Concept Drift
General” https://arxiv.org/pdf/2004.05785.pdf
MAB Specific:
Detecting Change in Data Stream → This doesn’t use ML, it’s just purely Statistics
the f-dsw TS algorithm combines these two estimations of the reward: the estimation of the entire (discounted) history of the rewards and the most recent n rewards. We propose to aggregate the two views using a function f(.) that selects one of the two values (e.g., max, min) or mix them (e.g., mean).