An adversarial reinforcement learning based system for cyber security

Song Xia, Meikang Qiu, Hao Jiang

Abstract: In this paper, we proposed a reinforcement learning based system for defending the network users from malicious network traffics. By training two reinforcement learning agents: network attack generation agent and network defense agent, and based on the environment of deep neural networks, this system not only aims to outperforme the traditional machine learning algorithm (such as CNNs and LSTM), but also can to detect the adversarial example, which is the one of the biggest challenges for current machine learning based intrusion detection system.