Meta-Thompson Sampling

Branislav Kveton 1 Mikhail Konobeev 2 Manzil Zaheer 1 Chih-wei Hsu 1 Martin Mladenov 1 Craig Boutilier 1
                     Csaba Szepesvari 3 2

             Abstract                which the prior distribution over mean arm rewards, a vital
    Efficient exploration in bandits is a fundamental     part of TS, is unknown. While the prior is unknown, the de-
    online learning problem. We propose a variant of      signer may know that it is one of two possible priors ...