Thompson Sampling with Information Relaxation
           Penalties

          Seungki Min                Costis Maglaras
        Columbia Business School          Columbia Business School
                    Ciamac C. Moallemi
                   Columbia Business School

                       Abstract
     We consider a finite-horizon multi-armed bandit (MAB) problem in a Bayesian
     setting, for which we propose an information relaxation sampling framework.
     With this framework, we defin ...