Online Learning 5 Lower Bound Analysis
why lower bound ?
在之前的UCB / greedy等算法中,我们实际上关注的是 “upper bound”, 也即“算法最坏也不会比$\sqrt{nk}$ 差“(以UCB 为例)。但很自然的会有一个问题是:$\sqrt{nk}$ 足够好了么? 会不会有regret为常数的算法?
lower bound analysis 解决的就是这一问题,我们通过证明问题的lower bound 来展示我们的算法足够优秀
The goal of this chapter is to formally introduce stochastic bandits. The model introduced here provides the foundation for the remaining chapters that treat stochastic bandits.