Deep Learning-Based Automated Attacker Models: A Hierarchical Attacker-Defender Framework under Partial Observability

This work examines how AI- and deep learning-enabled cyberattacks can be carried out and investigates how adaptive defence strategies can be developed to counter them. A conceptual model of an autonomous attacker is proposed and formalized within a hierarchical attacker-defender interaction framework. Based on this formulation, adaptive defence mechanisms and a theoretical evaluation framework are defined to assess the effectiveness of learning-based cyber offence and defence.