Few-Shot Classification with Multi-task Self-supervised Learning

Fan Shi, Rui Wang, Sanyi Zhang, Xiaochun Cao

Abstract: Few-shot learning aims to mitigate the need for large-scale annotated data in the real world. The focus of few-shot learning is how to quickly adapt to unseen tasks, which heavily depends on outstanding feature extraction ability. Motivated by the success of self-supervised learning, we propose a novel multi-task self-supervised learning framework for few-shot learning. To alleviate the deficiency of annotated samples in few-shot classification tasks, we introduce and analyze three different aspects, i.e., data augmentation, feature discrimination, and generalization, to improve the ability of feature learning. The proposed method achieves clear classification boundaries for different categories and shows promising generalization ability. Experimental results demonstrate that our method outperforms the state-of-the-arts on four few-shot classification benchmarks.