universal meta-learning

Meta-learning without meta-training on the training set or fine-tuning on the support set.