There has been growing progress on theoretical analyses for provably efficient learning in
MDPs with linear function approximation, but much of the existing work has made strong …