Large Margin Meta-Learning for Few-Shot Classification

It is demonstrated that the large margin principle can improve the generalization capacity of state-of-the-art meta-learning methods and lead to consistent and considerable improvements on few-shot classification.