Appendix Meta-Learning with Self-Improving Momentum Target A Overview of terminologies used in the paper

The meta-learner network, i.e., learns to generalize on new tasks. A dataset sampled from a given task distribution that is used for the adaptation. ANIL [ 36 ] only adapts the last linear layer of the meta-model to obtain . The aim of metric-based meta-learning is to perform a non-parametric classifier on top of the meta-model's embedding space Require: Distribution over tasks p (), adaptation subroutine Adapt (), momentum coefficient, weight hyperparameter, task batch size N, number of rollouts per task K, learning rate . Here, we describe the detailed objective of the meta-RL used in the experiments.