Best Policy Identification in Linear MDPs

Jerome Taupin, Yassir Jedra, Alexandre Proutiere

We investigate the problem of best policy identification in discounted linear Markov Decision Processes in the fixed confidence setting under a generative model. We first derive an…