Research on Joint Game-Theoretic Modeling of Network Attack and Defense Under Incomplete Information.

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Game Theory

Background:

• Cybersecurity threats are escalating, complicated by incomplete information and dynamic network environments.
• Traditional defense methods struggle with evolving attack strategies and network structures, hindering adaptive responses.
• High uncertainty in real-world scenarios challenges effective network attack-defense.

Purpose of the Study:

• To propose a robust multi-agent network defense approach for environments with incomplete information.
• To enhance the efficiency and coordination of defense strategies in dynamic attack-defense scenarios.
• To address limitations of traditional methods in handling complex and evolving cyber threats.

Main Methods:

• Developed JG-Defense (Joint Game-based Defense), integrating Bayesian game theory, graph neural networks, and proximal policy optimization.
• Introduced a Dynamic Communication Graph Neural Network (DCGNN) to model network dynamics and facilitate multi-agent communication.
• Designed a joint game loss function to optimize agent strategies for rationality and long-term benefits within reinforcement learning.

Main Results:

• JG-Defense demonstrated a 15.83% improvement in overall defense performance compared to the Cybermonic model.
• The DCGNN model, using a traditional PPO loss function, improved defense performance by 11.81% over the Cybermonic model.
• The integrated approach achieved superior global strategy coordination in dynamic, incomplete information attack-defense scenarios.

Conclusions:

• The proposed JG-Defense approach effectively enhances network defense decision-making under uncertainty.
• The integration of DCGNN and joint game modeling significantly improves coordination and performance in complex cybersecurity environments.
• JG-Defense offers a robust and adaptive solution for modern, dynamic network attack-defense challenges.