This talk has moved to Spring 2022 — Lower Bounds on Regret for Noisy Gaussian Process Bandit Optimization.