An Intelligent Penetration Path Algorithm Based on Deep Reinforcement Learning

An Intelligent Penetration Path Algorithm Based on Deep Reinforcement Learning

연구 분야: Strategies

학회: ICDLT '24: Proceedings of the 2024 8th International Conference on Deep Learning Technologies (ICDLT)

To address the limitations of traditional manual penetration testing methods, this study proposes an automated penetration testing mechanism based on reinforcement learning. Unlike traditional approaches, reinforcement learning algorithms can learn from interactions with the environment, continuously optimizing penetration testing strategies to enhance the efficiency and accuracy of attacker simulations. In response to potential issues such as training instability and slow convergence rates encountered by the traditional DQN algorithm during penetration testing tasks, this work introduces more efficient deep reinforcement learning techniques. We present an innovative algorithm for medium-sized enterprises, termed SADDTAP_PER, which uniquely integrates a priority experience replay mechanism, self-attention mechanisms, dueling optimization layers, and adaptive pooling layers. Experiments demonstrate that our algorithm achieves faster convergence rates on enterprise-level network topologies and maintains good convergence in scenarios with active defense measures.