Distributed Network Defence and Reinforcement Learning

The increasing number of security incidents against computer networks has made insufficient network management and intrusion detection approaches to maintain and to protect these complex systems. Even distributed intrusion detection seems to be not enough if it is used isolated from other disciplines. My research will focus in how network and security agents can learn to detect and to categorise abnormal states of the network using distributed reinforcement learning. We use reinforcement learning because of its capabilities to solve interactive problems when is difficult or impractical to obtain examples of the desired behaviour. We observed that network management systems have failed to offer a proper view of the state of complex network or in networks under security attacks. Intrusion Detection Systems have evolved to distributed systems and they have adopted artificial intelligence techniques to detect intrusions. However these approaches are not enough to identify complex attacks or attacks in a global scale. This work can also contribute in providing new scenarios to evaluate solutions to overcome important issues in the use of reinforcement learning in multiagent systems.