EMaQ: Expected-Max Q-Learning Operator
              for Simple Yet Effective Ofine and Online RL

          Seyed Kamyar Seyed Ghasemipour * 1 2 Dale Schuurmans 3 Shixiang Shane Gu 3

             Abstract                 1. Introduction
   Off-policy reinforcement learning (RL) holds the       Leveraging past interactions in order to improve a decision-
   promise of sample-efcient learning of decision-       making process is the hallmark goal of off-policy reinforce-
   making policies by  ...