Actually, googling 'learning to reinforcement learn' links to this paper - https://arxiv.org/abs/1611.05763