Q0(X))withthesecond moment regularizer while O(q

"the learner knows the logging policy": This isafairly standard assumption inprior work [1,5,19,22] and holds for10 exampleinonlineadvertisement.