MLSec-Project

Group Members:

Penetration Testing plays a critical role in evaluating the security of target systems by emulating real active adversaries.
However, the current approach demands significant manual effort, particularly for expansive and intricate networks, leading to outcomes heavily reliant on the expertise of pen-testers, thus diminishing repeatability.

Network Attack Simulator creates a detailed simulation of a real-life network topology and infrastructure
Scenario definition consists of: Network configuration, Host configurations and pen-tester configurations
Supports partially observable mode; reflecting the reality of pen-testing more accurately.
To address the challenge of achieving multiple objectives we model the solution as an Multi-Objective Markov Decision Process i.e. the reward R is a vector with n individual rewards, instead of a scalar reward.
We employ Proximal Policy Optimization (PPO) as it exhibits stable responsiveness to environmental changes, adjusting the gradient update step size optimally and promoting exploration.
These algorithms will train intelligent agents to maximize control over systems within a complex state space that simulates network topology.