Review for NeurIPS paper: Provably Efficient Neural GTD for Off-Policy Learning