Meta-learning with recurrent networks and with reinforcement learning