A Survey on Intrusion Detection Systems using Deep Reinforcement Learning

The significant increase in cyber-attacks particularly during the COVID-19 pandemic, calls for efficient cyber security systems which help organizations detect and mitigate cyber threats. Several traditional cyber defense systems such as Intrusion Detection Systems (IDS) are in place, but the ever evolving and complex novel attacks demand advanced and selflearning systems that can adapt to detect such threats. Such autonomous systems have been implemented using modern reinforcement learning (RL) techniques. This paper presents a survey on various IDSs implemented using Deep RL methods. We cover numerous architectures and present ways by which different authors address the common issues in IDSs such as the tradeoff between accuracy and the false positive rate (FPR), high computational requirements, etc. We expect that this comprehensive survey provides the foundations for and facilitates future studies on exploring the potential of implementing robust and advanced IDSs using DRL by addressing some of the common limitations like low accuracy, high FPR and high computational requirements.