Abstract: This paper studies the online reward poisoning problem, wherein an adversary deliberately manipulates the reward function during training to mislead the learning agent into adopting a ...