University of Twente Student Theses


Using adversarial reinforcement learning to evaluate he IMMUNE risk-assesment

Varris, V. (2020) Using adversarial reinforcement learning to evaluate he IMMUNE risk-assesment.

[img] PDF
Abstract:During the recent years, as the world becomes more digitalized, there has been a rise in cybercrime and rise of activity from stealthy threat actors such as advanced persistent threats, which poses a challenge for many companies to effectively secure their networks. The goal of the IMMUNE project is to focus on the security of modern industrial networks with an aim on self-defending resilient networks that use the modern networking paradigm software-defined networking. The internship investigates the feasibility of using reinforcement learning, which has enjoyed many breakthroughs in the past few years, in an adversarial setting. The main idea is to simulate 2 reinforcement learning agents: the attacker and the defender in a same partially observable network environment where they compete against each other with a goal for the network defender to learn the best countermeasures in each stage of the attack and therefore contribute towards the larger goal of the project of enabling autonomous network self-deference. As a result, a novel proof-of-concept reinforcement learning approach is devised, that allows defender to learn a policy of best reconfigurations provided by software-defined networking paradigm to take into account the given situational picture provided by the sensors in the network and information from risk assessment system that detects anomalous behaviour. This approach allows to take into account the known vulnerabilities in the system and devise countermeasures based on the behaviour they cause to the system.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science MSc (60300)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page