PaperReading20200229
CanChen
ggchen@mail.ustc.edu.cn
This paper about CB contains too much new information for me so I spent almost two days reading it.
A Contextual Bandit Bake-off
- Motivation: CB methods are very useful so this paper leverages the availability of large numbers of supervised learning datasets to empirically evaluate contextual bandit algorithms.
- Method: This paper introduces Reg-CB, Greedy, Cover, Cover-NU, Bagging for exploration and IPS, IWR, DR for exploitation.
- Contribution: Very useful for practitioners.