Automated Penetration Testing Based on Adversarial Inverse Reinforcement Learning

Automated Penetration Testing Based on Adversarial Inverse Reinforcement Learning

연구 분야: Analysis

학회: 2024 International Russian Smart Industry Conference (SmartIndustryCon)

This article is a study of modern trends in information security from the point of view of vulnerability analysis; the application of the Adversarial machine learning method is considered Inverse Reinforcement Learning (AIRL) in penetration testing. Integration of semantic rewards into the AIRL framework is proposed for deeper modeling of attacker strategies. The article analyzes the problems of penetration testing and proposes tools for collecting expert data, such as Deep Exploit, and discusses the prospects for using AIRL to train agents in dynamic network infrastructures. A model of an automated system for collecting expert data and penetration testing based on Adversarial Inverse Reinforcement Learning is proposed. The method used provides more effective simulation of vulnerabilities and increases the system's adaptability to modern information attacks and a changing environment. The use of AIRL with semantic rewards can significantly improve information security processes and provide new tools for analyzing the effectiveness of countering complex evolving cyber-attacks.