Characterizing the Gap Between Actor-Critic and Policy Gradient

           Junfeng Wen 1 Saurabh Kumar 2 Ramki Gummadi 3 Dale Schuurmans 1 3

            Abstract                 on a range of challenging tasks. Despite the success of AC
                                 methods, AC and PG have subtle differences that are only
   Actor-critic (AC) methods are ubiquitous in re-       partially characterized in the literature (Konda and Tsitsiklis,
   inforcement learning. Although it is understoo ...