A PyTorch reimplementation of MAML, replicating some of the experiments from Finn et al (2017): Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks.
Currently, MAML-Sines.ipynb reproduces (qualitatively) the supervised learning experiment from the paper, on a distribution of sine wave regression tasks. It is designed to be as simple and clear as possible while replicating the behaviour.