Quantum Reinforcement Learning Agent for Adaptive Firewall Rule Optimization Under Dynamic Attacks

Abstract

As attacks grow in sophistication, the requirements for intelligent and adaptive security structures that can make real-time decisions on whether to defend or not, exceed what can be achieved with a rule-based approach. This paper aims to explore a Cyber Security system that uses a Quantum Deep Q-Network (QDQN) framework to improve the level of threat detection and automatic responses to generate in the network in a cyber security dynamic. The proposed system uses a quantum-inspired reinforcement learning (QIRL) approach to the system along with a Web-based security management platform for evaluating the states in network traffic and deciding the optimal action for the firewall. This model is the model that has been implemented, using a 46-dimensional state representation, a four-qubit quantum-inspired feature transformation network and a Deep Q-Network Decision Maker layer that classifies threats and suggests one of five security options: Allow, Block IP, Rate Limit, Log and Monitor, or Isolate. It includes user logon and authentication, live threat analysis, visualization dashboards, reporting and model-driven inference services, enabling operational cybersecurity management. Experimental results show that the adaptive security policy learning approach which is proposed can learn the security policy using reinforcement learning and can distinguish between types of attacks such as Normal, DoS, Probe, R2L and U2R. The performance results from both training and testing datasets have shown the applicability of the QDQN model in automated threat neutralization and security decision making. This developed framework opens the door to the exciting possibility of quantum inspired reinforcement learning methods in the development of intelligent firewall systems and adaptive cyber defence infrastructure for the next generation.