Designing Military Network Scenarios for Offensive AI Training

Are you interested in cybersecurity, artificial intelligence, and intelligent agents? This thesis linked with Armasuisse CYD Campus explores how agents can be trained and evaluated in realistic, controlled network-security environments.

The goal of this BA thesis is to design a bundle of military network scenarios for NASimEmu [1], which is a framework for training deep reinforcement learning agents [2, 3] in offensive penetration-testing scenarios. NASimEmu combines a fast simulator with an emulated environment, allowing agents trained in simulation to be evaluated in a more realistic setting. The framework supports scenario generation, varying network configurations, and testing agents across multiple scenarios.

Sources:

[1]    https://github.com/jaromiru/NASimEmu and https://github.com/jaromiru/NASimEmu-agents
[2]    H. van Hasselt, A. Guez, and D. Silver, ‘Deep reinforcement learning with double Q-Learning’, in Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence, 2016, pp. 2094–2100.
[3]    J. Palanca, A. Terrasa, V. Julian, and C. Carrascosa, ‘SPADE 3: Supporting the New Generation of Multi-Agent Systems’, IEEE Access, vol. 8, pp. 182537–182549, 2020.

This thesis is suitable for a student with interest in:

• Python programming (we are using 3.10.x)
• Computer networks
• Cybersecurity basics
• Linux-based experimentation

Prior experience with reinforcement learning, machine learning, penetration testing tools is helpful, but not mandatory. The work will be conducted strictly in a controlled academic test environment and will focus on simulation, emulation, evaluation, and responsible cybersecurity research.