This academic paper enhances previous research on exfiltration path discovery using reinforcement learning (RL). In contrast to earlier approaches that focused solely on identifying optimal paths, this work incorporates protocol and payload considerations into the Markov Decision Process to emulate network-based exfiltration events more realistically. The proposed method aims to help emulate complex adversarial considerations such as payload size and protocol, thereby improving the identification of expected adversary behavior under various payload and protocol assumptions.
Publication date: 5 Oct 2023
Project Page: https://arxiv.org/abs/2310.03667v1
Paper: https://arxiv.org/pdf/2310.03667
