Clustering-based Attack Detection for Adversarial Reinforcement Learning

Table 1 shows the values of the parameters involved in the learning of π. For each domain, we describe the state space S and the action space A, the algorithm we use to learn the policy π, the learning rate α, and the maximum number of steps per episode K. Additionally, we set the number of episodes H to 1000 and the discount factor γ to 0.99 in all the domains.