meta-sgd (meta stochastic gradient descent)

A model-agnostic meta-learning method that learns an inner optimizer suitable for a range of learning tasks.