ARMS: Antithetic-REINFORCE-Multi-Sample Gradient for Binary Variables

                      Alek Dimitriev 1 Mingyuan Zhou 1

            Abstract                If b is discrete, however, Tφ is not differentiable, and f is not
   Estimating the gradients for binary variables is      always differentiable, e.g., if it corresponds to the reward
   a task that arises frequently in various domains,     function in reinforcement learning. This has inspired work
   such as training discrete latent v ...