Robust Policy Gradient against Strong Data Corruption

                Xuezhou Zhang 1 Yiding Chen 1 Jerry Zhu 1 Wen Sun 2

            Abstract                 highly noisy data, such as autonomous driving, quantitative
                                 trading, or medical diagnosis.
   We study the problem of robust reinforcement
   learning under adversarial corruption on both re-      In fact, data corruption can be a larger threat in the RL
   wards and transitions. Our attack model assumes  ...