Asymptotic expansion for batched bandits

Yechan Park, Nakahiro Yoshida

In bandit algorithms, the randomly time-varying adaptive experimental design makes it difficult to apply traditional limit theorems to off-policy evaluation of the treatment effect…