batched-learning

This paper studies the joint effect of memory width and batch depth in stochastic Lipschitz bandits, characterizing the minimax pseudo-regret tradeoff up to logarithmic factors and showing that state width and update depth are not interchangeable.