Targets in Reinforcement Learning to solve Stackelberg Security Games
This is an incremental review aimed at researchers in security and reinforcement learning, with no new empirical results.
The paper reviews how reinforcement learning models Stackelberg security games, focusing on improving target representations to enhance algorithm performance in security scenarios.
Reinforcement Learning (RL) algorithms have been successfully applied to real world situations like illegal smuggling, poaching, deforestation, climate change, airport security, etc. These scenarios can be framed as Stackelberg security games (SSGs) where defenders and attackers compete to control target resources. The algorithm's competency is assessed by which agent is controlling the targets. This review investigates modeling of SSGs in RL with a focus on possible improvements of target representations in RL algorithms.