Commissioning Federated Reinforcement Learning to Envision Network Security Strategies

Federated Learning (FL) has been used in various cyber security tasks to train ML models in a decentralized and privacy-preserving manner. This has been coupled with the application of reinforcement learning in various simulated environments for effective training in defense and intrusion response scenarios for computer networks. In this work, we presented a FL - based method in which reinforcement learning agents are trained in a customized environment inspired by real-world schemes, to defend computer networks in a privacy-preserving manner. The results show that federated reinforcement learning based agents outperform the reinforcement learning agent.