Meta-learning automatically infers an inductive bias by observing data from a number of related tasks. The is encoded hyperparameters that determine aspects the model class or training algorithm, such as initialization learning rate. assumes tasks belong to task environment, and are drawn same environment both during meta-training meta-testing. This, however, may not hold true in practice. In t...