The Role of Global Labels in Few-Shot Classification and How to Infer Them

Few-shot learning is a central problem in meta-learning, where learners must quickly adapt to new tasks given limited training data.