Adapting to Misspecification in Contextual Bandits with Offline Regression
                  Oracles

               Sanath Kumar Krishnamurthy 1 Vitor Hadad 2 Susan Athey 2

             Abstract                whose distribution may depend on the context and action.
    Computationally efficient contextual bandits are     The objective of the algorithm is to interactively learn a map-
    often based on estimating a predictive model of      ping from contexts to actions so as to maximize the r ...