agents with better capabilities (i.e. creating Active Directory simulations before applying the agent to the real-world).
Adapting current simulations to work with natural language, to facilitate the usage of LLMs in combination with Reinforcement Learning.
Making current simulations faster to enhance the speed at which agents can interact with the environment. This involves rewriting environments in C, and interfacing it with Python.

Don’t hesitate to reach out to us for more information!

Goal

The actual goals will depend on the chosen topic, and are to be discussed with the supervisor.

Expected outcome

1 blog post
1 poster
a project report documenting the contribution

Required skills

To start this project you should have some knowledge of:

Computer Science
Basic understanding of Cyber Security

You’ll also be able to learn other skills during your project.

Conditions

Applicant’s country of origin must be a member of EU or NATO

Tools and technologies

To achieve this project, you will use following tools and technologies:

git to manage your source code
GitLab to implement Continuous Implementation (CI)