Abstract: Few-shot classification is the task of recognizing unseen classes using a limited number of samples. In this paper, we propose a new contrastive learning method called Feature-Level ...