Meta-Learning and Universality: Deep Representations and Gradient Descent can Approximate any Learning Algorithm.

Chelsea Finn, Sergey Levine

Browse the full ICLR paper archive.